A design tool for Time-of-Flight (TOF) LiDAR sensors used in autonomous vehicles, surveying and robotics. Vary the laser power, pulse width, receiver bandwidth, wavelength and target reflectivity, and the range resolution, SNR, received power, distance precision and max unambiguous range update in real time.
Parameters
LiDAR type
Flash runs at 40 fps; the others at 10 fps
Wavelength
1550 nm is eye-safe (Class 1) and allows higher peak power
Laser power
mW
Pulse width τ
ns
Receiver bandwidth
MHz
Target range R
m
Target reflectivity ρ
%
Black car body 5-10%; concrete 30%; white sign 80%
Ambient (background) light
lux
Night 0, dusk 1000, bright sunlight 100000
Results
—
Range resolution (m)
—
Received power (μW)
—
SNR (dB)
—
Distance precision (cm)
—
Max unambiguous range (m)
—
Lateral resolution @target (cm)
—
LiDAR visual — pulse, reflection, 3D point cloud
The LiDAR unit fires a laser pulse, the target reflects it and the time of flight gives the distance. The bottom strip is an indicative 3D point cloud.
Professor, how is LiDAR different from a camera or radar? It is that spinning thing on top of self-driving cars, right?
🎓
Right, the most famous version is the mechanical-rotating Velodyne or Ouster unit. LiDAR (Light Detection And Ranging) fires short laser pulses and measures the time of flight (TOF) until the echo comes back. The formula is simple: d = c·τ/2, the speed of light times the round-trip time divided by 2. 1 m round-trip is ~6.67 ns, 100 m is 667 ns. As a rough rule, cameras are great at color and texture, radar at velocity, and LiDAR at 3D shape and absolute distance accuracy.
🙋
The default range resolution is 0.75 m. Does that mean a self-driving car can only measure things in 75 cm steps? It would not be able to tell a pedestrian from a sign…
🎓
Great question. δd = c·τ/2 is the minimum gap at which two adjacent objects can still be told apart. If a pedestrian has another object within 75 cm in front or behind, they will be merged into one return. But the distance precision on a single object is much better: δd / √SNR. With SNR = 100 (20 dB) you get 7.5 cm; with SNR = 1000, 2.4 cm. That is the Cramer-Rao lower bound. So a real sensor might be advertised as 75 cm resolution but 2 cm precision.
🙋
So I need to boost the SNR. The received power says 0.001 μW with a 50 mW laser. Is the return really that tiny?
🎓
Yes, and that is the hardest part of LiDAR design. The lidar equation gives P_rec = P_tx·ρ·A_rec / (π·R²). At 100 m with 30% reflectivity and a 5 cm aperture, you only get something like 10⁻⁹ W ≈ 1 nW back. The 1/R² law means 200 m is a quarter and 300 m a ninth of that. Production LiDARs fight back with three tricks: (1) average hundreds of pulses to gain √N in SNR, (2) coherent (FMCW) detection to push past the shot-noise limit, and (3) gated detection to reject ambient light.
🙋
There is a wavelength selector for 905 nm and 1550 nm. The 1550 nm option says eye-safe — does that mean 905 nm is dangerous?
🎓
905 nm is still Class 1 in compliant designs, but the peak-power cap is tight. 1550 nm is absorbed by the cornea and lens before reaching the retina, so under the same Class 1 limit you can emit 10-50× more peak power. That is why long-range premium LiDARs (Luminar, Aeva, >250 m) use 1550 nm InGaAs. 905 nm wins on cost because Si APDs and SPADs are cheap, so Velodyne, Hesai and most Innoviz units stay there. The trade-off: 1550 nm sits in the SWIR band near water-vapor absorption lines, so fog and rain can attenuate it more.
🙋
I have heard Tesla famously switched to a camera-only stack. Is LiDAR really necessary?
🎓
The industry is split. Tesla dropped LiDAR and radar from around 2021 and bets everything on vision plus a neural net for depth estimation. Waymo, Cruise, Mobileye and Zoox take the opposite view and rely on LiDAR for redundancy in low-light, low-contrast and small-object scenarios. The hot trend is FMCW LiDAR (Aurora, Aeva): unlike ToF it returns Doppler velocity together with range, so a single frame already tells you that the car ahead is 30 m away and closing at 50 km/h. More expensive than 905 nm ToF, but increasingly seen as the right stack for L4.
Frequently asked questions
For a time-of-flight LiDAR the range resolution is δd = c·τ/2, where τ is the pulse width and c is the speed of light 2.998×10^8 m/s. A 5 ns pulse therefore yields δd = 0.75 m, a 1 ns pulse 15 cm, and a sub-ns 0.1 ns pulse reaches the 1.5 cm class. In practice this is limited by the APD receiver response time and circuit bandwidth, so 905 nm systems typically sit in 1-10 ns and state-of-the-art SPAD arrays reach the 100 ps range.
This tool uses a simplified lidar equation P_rec = P_tx·ρ·A_rec / (π·R²) that assumes a Lambertian diffuse target. ρ is target reflectivity, A_rec is the receiver aperture area and R is the target distance. The most important feature is the 1/R² fall-off: doubling the range cuts received power to a quarter. Atmospheric attenuation exp(-2αR) and optical-system transmittance are not included, so for fog, rain or ranges beyond 200 m the real receive level can easily be half of this estimate.
905 nm is cheap because Si APDs and SPADs are mature, so most low-cost automotive LiDARs (Velodyne, Hesai, many Innoviz units) use it. 1550 nm needs more expensive InGaAs detectors, but light at that wavelength is absorbed by the cornea and lens before reaching the retina, so under the same eye-safe Class 1 limit you can emit 10-50× more peak power. That gives 1550 nm an edge for long range (>200 m) and bad weather (Luminar, Aeva). The price is that 1550 nm sits in the SWIR band near water-vapor absorption lines, so fog and rain can attenuate it more than 905 nm.
Tesla phased out both LiDAR and radar from 2021 onwards and built AP/FSD around cameras plus a neural network for depth estimation. The reasons are (1) LiDAR BOM cost, (2) software complexity of sensor fusion, and (3) Elon Musk's bet that 'if a human can drive with eyes alone, so can an AI'. Waymo, Cruise, Mobileye and Zoox argue the opposite and keep LiDAR for redundancy in low-light, low-contrast and small-object scenarios. FMCW LiDAR (Aurora, Aeva) is a separate trend: unlike ToF it returns Doppler velocity along with range, so a single frame already tells you the velocity vector of the car ahead.
Real-world applications
Autonomous vehicles and ADAS: Waymo Driver's roof-mounted Honeycomb LiDAR, Velodyne Alpha Puck, Hesai AT128, Luminar Iris and so on. Typical requirements are 200 m detection at 10% reflectivity, ±5 cm distance precision and 0.1° angular resolution. The default 100 m / 30% case in this tool yields just 4 dB SNR, which is harsh by production standards — real units recover 20+ dB through pulse averaging and coherent detection.
Geographic and topographic surveying: Airborne and drone LiDARs (RIEGL, Leica) are the standard for GIS, terrain modelling and forestry. Terrestrial LiDARs (Faro, Leica BLK360) handle 3D building scans and BIM. Because the target is static or slow, long averaging gives sub-cm precision and 10 m to several km coverage.
Agricultural and forestry robotics: Companies like John Deere and Naïo Technologies fit low-cost 905 nm solid-state LiDARs to autonomous tractors and weeding robots for row navigation and crop-vs-weed classification. The same units are the standard sensor for SLAM (Simultaneous Localization and Mapping).
Atmospheric LiDAR: Mie-scattering aerosol LiDARs and Doppler wind LiDARs (NOAA, JMA) measure atmospheric composition, cloud-base height, PM2.5 distribution and upper-air winds. The ToF ranging principle is the same as in this tool, but the target is a scattering volume rather than a hard point, so range resolution is set by the receiver gate width.
Common misconceptions and gotchas
The first trap is confusing range resolution with distance precision. The δd = c·τ/2 = 75 cm value the tool returns for a 5 ns pulse is the minimum gap at which two adjacent objects can still be separated; the precision on a single object is much better, δd/√SNR. The 45-60 cm precision in the default case is dominated by an SNR of only ~4 dB. In a real product, averaging brings SNR to 30 dB (×1000), pushing precision below 3 cm with the same pulse width. The right answer to "can a 75 cm resolution sensor see a pedestrian's legs and torso separately?" is "no, but the pedestrian's center of mass is known to the cm".
The second trap is extrapolating the lidar equation linearly. The simple P_rec = P_tx·ρ·A/(π·R²) used here assumes (a) a Lambertian diffuse target, (b) zero atmospheric attenuation and (c) 100% receiver-optics throughput. Real systems will (1) return either zero or a saturating spike from specular surfaces (bumpers, glass), (2) lose 10-30 dB to fog/rain/snow and (3) drop 30-50% in the receive optics. Use the formula with 6-10 dB of margin at design time and calibrate against the actual sensor, and never exceed the Class 1 eye-safe limit.
The third trap is arguing about range resolution while ignoring angular resolution. This tool fixes angular resolution at a typical 0.5 mrad, but it is the parameter that decides lateral (cross-range) resolution. At 100 m, 0.5 mrad equals 5 cm; at 200 m it is 10 cm. Anything smaller (a pedestrian's arm, a bicycle frame) shows up as a single point. A 32-line Velodyne Puck with 0.2° horizontal step produces only 4-5 returns per pedestrian at 200 m, which forces a lot of work into the perception software. Variable-resolution sensors like Luminar Iris dynamically concentrate sampling at 0.05° in the region of interest.
How to Use
Enter laser power in milliwatts (typical 10–100 mW for automotive LiDAR) and pulse duration in nanoseconds (5–50 ns for standard TOF sensors).
Set receiver bandwidth in MHz (50–500 MHz determines temporal resolution) and target range in meters (0.5–200 m for surveying/autonomous vehicles).
Click simulate to compute range resolution, received power at the detector, SNR in dB, distance precision, maximum unambiguous range, and lateral resolution at target distance.
Worked Example
For a 50 mW automotive LiDAR with 10 ns pulse duration, 200 MHz receiver bandwidth, scanning a building facade at 50 m: range resolution is approximately 0.75 m (limited by c·Δt/2 = 3×10⁸ m/s × 10 ns / 2); received power drops to ~2.1 μW due to inverse-square law; SNR reaches 18 dB; distance precision improves to 7.5 cm; maximum unambiguous range extends to 15 m (determined by 2·c·τ_pulse); lateral resolution at 50 m with 0.1° beam divergence yields 8.7 cm spot size.
Practical Notes
For autonomous vehicle (AV) applications, maintain SNR above 15 dB at maximum operating range; lower receiver bandwidth reduces noise but sacrifices axial resolution.
Surveying-grade systems (e.g., Riegl, Leica) use 100+ mW lasers and nanosecond-level pulse control to achieve centimeter accuracy over 300+ m.
Lateral resolution degrades linearly with distance; at 100 m, a 0.1° beam creates 17.5 cm spots—critical for feature extraction in robotic navigation.
Pulse overlap occurs when unambiguous range is exceeded; use longer pulses or lower repetition rates for distant targets in surveying campaigns.