Mitigating Ghost Images in LiDAR Optical Windows

Introduction#

Automotive LiDAR systems must operate reliably in harsh lighting conditions, including direct low-angle sunlight and oncoming high-beam headlights. In these strong backlight scenarios, the optical window that protects the transmitter and receiver apertures can become a significant source of ghost images. A ghost image is a secondary, displaced reflection of a bright source that reaches the detector and competes with the true return signal. For a time-of-flight LiDAR, even a faint ghost can trigger false detections or raise the noise floor enough to reduce ranging accuracy at long distances.

This case study examines a roof-mounted LiDAR module whose window produced a measurable ghost at roughly 1.2 degrees off-axis under a 100 klux backlight. The engineering team combined optical simulation in Zemax and LightTools with physical prototyping to trace the stray light path, then applied three coordinated countermeasures: aspheric surface optimization, threaded baffle design inside the lens barrel, and a multilayer broadband anti-reflective coating. The result was a reduction of the total stray light coefficient (TSG) from 2.8% to 0.42%, comfortably below the 0.5% target.

Physical Origins of Ghost Images Under Backlight#

Ghost images in a LiDAR window arise from multiple reflections between optical surfaces. The window itself has two surfaces; each reflects roughly 4% of incident light at normal incidence when uncoated, and the reflection rises sharply at grazing angles. Strong backlight enters the window, reflects off the inner surface, bounces back off the outer surface, and then travels toward the detector as a displaced secondary image. The lens elements behind the window add further surface pairs, creating a cascade of possible ghost paths.

The severity depends on surface curvature, aperture stops, and the angular acceptance of the detector. A plano window is often assumed to be benign, but its parallel surfaces form a weak etalon that produces a laterally shifted ghost. When the window is tilted or curved for aerodynamic integration, the ghost displacement changes with field angle, making it harder to mask with a fixed baffle. Scattering from lens barrel walls and mounting interfaces further contributes a diffuse halo that raises the overall stray light floor.

Simulation Workflow with Zemax and LightTools#

The team first built a sequential model in Zemax to capture nominal imaging performance and surface-by-surface reflection coefficients. Non-sequential ray tracing in LightTools then modeled the full window, barrel, and detector geometry, including measured surface roughness and coating data. A Monte Carlo source with a 100 klux collimated beam at multiple incidence angles was launched, and rays were classified by path length and surface interaction count to isolate double-reflection ghosts from single-scatter background.

LightTools' stray light analysis produced a path map showing that 61% of the ghost energy came from the window's inner-outer surface pair, while 27% originated from the front lens element and 12% from barrel wall scatter. This ranking directed the design effort: the window coating and tilt were prioritized, followed by the front lens asphere and the barrel threading. Zemax optimization then evaluated aspheric coefficients to shift ghost foci away from the detector active area without degrading the nominal modulation transfer function.

Aspheric Optimization and Baffle Threading#

Replacing the spherical front element with an aspheric surface allowed the team to control the radius of curvature across the aperture. By adjusting the conic constant and higher-order coefficients, the ghost reflection from that element was defocused and displaced outside the detector footprint. Zemax merit functions combined spot size, distortion, and a custom ghost displacement operand, ensuring the optimization did not trade imaging quality for stray light suppression. The final asphere reduced front-element ghost contribution by 74%.

Inside the lens barrel, conventional smooth walls acted as a light pipe. The team introduced fine-pitch extinction threads with a V-groove profile and a matte black anodized finish. LightTools simulations compared thread pitches of 0.5 mm, 0.8 mm, and 1.2 mm; the 0.8 mm pitch with a 60-degree groove angle trapped the most grazing rays. Physical prototypes confirmed the simulation, showing a 3.1x reduction in barrel-scattered light. A secondary baffle ring at the window mounting flange blocked the direct backlight path that had fed the strongest ghost.

Multilayer Broadband Anti-Reflective Coating#

The window received a multilayer broadband anti-reflective coating tuned for the 905 nm LiDAR band and the visible spectrum used by the camera fusion module. The coating stack used alternating high- and low-index dielectric layers, with thicknesses optimized by a needle-synthesis algorithm to minimize reflectance across 400 to 1000 nm and across incidence angles from 0 to 60 degrees. Average reflectance dropped from 4.1% per surface to 0.18% at normal incidence and stayed below 0.9% at 60 degrees.

Coating durability was validated with adhesion, abrasion, and humidity testing to automotive standards. The team also verified that the coating did not introduce angular color shift that could confuse the fused camera. Because the coating reduced both window surface reflections, the double-reflection ghost path lost most of its energy at the source, complementing the geometric fixes rather than duplicating them.

Results and Real-World Validation#

After integrating all three countermeasures, the module was tested in a solar simulator at 100 klux with incidence angles swept from 0 to 75 degrees. The measured total stray light coefficient fell to 0.42%, with the worst-case ghost peak suppressed by 18 dB relative to the original design. Long-range ranging tests at 200 meters showed no false detections under direct backlight, and the noise floor remained within 3% of the dark-condition baseline.

The approach has since been applied to a second-generation window with a more aggressive curvature for styling. The same workflow, sequential Zemax for nominal design, non-sequential LightTools for stray light path analysis, and iterative coating optimization, reduced development time by roughly 30% compared with the first program. The case demonstrates that ghost image suppression in LiDAR windows is best treated as a system problem: optical surface shape, mechanical light trapping, and thin-film coating must be co-optimized to push the stray light coefficient below the 0.5% threshold reliably.