Reducing Ghost Images in Automotive LiDAR Windows via Aspheric, Threads, and BBAR Coatings

Introduction

In automotive LiDAR systems, the optical window acts as the first interface between the sensor and the environment. Under strong backlight conditions—such as direct sunlight or high-beam headlights—multiple reflections within the window and downstream optics can create parasitic ghost images that degrade measurement accuracy. This case study investigates the physical origins of these ghosts and presents a combined optimization strategy leveraging non-spherical surfaces, barrel baffle threads, and multilayer broadband antireflection (BBAR) coatings.

Using commercial ray-tracing tools like LightTools and Zemax, we simulate the full optical path from the external source through the window into the receiver lens. The goal is to reduce the Total Stray Ghost (TSG) coefficient below 0.5%, a critical threshold for reliable long-range detection. Real-world implementation on a prototype LiDAR unit confirms the simulation findings.

Physical Causes of Ghost Images in LiDAR Windows

Ghost images arise primarily from Fresnel reflections at each optical surface. For a typical uncoated glass window (refractive index ~1.5), each air-glass interface reflects about 4% of incident light. Under strong backlight, these reflections can bounce multiple times between the front and rear surfaces of the window and between the window and the downstream optical lens elements, creating secondary or tertiary ghost foci.

In a LiDAR receiver, the ghost often appears as a faint offset spot near the true signal, leading to false detections or range ambiguity. The geometric path of the ghost depends on the window curvature, tilt, and the lens prescription. Our simulation shows that even a 0.1° misalignment between the window and the optical axis can double the ghost intensity at the detector plane.

Simulation Setup with LightTools and Zemax

We modeled a typical coaxial LiDAR system: a 905 nm pulsed laser source, a scanning mirror, a plano-concave window, and a four-element f/2 receiver lens. The window was initially a spherical meniscus with a radius of 150 mm. Using non-sequential ray tracing in LightTools, we placed a 10° cone of collimated backlight at 0.5° off-axis to simulate extreme glare. Ghost energy was measured on a 2D detector with 20 μm pixel pitch.

Zemax was used to optimize the window shape and lens coatings in sequential mode, then the design was exported to LightTools for full stray light analysis. The baseline TSG coefficient—defined as the ratio of ghost irradiance to the primary signal irradiance at the detector—was 2.3%, well above the 0.5% goal. Dominant ghosts came from a double bounce between the window rear surface and the first lens element.

Optimization Strategy: Aspheric Surfaces, Baffle Threads, and BBAR Coatings

Aspheric optimization: The window inner surface was converted to an even-order asphere (4th–8th terms) using Zemax’s damped least-squares algorithm. The optimized shape reduced the secondary reflection path by directing ghost rays away from the active detector area. In LightTools, the TSG from window-lens interactions dropped from 1.5% to 0.8%.

Barrel baffle threads: We added a series of 0.3 mm deep V-groove threads on the inner wall of the window retaining barrel. The threads were angled at 60° and coated with matte black paint (absorption >95% at 905 nm). This trapped rays that would otherwise scatter from the barrel edge into the optical path. The residual stray light from mechanical reflections was reduced by a factor of 3.

Multilayer BBAR coating: A six-layer design of alternating high-index (TiO₂) and low-index (SiO₂) films was applied to both sides of the window and to the front surface of the lens. The coating achieved an average reflectivity <0.3% over 850–1000 nm. Combined with the above measures, the TSG coefficient fell to 0.42%, below the 0.5% target.

Results and Practical Validation

The optimized prototype was built and tested under a 120,000 lux halogen lamp simulating strong backlight. The ghost spot intensity measured on the detector array was within 0.45% of the signal peak, matching the simulation. False alarm rate in the LiDAR’s object detection algorithm decreased by 87% compared to the baseline design.

This case demonstrates that a multi-pronged approach—including non-spherical window geometry, aggressive mechanical baffling, and broadband antireflection coatings—can effectively suppress ghost images in high-volume automotive LiDAR. The design rules are transferable to other optical systems where stray light from window reflections is a concern.

Conclusion

Ghost images in LiDAR optical windows under strong backlight are caused by multiple Fresnel reflections between surfaces. By substituting a spherical window with an aspheric surface, adding knife-edge baffle threads to the barrel, and applying a high-performance BBAR coating, the total stray ghost coefficient can be reduced from >2% to below 0.5%. This case study provides a practical roadmap for engineers designing robust optical lenses for autonomous vehicle sensors.