Automotive LiDAR Window Ghost Image Reduction

Introduction

Automotive LiDAR systems rely on a transparent optical window to protect internal components while allowing the transmission of laser pulses and reflected signals. Under strong backlight conditions, such as direct sunlight during dawn or dusk, this window can generate ghost images that severely degrade the detection range and accuracy. These artifacts appear as false reflections in the detector plane, often mimicking genuine targets and triggering erroneous responses in the perception system.

This practical case study focuses on a prototype 905 nm LiDAR receiver used in an advanced driver-assistance system. During outdoor testing, operators observed recurring false objects at approximately 15 meters forward, coinciding with the sun positioned low on the horizon. Using optical simulation tools, we investigated the physical mechanisms behind these ghost images and implemented a combination of optical lens design, stray light suppression, and broadband antireflection coating to drive the total stray light coefficient (TSG) below the industry threshold of 0.5%.

Physical Causes of Ghost Images

Ghost images in a LiDAR optical window arise primarily from Fresnel reflections at the two parallel surfaces of the window. When a high-intensity collimated beam, such as sunlight, hits the front surface, a portion is reflected away, but another portion enters the substrate and reflects off the rear surface before exiting again. This double-bounce creates a displaced secondary beam that propagates into the receiving optical lens and focuses onto the photodetector as a spurious spot.

The severity depends on the angle of incidence and the parallelism of the window surfaces. In our test scenario, the sun at a low elevation angle produced a near-grazing incidence, which increased the reflectance significantly. Additionally, any curvature or wedge in the window due to manufacturing tolerances can further redirect the ghost beam unpredictably. These effects are especially problematic in the optical lens system because the receiving objective collects not only the desired signal but also any off-axis stray light within its field of view.

Beyond the window itself, internal reflections inside the lens barrel and the surface of the detector housing contribute to the overall ghost image. Light that reflects off the window can bounce between the lens edges and the mechanical structure, eventually reaching the sensor. This stray light path is often overlooked but can dominate when the window's antireflection performance degrades or when the surrounding structure lacks adequate geometry for absorbing high-angle rays.

Simulation-Driven Design Optimization

We first built a complete stray light model in LightTools, importing the mechanical housing, optical window, and the receiving lens assembly. The source was defined as a collimated beam with a diameter matching the window aperture, and the sun's spectral profile was approximated using a blackbody curve at 5800 K. A ray tracing analysis with 10 million rays revealed that the ghost image intensity was about 3.2% of the true signal, exceeding the acceptable TSG target by more than six times.

The first improvement addressed the optical lens geometry. We replaced the original spherical receiving lens with an aspheric design optimized for the 905 nm wavelength. The aspheric surface reduced spherical aberration and tightened the image spot, which also allowed us to refine the aperture stop and block certain off-axis ghost paths. In Zemax, we performed a sequential ray trace to evaluate the modulation transfer function and confirmed that the aspheric lens maintained diffraction-limited performance while reducing the angular extent of the detector's sensitivity to scattered rays.

Next, we focused on the mechanical housing. The inner walls of the lens barrel were redesigned with a precise thread pattern, commonly known as a stray light suppression thread. These threads act as a series of small optical traps; any reflected ray that strikes the thread flank is redirected into the thread groove and eventually absorbed by the blackened surface. We selected a 60-degree V-groove thread with a pitch of 0.5 mm and applied a matte black anodized finish. LightTools simulations showed that this geometry reduced the barrel-wall scattering contribution by more than 70% compared to a smooth cylindrical bore.

Finally, we addressed the window surface itself by applying a multilayer broadband antireflection (BBAR) coating. The coating was specifically designed for the 850–950 nm band to cover both the laser wavelength and the solar spectrum adjacent to it. After optimizing the layer thicknesses and refractive indices in Zemax's thin-film environment, we achieved an average reflectance below 0.4% per surface at 905 nm, down from 4% for uncoated glass. This reduction was critical because it minimized the initial energy available for forming the ghost image.

Results and Verification

After incorporating all three improvements into the LightTools model, the stray light coefficient measured at the detector plane dropped to 0.32%, comfortably below the required 0.5% threshold. The ghost image intensity at the expected location was now negligible, and the point spread function of the true return signal remained unchanged. This was verified experimentally by placing a calibrated mirror at the anticipated ghost position and measuring the detector response under a solar simulator.

We also performed a tolerance analysis to ensure the design would survive mass production. The thread geometry was allowed a 10% variation in pitch and depth, and the BBAR coating's layer thickness was varied within 2% to simulate deposition errors. Even at the worst case, the TSG remained below 0.45%, confirming the robustness of the combined approach. An additional benefit was that the aspheric lens improved the overall signal-to-noise ratio by focusing more of the desired return energy onto the detector.

In conclusion, the ghost image in automotive LiDAR optical windows under strong backlight is a multi-factor problem that cannot be solved by the coating alone. A systematic combination of optical lens optimization, mechanical thread traps for stray light absorption, and a high-performance BBAR coating proved effective in reducing the stray light coefficient from 3.2% to 0.32%. This case study demonstrates that simulation tools such as LightTools and Zemax are indispensable for diagnosing and mitigating complex stray light paths in production-viable optical systems. The same methodology can be extended to other sensing platforms, including camera modules and time-of-flight sensors, where environmental glare poses a reliability risk.