Wavefront Coding for Accommodation-Invariant Near-Eye Displays

TL;DR

This work tackles the vergence–accommodation conflict in near-eye displays by achieving accommodation-invariance through wavefront coding with a diffractive optical element (DOE) and end-to-end optimization. A differentiable retinal image-formation model and a pre-processing CNN are jointly trained to optimize the DOE phase profile and image correction, yielding a static-eye AI-NED that extends DoF up to $4 ext{ D}$ without gaze tracking. Simulations and benchtop experiments demonstrate more uniform frequency response across depth, robust performance to fabrication noise, and perceptual quality improvements over state-of-the-art AI-NED approaches. The approach offers a practical path toward VAC mitigation with a compact, static optics solution suitable for wearable near-eye displays and potential human-subject validation in future work.

Abstract

We present a new computational near-eye display method that addresses the vergence-accommodation conflict problem in stereoscopic displays through accommodation-invariance. Our system integrates a refractive lens eyepiece with a novel wavefront coding diffractive optical element, operating in tandem with a pre-processing convolutional neural network. We employ end-to-end learning to jointly optimize the wavefront-coding optics and the image pre-processing module. To implement this approach, we develop a differentiable retinal image formation model that accounts for limiting aperture and chromatic aberrations introduced by the eye optics. We further integrate the neural transfer function and the contrast sensitivity function into the loss model to account for related perceptual effects. To tackle off-axis distortions, we incorporate position dependency into the pre-processing module. In addition to conducting rigorous analysis based on simulations, we also fabricate the designed diffractive optical element and build a benchtop setup, demonstrating accommodation-invariance for depth ranges of up to four diopters.

Figures (17)

• Figure 1: Existing near-eye display architectures to address the VAC. Each method incorporates one or more display planes as well as a light modulator such as a refractive lens or a microlens array. Depending on the architecture, the focal surfaces with varying numbers and shapes can be created, shown as the solid black lines within the scene. Some parts of the display can also be dynamically adjusted as illustrated with arrows, in order to manipulate the focusing mechanism. Please, note that the drawing is not to scale and some elements are exaggerated in size to illustrate the underlying principles.
• Figure 2: Top: Illustration of a typical NED system, including the viewer's eye. The viewing module consists of a 2D display and a magnifying lens. The lens focuses the display image onto a fixed virtual image plane (red line). Accommodation, on the other hand, is expected to dynamically change with respect to the distance of the virtual object, shown as the dash-lined accommodation plane. Bottom: Frequency analysis through the varying accommodation range of 0-4 diopter (D), illustrated via the MTF (left) as well as the MTF gradient (right). The display is capable of presenting high-frequency information at the virtual image plane (red line), around which the frequency response decreases rapidly.
• Figure 3: The frequency analysis illustrating the effect of the wavefront coding to DoF extension in NEDs. Here we assume the cubic phase mask Dowski as the underlying phase plate. Left: One-dimensional cross-sections of the frequency responses through the target depth range of 0-4 D. Right: The gradient of the MTFs with respect to changing depth.
• Figure 4: The proposed end-to-end learning procedure for AI display optimization.
• Figure 5: Near-eye display setup including the viewer. For each pixel on the display, only a subsection of the incoming light enters the retina, limited by the eye pupil. The subsection can be introduced via a virtual sub-aperture at the lens plane. The center of the sub-aperture as well as the angle of incidence to the eye, $(\theta,\phi)$, shifts with the pixel location, $(\xi,\eta)$.