Exoplanet Detection via Differentiable Rendering

TL;DR

This work tackles the challenge of detecting faint exoplanets in direct-imaging data plagued by speckle noise from wavefront aberrations. It introduces a differentiable renderer that models wave-space light propagation through a coronagraphic telescope, allowing gradient-based refinement of outdated wavefront estimates using wavefront-sensing data. The approach yields substantial improvements in starlight subtraction and planet detectability, approaching fundamental photon-noise limits and outperforming KLIP in simulations modeled on JWST/NIRCam. The framework is robust to drift, supports multiple measurements and planets, and offers a scalable, wave-space alternative to traditional reference-image–driven post-processing, with potential extensions to broadband, real JWST data, and other high-contrast imaging platforms.

Abstract

Direct imaging of exoplanets is crucial for advancing our understanding of planetary systems beyond our solar system, but it faces significant challenges due to the high contrast between host stars and their planets. Wavefront aberrations introduce speckles in the telescope science images, which are patterns of diffracted starlight that can mimic the appearance of planets, complicating the detection of faint exoplanet signals. Traditional post-processing methods, operating primarily in the image intensity domain, do not integrate wavefront sensing data. These data, measured mainly for adaptive optics corrections, have been overlooked as a potential resource for post-processing, partly due to the challenge of the evolving nature of wavefront aberrations. In this paper, we present a differentiable rendering approach that leverages these wavefront sensing data to improve exoplanet detection. Our differentiable renderer models wave-based light propagation through a coronagraphic telescope system, allowing gradient-based optimization to significantly improve starlight subtraction and increase sensitivity to faint exoplanets. Simulation experiments based on the James Webb Space Telescope configuration demonstrate the effectiveness of our approach, achieving substantial improvements in contrast and planet detection limits. Our results showcase how the computational advancements enabled by differentiable rendering can revitalize previously underexploited wavefront data, opening new avenues for enhancing exoplanet imaging and characterization.

Figures (16)

• Figure 1: A 1D toy example illustrating the basic principles and challenges of the problem addressed in this paper, simplified by examining the PSF in 1D. The first panel shows the photon counts resulting from a star and a planet, including the effects of wavefront aberration. The second panel additionally displays an incorrect star PSF used in practice (measured under a different wavefront aberration), and the level of photon noise in observations. The third panel shows 1) Outdated: subtraction using the outdated star PSF, which leaves significant residuals that can be mistaken for the planet signal, posing the risk of false positive detection or too low signal-to-noise; 2) Optimal: perfect starlight subtraction, representing the upper bound of performance, limited only by noise. The core imaging problem is separating the planet signals from the star signals. Our goal is to improve on the outdated starlight subtraction result via differentiable optimization, achieving a more accurate separation of the star and planet signals. This will effectively result in accurately reducing starlight residuals and improving planet detection sensitivity.
• Figure 2: Schematic diagram of the Lyot Coronagraph equivalent to the one in JWST NIRCam, showing the role and impact of each optical element. This optical system forms the basis of our differentiable rendering approach and illustrates the complexity of the imaging process we aim to model. Light enters from the left, reflecting off the primary mirror (aperture) in the pupil plane, then focuses on the image plane where a coronagraphic mask blocks the central region containing the star. In the subsequent pupil plane, a Lyot stop suppresses residual diffracted starlight before the light undergoes instrument-induced wavefront aberrations and is refocused onto the image plane detector. The plots show (a) aperture amplitude transmission, (b) coronagraphic mask amplitude transmission, (c) Lyot stop amplitude transmission, (d) static optical aberrations in the NIRCam instrument, and (e) resulting point spread function (PSF) at the detector, displaying measured intensity at each pixel.
• Figure 3: Illustration of the sources of wavefront aberrations. In space-based instruments like the James Webb Space Telescope (JWST), aberrations stem from mirror misalignments, deformations, or tilt errors (purple curves) and slowly-evolving non-common path aberrations (blue curves); the former are measured typically every two days (not simultaneous with the science observations), while the latter are not measured directly. See the Appendix \ref{['s:appendix_A']} for details specific to JWST and NIRCam. Wavefronts are partially corrected by deformable mirrors, but residual distortions persist in the final image due to optical imperfections.
• Figure 4: Overview of the proposed differentiable rendering framework for enhanced exoplanet imaging. The method employs gradient-based optimization to improve the initial, outdated wavefront aberration estimate $\hat{\phi}$ based on the discrepancy between the rendered image $\hat{y}$ and the observed science image $y$. The differentiable renderer facilitates efficient backpropagation and gradient descent updates. It is important to note that while $y$ contains signals from both the star and the planet (along with noise and background), $\hat{y}$ only considers the star PSF. Since the optimization changes the structure of the estimated wavefront aberration, which primarily affects the star PSF, it is unlikely to overfit localized features like a planet as it would introduce larger inconsistencies with the dominant star signals. Upon convergence, the refined star PSF estimate is used for PSF subtraction, producing a final image where the planet signals are significantly enhanced and speckle noise is effectively suppressed.
• Figure 5: Visualization of wavefront aberration evolution with real JWST data. We show consecutive optical path difference measurements (taken around three days apart) and their resulting star PSFs, as well as the difference between both. The figure displays consecutive optical path difference (OPD) measurements taken approximately three days apart (a)-(b), their resulting star point spread functions (PSFs) (d)-(e), and the differences between both measurements (c),(f). The OPD color scale is limited to $(-200, 200)$ nm to highlight the overall structure, although a significant meteoroid impact on the lower right mirror segment caused local OPD values to reach 800 nm. This visualization demonstrates the temporal evolution of wavefront aberrations and their impact on the star PSF.