Point spread function subtraction using telemetry data: one more step on the path to reaching the photon-noise limit in high-contrast imaging

Abstract

Current HCI systems are limited by residual starlight, resulting from atmospheric turbulence unseen by the adaptive optics system or stemming from imperfections in the optical system itself. This leads to speckles in the PSF of the final images inhibiting our detection capabilities. Reaching the photon-noise limit in high-contrast imaging (HCI) is the next major breakthrough to enable the observation of fainter planets, located closer to their host stars, ultimately enabling the detection of Earth analogs. Standard PSF subtraction techniques consist of differential imaging (DI) methods such as reference-star DI (RDI), angular-DI (ADI) and spectral-DI (SDI) which provide a discriminant to separate stellar light from circumstellar light in post-processing. These methods complement the adaptive optics (AO) system corrections and are essential to push contrast limits, typically by a few orders of magnitude. However, they offer limited performance at small angular separations where Earth-like planets are expected, due to strong temporal speckle variation and smaller differential leverage. We present our novel telemetry-based PSF-reconstruction method, using supervised cross-decomposition to learn and reconstruct speckle patterns from WFS telemetry. Using simulated, testbed, and on-sky data, we show that our custom cross-decomposition technique can accurately capture the quadratic relationship between the wavefront sensor images and the camera images, thus accurately reconstructing focal plane speckles and providing an improvement in contrast by an order of magnitude. This approach enables high contrast imaging systems equipped with wavefront sensors (such as Dual-Purpose Focal Plane Mask) to reach the contrast limit set by the photon-noise limit instead of the speckle noise limit. Ultimately, cross decomposition might be of interest to other applications in astronomy for which a sensor / science product with correlated relationships can be found.