End-to-end modelling and speckle subtraction with the JWST NIRCam coronagraph

The limiting noise floor in almost all coronagraphic observations is from uncertainty in the field of speckles around a bright star, in which optical aberrations can easily hide a faint planet. Data-driven methods for subtracting this field have been very successful, but the exquisite precision of the James Webb Space Telescope instruments implies much deeper fundamental contrast and resolution limits than are routinely being achieved. Recent advances in differentiable optics models and machine learning have made it possible to simulate and perform phase retrieval on the optical system end-to-end, trained on science data, and provide much more accurate predictive models of the speckle field in the James Webb NIRISS Interferometer. We present the early results of adapting these methods to the NIRCam coronagraph, achieving improvements in wavefront sensing and metrology and a path to deeper coronagraphy with JWST and other space telescopes.