Exploring Neural Network Approaches for Pyramid Wavefront Sensor Denoising in Adaptive Optics

Adaptive optics (AO) systems relying on pyramid wavefront sensors (pyrWFS) are essential for high-contrastimaging, particularly in exoplanet detection. However, noise in pyrWFS images—exacerbated by faint guidestars—degrades wavefront reconstruction accuracy and limits correction performance. This paper presents adeep learning-based denoising framework using a pyramidal convolutional autoencoder, designed for integrationin AO control loops. The model is trained on large-scale simulated datasets generated with COMPASS [1], cover-ing diverse observing conditions (seeing, star magnitudes). Our approach demonstrates significant improvementsin image quality (PSNR) while preserving critical wavefront features, enabling more accurate phase reconstruc-tion. Preliminary results show enhanced Strehl ratio and reduced residual noise in closed-loop AO simulations,validating the method’s potential for on-sky deployment. This work paves the way for denoising solutions innext-generation AO systems.