Wavefront Reconstruction for a Holographic Modal Wavefront Sensor Based on Extreme Learning Machine

Abstract

The intermodal crosstalk effect as well as the limited dynamic range of holographic modal wavefront sensors (HMWFSs) significantly affect their wavefront-sensing accuracy. Thus, this study was aimed at proposing an extreme learning machine (ELM)-based wavefront-reconstruction algorithm for holographic HMWFSs to overcome the errors caused by crosstalk as well as extend the dynamic range of the sensors. The simulation results indicated that the proposed ELM-based algorithm reduced the crosstalk-induced residual wavefront root mean square error to 4.7% of the initial value, and this was 84.6% lower than the reduction achieved by the conventional sensitivity-matrix method. After selecting the optimal range of training samples, the ELM model further reduced the residual error by approximately 74% under aberration conditions, where the conventional method reached its convergence limit. Thus, we proposed an ELM model for mitigating the issue of the linear regression relationship between the differential signals measured by HMWFS and the incident-wavefront Zernike-mode coefficients under the aberration-mode crosstalk effect.