Wavefront Shaping via a Learning Algorithm Guided, Adaptive Optics Device for Focal Volume Control
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
ABSTRACT
The topic of this dissertation is the development of a learning algorithm for the shaping of a lasers three-dimensional focal intensity distribution, specifically for the interest of generating non-diffracting beams. These beams have garnered growing interest in fields such as microscopy and laser physics. The learning algorithm developed is based on a genetic algorithm (GA) approach. The goal of the algorithm is to find the necessary bi-dimensional phase (or wavefront) to apply to a Spatial Light Modulator (SLM) in order to optimize the laser spatial profile; which is measured by a Charged Coupling Device (CCD) camera. Two versions of the GA were developed, validated and used, the first one performing wavefront correction (or focal spot optimization), and the second one focal volume optimization.
The first version of GA is called single-plane GA, as it uses the signal of the CCD camera in one single transverse plane. This approach is similar to what is standard in wavefront correction schemes that are not using - for optimization - the direct measurement of the laser wavefront, but instead a signal generated by the laser that is dependent on the wavefront aberrations (or shape). An experimental setup is described, and wavefront optimization experiments validate the use of the single-plane GA as a way of correcting aberrations purposely induced on a Gaussian beam laser. The feedback signal, which can be maximized or minimized, is the spatial profile maximum intensity or its beam width.
As the single-plane GA uses the spatial profile, measured in one transverse plane, it is unable to optimize the lasers focal distribution in a predictable way. An extension of the method, called multi-plane GA, is later proposed. This novel approach to beam shaping uses, as feedback for algorithm optimization, the spatial profiles measured in multiple transverse planes. Recording multiple transverse planes at different longitudinal positions is achieved using the SLM, which applies a defocus term without physical translation of the CCD camera. This multiple recording capability requires a proper defocus calibration of the SLM, which was performed and validated via measurement of a non-aberrated Gaussian beam, as well as a SLM-generated astigmatic Gaussian beam, with comparisons taking the 3D recording of the focal distribution using manual translation of the CCD camera.
Different fitness methods using multiple transverse planes were developed and their implications and limits tested. The balance between the algorithm convergence speed, the degrees of freedom of the 2D phases induced by the SLM, the size of the algorithm search space, and ability of the GA to perform a given focal volume shaping goal, are discussed. Using the minimization of energy, size or similitude between spatial profiles in the five planes, convergence of the multi-plane GA towards a non-diffracting Bessel beam and Airy beam was demonstrated. Compared to the initial starting corrected Gaussian beam, the multi-plane GA was able to increase the laser depth of field up to 32 times for the Bessel beam, and up to 7 times for the Airy beam.