Machine Learning Optimization of OAM-mediated Nonlinear Fiber Optics

Project Description

Structured light – i.e. light with spatially-tailored intensity, phase, or polarization properties – has become a ubiquitous tool across myriad applications in optics; including biological imaging, quantum optics, telecommunications, and more. In particular, it has been shown that light beams which carry orbital angular momentum (OAM) can propagate for kilometer-scale distances in both free-space and optical fibers, opening up many new avenues for linear and nonlinear photonics due to the robustness and multiplicity of this, in principle, infinite basis. Targeting specific single, or superpositions of, OAM modes generally requires the use of digital holograms encoded by spatial light modulators (SLMs). These SLM-based generation systems can flexibly excite pure OAM modes; however, optimization and alignment of these systems is non-trivial, and they are often plagued by alignment drift during long-duration experiments.

Recently, the High Dimensional Photonics Lab has had encouraging results that indicate that the combination of Multi-Plane Light Conversion (MPLC) – a spatial phase modulation technique for exciting OAM modes with nearly zero theoretical loss – and machine learning methods can be combined to facilitate automatic optimization and alignment of OAM excitation systems. In this project, we will build upon these results to directly apply machine learning algorithms to optimize and stabilize high power laser generation at novel colors using the nonlinear interaction between OAM modes in optical fibers.

Mentors

Siddharth Ramachandran, PI    Purva Bhumkar Jeff Demas

In this project we will implement a Genetic Algorithm (GA) to control 24 parameters affecting the pure generation of OAM modes. We will measure the power of a laser beam generated by the interaction between these OAM modes, which in turn will provide feedback for the optimization of our SLM-based MPLC scheme. We will study the convergence of the algorithm to determine the robustness of the output solution space, as well as how the performance of the algorithm compares to human alignment and optimization. Informed by these data, we hope to determine which are the crucial parameters within the initial 24 considered, and prune or add degrees of freedom as needed to improve the algorithm in terms of both output laser power and speed of convergence. Using the improved algorithm, we will implement a scheme to periodically compensate alignment drift in our system – allowing for long-duration experiments crucial for achieving our project’s ultimate goal: the generation of a high-power blue laser for underwater telecommunications and sensing applications.
This project will allow for learning opportunities in hardware, software, and physics. With respect to hardware, the student will learn about polishing, cleaving, and mounting customized multi-mode optical fibers. They will be trained in the safe use of high-power laser systems, as well as how to align multi-component free space optical systems. The student will learn to use and modify custom software to program the output of spatial light modulators. They will also learn about the implementation of Genetic Algorithms, and potentially other machine learning optimization methods. The student will gain experience in writing software to analyze and display data in a scientifically rigorous fashion. Additionally, the student will learn about the propagation of light in optical fibers and OAM. They will learn about the fundamental conservation rules that dictate the conversion of energy between different color light beams through nonlinear optics and specifically, the process of four-wave mixing.

Timeline