Event Title

Experimental Implementation of Wavefront Sensorless Real-time Adaptive Optics Aberration Correction Based Upon a Deep Neural Network

Faculty Advisor

Narendra Jaggi

Faculty Advisor

Gabriel Spalding

Graduation Year

2020

Location

Center for Natural Sciences

Start Date

4-4-2020 9:00 AM

End Date

4-4-2020 10:00 AM

Description

Traditional adaptive optics (AO) aberration correction algorithms require multiple iterations, so are too slow for purposes such as free-space optics communications, which suffer from fast distortions due to atmospheric aberrations. Recently, deep neural network (DNN) based methods were proposed, in papers using simulations alone to demonstrate the high-speed aberration correction capabilities of such approaches. We describe experimental implementation of such techniques, using a multiplexable spatial light modulator (SLM), where atmospheric aberration is achieved in lab by using a heating element. Graphics processing unit (GPU) acceleration is used for both the neural network and phase profile generation to enable real-time aberration correction. We are capable of real-time aberration correction at 50 frames per second, which is limited by the camera acquisition speed.

This document is currently not available here.

Share

COinS
 
Apr 4th, 9:00 AM Apr 4th, 10:00 AM

Experimental Implementation of Wavefront Sensorless Real-time Adaptive Optics Aberration Correction Based Upon a Deep Neural Network

Center for Natural Sciences

Traditional adaptive optics (AO) aberration correction algorithms require multiple iterations, so are too slow for purposes such as free-space optics communications, which suffer from fast distortions due to atmospheric aberrations. Recently, deep neural network (DNN) based methods were proposed, in papers using simulations alone to demonstrate the high-speed aberration correction capabilities of such approaches. We describe experimental implementation of such techniques, using a multiplexable spatial light modulator (SLM), where atmospheric aberration is achieved in lab by using a heating element. Graphics processing unit (GPU) acceleration is used for both the neural network and phase profile generation to enable real-time aberration correction. We are capable of real-time aberration correction at 50 frames per second, which is limited by the camera acquisition speed.