Machine Learning-based Design of Structured Laser Light to Improve Data Transfer Rates in Underwater Communication
Machine Learning-based Design of Structured Laser Light to Improve Data Transfer Rates in Underwater Communication
MIDN 1/C William A. Jarrett
Prof. Svetlana Avramov-Zamurovic
Prof. Joel Esposito
Trident Scholar Project
MIDN 1/C Jarrett successfully completed his Trident Scholar Project in the Spring of 2022. His project was awarded the Class of 1979 Prize for the best project. Additionally, the work from his Trident Scholar Project gave rise to conference presentations at SPIE and Optica, and a conference poster at CLEO. His work was submitted to Optics Communications, with further work being in process currently.
Summary Statement: The goal of this project is to investigate and expand the capabilities of underwater wireless optical communication systems to meet a growing need for high data transfer rates in the undersea environment. This will be done by using machine learning to inform the design of structured light carrying orbital angular momentum to create complex intensity patterns that carry spatially multiplexed information. The performance of this communication system will be evaluated through differing levels of optical turbulence by measuring the bit error rate.
Project Abstract
Whether for business, scientific study, or military operations, there is a growing demand for communication systems capable of high bandwidth underwater transmission. To meet this demand, a communication system using Laguerre-Gaussian (LG) beams of structured light and deep machine learning convolutional neural networks (CNNs), is proposed [1]. In this system, the structured beams of light are spatially encoded to carry more information and maintain a resistance to underwater attenuation. The spatial encoding is the result of combining basis beams, possible due to the orthogonality of the beams, with each combination resulting in a distinct image. This creates an alphabet of 2^N images, where N is the number of basis beams. The basis beams can be switched on and off to form the different combinations, achieving image transmission rates of ~100 MHz [2]. Each added basis beam encodes an additional bit of information that can be passed with each image. However, as the alphabet size increases, achieving a high accuracy with the machine learning classification grows more difficult. Additionally, in the oceanic environment, classification ability is negatively impacted by attenuation, due to scattering and absorption, and optical turbulence, the result of changes in the refractive index of the medium due to changes in the salinity, temperature, and pressure of the water. These challenges motivate investigation into beam design to optimize system bandwidth, and establishing error ratings for practical use.
The first goal of this Trident Project is to investigate the optimal selection of basis beams for designing a system that best preserve the intensity features necessary for reliable classification. Basis beam selection is limited by the resolution of the spatial light modulator (SLM) used to impart the orbital angular momentum (OAM) on the beam, creating a parameter space of topological charges and Laguerre polynomial orders of 0-3. The basis beam candidates will be propagated underwater through optical turbulence. The objective is to determine which beams are preserved better for classification by the machine learning (ML) network. The primary features the ML network uses to classify the basis beams, such as vortex size and intensity correlation, will be investigated and utilized to inform basis beam selection. The hypothesis is that basis beams that prove resilient to optical turbulence can be decoded more reliably.
From this, the project will build on the work in Ref. 2, increasing the bandwidth of the system and experimentally establish a bit error rating (BER) of the system with 8, 9, and 10 basis beams, utilizing optimal basis beam selection. Each additional basis beam increases the bandwidth of the system, and the BER provides a metric to compare the system to ones currently used. The alphabet will be propagated through the same underwater environmental conditions and the results will be quantified using the ML network. The resulting activations, confusion matrix, confidence scores, and classification accuracy will tell us which features of the beams are the most important to the machine learning algorithm to reliably classify the beams, even in the presence of significant optical turbulence. It is important to approach beam design from the ML point of view, because it is the ML network that decodes the transmitted information. This investigation will aid practical use of this underwater optical communication system by optimizing beam design, increasing system bandwidth, and establishing a BER.
Trident Scholar Proposal and Presentation
MIDN 2/C Jarrett presenting his Trident proposal to the USNA Trident Committee
Trident Scholar Report
Trident Scholar Poster
Trident Scholar Presentation
MIDN 1/C Jarrett presenting his Trident Scholar Project to the USNA Trident Committee, as well as various members of the USNA community
SPIE Remote Sensing 2021