Investigation of Machine Learning Techniques in Stabilization of Co-Propagating Polarization-Encoded Photons - Thesis

Aug 5, 2026

Quantum communication protocols and networks have progressed in laboratory experiments, yet practical operation is often limited by short time windows and substantial environmental noise that degrades both time-coded and polarization-encoded photon systems. Polarization encoding is widely used, but its stability remains a core constraint. This thesis applies modern machine learning to integrate classical and quantum signals, evaluating direct prediction, variable recalibration for the quantum channel, and reinforcement-learning-based calibration. Experimental testing over hours to days assesses attention models, sliding-window time series predictors, and reinforcement learning for enhanced longevity, reliability, and robustness.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages38

Reading Time1 h 36 min

AccessFree to download

Chapter 1 Introduction and Background

Introduction to Quantum Communication Systems

Challenges in Polarization Stabilization

Machine Learning for Quantum Systems

Problem Statement

Proposed Schemes

Thesis Structure

Chapter 2 Literature Review

Photon Encoding Techniques in Optical Fiber Quantum Communication Networks

Stabilization Strategies for Fiber-Optic Quantum Channels

Chapter 3 Methodology

Preface

Final Scheme: Direct Prediction Across Bands

Other Approaches

INVESTIGATION OF MACHINE LEARNING TECHNIQUES IN STABILIZATION OF CO-PROPAGATING POLARIZATION ENCODED PHOTONS BY YUEZE LIU THESIS Submitted in partial fulfillment of the requirements for the degree of Master of Science in Electrical and Computer Engineering in the Graduate College of the University of Illinois Urbana-Champaign, 2025 Urbana, Illinois Adviser: Professor Eric Chitambar ii ABSTRACT Quantum communication protocols and networks have advanced considerably in experimental laboratories; however, their operation is frequently constrained by short timeframes and significant environmental noise. These challenges affect both time-coded and polarization-encoded photon systems. Polarization encoding has become the prevalent method, despite its intrinsic issues with stability. This thesis investigates the integration of classical and quantum signals through the application of modern machine learning techniques, with the goal of overcoming the stochastic challenges that have historically impeded system performance. We evaluate three distinct schemes: (1) direct prediction of the quantum signal from the classical signal,(2) variable recalibration queue for the quantum channel, and (3) calibration of the quantum signal via reinforcement learning. Each approach is characterized by unique strengths and limitations in terms of measurement overhead, implementation complexity, and operational stability. Through extensive experimental testing over durations ranging from hours to days, we analyze the performance of attention-based models, sliding window time series predictors, and reinforcement learning frameworks in predicting and stabilizing polarization states. Our findings indicate the potential of machine learning approaches to enhance the longevity and reliability of quantum communication systems, while also highlighting the challenges of model generalization and data requirements for robust performance. This research contributes to the advancement of quantum communication infrastructure by demonstrating practical methods for maintaining stable polarization encoded quantum channels, essential for the development of long-distance secure quantum networks. iii To my family, who have supported me throughout this journey. iv ACKNOWLEDGEMENTS I would like to express my deepest gratitude to my advisor, Dr. Eric Chitambar, for his invaluable guidance, support, and encouragement throughout my doctoral journey. His expertise and insights have been instrumental in shaping this research. I am grateful to the members of the Quantum Information Group at the University of Illinois at Urbana-Champaign for creating a collaborative and stimulating research environment. Special thanks to my lab colleagues who contributed to discussions and provided assistance during experimental setups. I would like to acknowledge the support of the Department of Electrical and Computer Engineering for providing the necessary resources and facilities for conducting this research. Additionally, I appreciate the funding support from the Department of Energy’s InterQnet project and the mentorship and collaboration of Professor Prem Kumar at Northwestern University for their efforts in collecting the data and everything hardware-related to realize this project. Finally, I extend my heartfelt thanks to my family and friends for their unwavering support, patience, and understanding during the course of this work. This achievement would not have been possible without their encouragement and belief in me. v Contents List of Figures ......................................................................... viii Chapter 1 Introduction and Background ............................................. 1 1.1 Introduction to Quantum Communication Systems .............................. 1 1.2 Challenges in Polarization Stabilization ......................................... 1 1.2.1 Environmental Factors .......................

FAQs

What problem does the thesis address in polarization-encoded quantum photonic systems?

It targets instability caused by environmental noise and operational constraints, which can limit both time-coded and polarization-encoded photon systems. The work focuses on stabilizing polarization-encoded photons over realistic timeframes.

What machine learning-based schemes are evaluated for stabilization?

Three schemes are assessed: direct prediction of the quantum signal from the classical signal, variable recalibration using a queue for the quantum channel, and reinforcement learning to calibrate the quantum signal.

How is model performance analyzed and what are the main limitations identified?

Extensive experiments run over durations from hours to days and compare attention-based models, sliding-window time series predictors, and reinforcement learning. Results emphasize gains in longevity and reliability while also highlighting challenges in model generalization and data requirements.

Aug 5, 2026

Upload by Valentina