ENABLING MULTI-SCALE SENSING WITH WIRELESS-INFORMED MACHINE LEARNING: APPLICATIONS IN EARTH AND SPACE - Dissertation

Aug 3, 2026

Upload by Eden

This dissertation investigates how next-generation networks, including 5G and beyond, can enable a dynamic multi-scale sensing ecosystem by integrating global satellite systems with localized wireless technologies. It targets near real-time environmental monitoring through satellites and precise indoor sensing for healthcare, security, and smart infrastructure. The work addresses satellite bottlenecks, wireless data needs for machine learning, and privacy barriers to large-scale deployment. It develops wireless-aware learning frameworks using self-supervised and generative techniques to improve predictability, reduce overhead, and add privacy-preserving mechanisms, enabling scalable, efficient, and robust sensing across global and localized settings.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages174

Reading Time7 h 18 min

AccessFree to download

Abstract

Acknowledgments

© 2024 Jayanth Shenoy ENABLING MULTI-SCALE SENSING WITH WIRELESS-INFORMED MACHINE LEARNING: APPLICATIONS IN EARTH AND SPACE BY JAYANTH SHENOY DISSERTATION Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Computer Science in the Graduate College of the University of Illinois Urbana-Champaign, 2024 Urbana, Illinois Doctoral Committee: Assistant Professor Deepak Vasisht, Chair Professor Brighten Godfrey Professor Matthew Caesar Dr. Vaishnavi Ranganathan, Microsoft Research ABSTRACT This dissertation investigates how next-generation (next-gen) networks, including 5G, and beyond, can enable a dynamic, multi-scale sensing ecosystem by integrating global satellite systems with localized wireless technologies. These networks promise transformative capabilities, from near real-time environmental monitoring via satellites to precise indoor sensing for healthcare, security, and smart infrastructure. However, their growing scale introduces significant challenges. Satellite systems face data transfer bottlenecks, high mobility, and limited downlink capacities, while wireless sensing applications require machine learning models that depend heavily on large, annotated datasets. Additionally, privacy concerns surrounding the pervasive use of wireless sensing present further barriers to scalability. This dissertation focuses on answering the following key research question: How can wireless signal propagation models be integrated with machine learning to help enable multi-scale sensing? This work answers this question by developing novel machine learning frameworks that incorporate wireless domain knowledge to enhance predictability and reduce system overhead. By leveraging self-supervised and generative learning techniques, these frameworks address data inefficiencies, optimize resource allocation, and introduce privacypreserving mechanisms for wireless sensing. The proposed approaches streamline the coordination of complex satellite and wireless sensing systems, reducing reliance on expensive hardware and labor-intensive configurations. This research demonstrates how integrating domain knowledge into machine learning models enables next-gen networks to achieve scalability, efficiency, and robustness across both global and localized sensing applications. Ultimately, this work highlights the potential for intelligent, adaptable sensing architectures that bridge large-scale monitoring with fine-grained precision. To my beloved Avni, for her love and support. iii ACKNOWLEDGMENTS The work presented in this dissertation is the culmination of years of hard work throughout my years as a graduate student at UIUC. It would not have been possible without the help and support of a large group of people to whom I owe my deepest of gratitude. First and most importantly, I am thankful to my advisor Deepak Vasisht for putting in years of effort into mentoring me and helping me grow as a researcher. Deepak has always been one of the most insightful and enterprising people I know. Over the past years, he has always been accommodating as an advisor, helping me with brainstorming research ideas, writing papers, and critiquing presentations. Deepak has always encouraged me to think creatively and adhere to the highest standards of ethics when researching new ideas. I will never forget his words “just do good research”. I am honored to be one of his first students and to have been a part of the journey in founding his new lab at UIUC. I am also grateful to my committee members Vaishnavi Ranganathan, Matt Caesar, and Brighten Godfrey. Their expertise and insightful perspectives have not only shaped the direction of this research but also deepened my understanding of the broader implications of this work. I am thankful for their active participation and thoughtful feedback which have helped me tremendously in improving the work presented in this dissertation. I would also li

FAQs

What sensing problem does the dissertation address?

It examines how next-generation networks can integrate satellite and wireless technologies to support dynamic multi-scale sensing from near real-time monitoring to fine-grained indoor sensing.

Why are scalability challenges significant for multi-scale sensing?

Satellite systems face data-transfer bottlenecks, high mobility, and limited downlink capacity, while wireless sensing relies on machine learning models requiring large annotated datasets; pervasive wireless sensing also raises privacy concerns.

How does the dissertation improve multi-scale sensing using machine learning?

It develops machine learning frameworks that incorporate wireless domain knowledge, using self-supervised and generative learning to reduce data inefficiencies, optimize resource allocation, and introduce privacy-preserving mechanisms.

Aug 3, 2026

Upload by Eden