Graduation Semester and Year

Spring 2025

Language

English

Document Type

Thesis

Degree Name

Master of Science in Computer Engineering

Department

Computer Science and Engineering

First Advisor

Dr.Debashri Roy

Second Advisor

Dr.Diego Patino

Third Advisor

Dr.Xiaojun Shang

Abstract

The swift evolution of wireless communication technologies,particularly in the field of rf signals or in CBRS bands,demands increasingly sophisticated signal processing techniques to ensure efficient transmission, reception, and spectrum management.Traditional approaches to signal generation and reconstruction, although effective in controlled environments, often struggle to cope with the challenges presented by real-world noisy conditions, hardware constraints, and limited access to large-scale datasets. In response to these limitations, this thesis explores the application of diffusion models—a class of generative models known for their ability to produce high-fidelity samples—to the domain of spectrogram generation for communication signals.

Different from conventional strategies to simulate a spectogram like MATLAB, which often demands significant resources, diffusion models offer an open source and data driven alternative for spectogram generation.By leveraging the signal properties diffusion models tend to learn the underlying part or the distribution of the signal,diffusion models can synthesize and create new signals for analysis under different conditions.is capability is especially important for research communities seeking accessible solutions for data generation, spectrum sensing, signal classification, and anomaly detection.

Inspired by the groundbreaking work of Jonathan Ho et al. on Denoising Diffusion Probabilistic Models (DDPMs)—which demonstrated that diffusion models could outperform GANs in image synthesis—this study explores how diffusion architectures can be adapted to the spectrogram domain. The diffusion framework consists of a forward process, where structured data is progressively corrupted with noise, and a learned reverse process, where the model reconstructs clean data step-by-step from pure noise. This stochastic sampling mechanism allows the generation of diverse spectrograms while preserving key spectral features critical to communication signals.

Furthermore, this research investigates the broader applications of spectrogram diffusion beyond mere data generation. Potential use cases include synthetic data augmentation for federated learning in wireless networks, enhancing model robustness under spectrum scarcity, and transfer learning of the wireless data as well as improved resilience of the signals under noise conditions.By demonstrating both the practical and theoretical advantages of diffusion models for signal representation, this thesis aims to highlight the transformative role of generative AI in advancing wireless communication systems, ultimately enabling more efficient spectrum management, robust signal processing, and intelligent network design.

Keywords

Diffusion, Network, Radio-Frequency signals, DDPM, Probabilistic models

Disciplines

Artificial Intelligence and Robotics | Data Science | Digital Communications and Networking | Statistical Models

License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Available for download on Tuesday, May 19, 2026

Share

COinS