Signal Processing Commons^™

Universal Sound Separation: Distance-Aware Mixture Simulation, Co-Occurrence Conditioning, And Chain-Of-Inference, Wonjun Park

Computer Science and Engineering Theses - Archive

Universal Sound Separation (USS) -- the task of disentangling arbitrary sound sources from a single-channel acoustic mixture -- remains an open challenge due to the ill-posed nature of the problem and the distributional gap between synthetic training data and real-world recordings. This thesis addresses three distinct bottlenecks in the USS pipeline: training data realism, inference strategy, and conditioning richness. We first present two knowledge-guided approaches to sound source separation. The first is a distance-aware mixing strategy that leverages Large Language Models (LLMs) to assign plausible loudness relationships between audio sources during training data synthesis. By querying an LLM about the …

Deep Learning For Wireless Communications, Swarada Ajit Kulkarni

Electrical Engineering Dissertations

The rapid evolution of wireless communication imposes stringent requirements for ultra-reliable, low-latency transmission in dynamic, interference-prone environments. Traditional model-driven signal processing struggles to adapt to nonlinear hardware effects, time-varying channels, and complex interference patterns. Deep learning (DL) offers a transformative, data-driven alternative, enabling end-to-end optimization and robust adaptation under uncertain propagation conditions.

This dissertation investigates deep learning architectures for intelligent and resilient wireless communication through three complementary contributions. The first introduces a Vision Transformer (ViT)-based modulation classification framework that leverages self-attention to capture local and global dependencies in spectrogram representations of Quadrature Amplitude Modulation (QAM) signals. The ViT achieves superior …

Development Of A Cost-Effective Daq For Measuring Brake Performance In Race Cars, Adrin Alias

2025 Spring Honors Capstone Projects - Archive

This project explores the feasibility of creating a cost-effective data acquisition (DAQ) system for high-speed, real-time brake performance testing of Formula SAE racecars. The research addresses the limitations of the current MoTeC DAQ system currently employed by the team, which is costly and time-consuming to set up for on-car testing. The team will use a brake dynamometer for steady-state comparisons of different brake pad compounds (senior design project), but evaluating real-world performance requires on-car testing. By systematically comparing various hardware platforms, sensors, communication protocols, and storage solutions, this project aims to balance cost-efficiency with reliability and performance. The research evaluates …

Novel Electrophysiological Biomarkers In Pediatric Drug Resistant Epilepsy And Genetic Epilepsy Syndromes, Sakar Rijal

Bioengineering Dissertations - Archive

Pediatric epilepsies, particularly those that are drug-resistant or genetically driven, represent some of the most complex neurological disorders encountered in childhood. Central to their pathophysiology is a disruption in the delicate balance between cortical excitation and inhibition (E/I), often resulting from impaired GABAergic interneuron function. This imbalance manifests as aberrant network dynamics and altered neural oscillations, giving rise to seizures and long-term cognitive impairments. In this thesis, we developed a translational framework to identify electrophysiological biomarkers that (i) assess cortical E/I imbalance and (ii) map epileptogenic zones, with the aim of enhancing diagnosis, guiding surgical planning, and informing therapeutic monitoring …

Smart Irrigation System Using Iot And Lstm For Optimal Water Management, Farley Y. Ruiz

Electrical Engineering Theses - Archive

This thesis presents the design and implementation of a smart irrigation system that combines Internet of Things hardware with a Long Short-Term Memory (LSTM) neural network for predictive soil moisture management. The goal is an affordable and reliable solution that uses real-time sensor data and environmental data to schedule irrigation before the substrate moisture drops below its target range. The system integrates soil moisture, temperature, humidity, and sensors on an Arduino Nano that communicates wirelessly with a Raspberry Pi. The Raspberry Pi runs a Python/Flask backend that collects and processes data, executes the LSTM model, and serves a secure web …

Power Quality Event Diagnosis Using Multi-Rate Neural Networks, Jordan D. Lloyd

Electrical Engineering Theses - Archive

Power quality disturbances (PQDs) are among the primary challenges facing modern electrical systems, as they degrade the performance and lifespan of connected equipment. This thesis investigates the relationship between the rate at which voltage waveform data are sampled, the reliability of these measurements, and the ability of deep neural networks to classify PQDs accurately. A one-dimensional convolutional neural network (CNN) was trained and evaluated across multiple sampling rates and signal-to-noise ratios to quantify how information loss in the temporal and spectral domains affects classification reliability. The results demonstrate that model accuracy degrades nonlinearly as sampling rate and signal-to-noise ratio (SNR) …

Study Of Deep Neural Network Trained With Salient, Compressed Medical Video Data For Enhanced Predication, Aileen Sengupta

Electrical Engineering Dissertations - Archive

The rapid growth of surgical video analysis presents a need for efficient deep learning models for surgical training, while reducing the need for excessive image and video image storage. Traditional training approaches typically rely on uniformly compressed video data, instead of selectively preserving the most surgically relevant regions. This dissertation investigates the impact of training deep neural networks (DNNs), both convolutional and transformer-based architectures, on saliency-guided, differentially compressed surgical video sequences. The study systematically evaluates how such compression influences prediction accuracy, computational efficiency and storage requirements. Experimental results demonstrate that models trained on ROI-focused compressed data combined with motion vectors …

Neurovascular Coupling Impairments In Acute Traumatic Brain Injury: An Eeg-Nirs Analysis, Zachary Armstrong

Bioengineering Theses - Archive

Traumatic brain injury (TBI) is a major cause of neurological impairment, often leading to variable recovery and uncertain prognosis in the neurocritical care setting. There is a pressing clinical need for robust, physiologically grounded biomarkers to inform prognosis and therapeutic decision-making in acute TBI. This thesis investigates neurovascular coupling (NVC), the physiological coordination between neuronal activity and cerebral blood flow, as a candidate biomarker for brain function and recovery after injury.

A prospective cohort study was performed using simultaneous electroencephalography (EEG) and near-infrared spectroscopy (NIRS) recordings in patients with moderate-to-severe TBI and healthy controls. Wavelet transform coherence (WTC) analysis was …

Exploring Smart Thermostat, Don P. Dang

2024 Fall Honors Capstone Projects - Archive

This study examines the security, privacy, and compatibility challenges associated with smart thermostats in smart home systems. Smart thermostats, as part of the growing Internet of Things (IoT) ecosystem, face vulnerabilities such as unauthorized access, data breaches, and inconsistent security protocols. Using a mixed-methods approach, this research evaluates encryption techniques, communication protocols (Zigbee, Z-Wave, Wi Fi), and user behaviors that impact system security and efficiency. Key findings indicate that 35% of users are concerned about hacking risks, while 25% express data privacy concerns. Many users lack awareness of security measures, such as firmware updates and password management, which increases vulnerabilities. …

Unsupervised Machine Learning In Wireless Sensor Measurements And Hyperspectral Imaging, Abrar U. Alam

Electrical Engineering Dissertations - Archive

The increasing demand for real-time analysis of sensor data in dynamic environments necessitates innovative approaches to data clustering. This work introduces a novel Online Kernel Clustering (OKC) framework that efficiently determines time-varying clustering configurations without requiring training data. The proposed method employs sparse kernel factorization, guided by a time-dependent metric to quantify the closeness of kernel similarity matrices to a block diagonal structure. By processing data sequentially, the OKC framework is tailored for non-stationary settings. The optimization process integrates block coordinate descent, difference-of-convex functions minimization, and projected sub-gradient descent to iteratively update kernel covariance matrices and cluster memberships online. Numerical …

All-Optical Signal Processing With Fiber-Based Parametric Wavelength Converters, Cheng Guo

Electrical Engineering Dissertations - Archive

The optical signal degradation by optical amplifier noise set the fundamental limit of link reach in the fiber-optics networks. The industrial solution is to use the optical-electrical-optical (OEO) regenerator to clean up the noise at the expense of high-speed electronics and extra cost of laser and photodetectors. All-optical signal processing, enabled by nonlinear optics and optical fiber, intrinsically provides 2-order of magnitude higher processing bandwidth and seamless interface to fiber communication channels. However, there is no robust phase-preserving regenerator that has been experimentally demonstrated without sophisticated polarization tuning and without instable interferometric structure. In this project, we explore the applications …

Neural Net Estimation Of Discriminants Posterior Probability Vector, Harshvardhan Harshvardhan

Electrical Engineering Theses - Archive

Interpreting multi-layer perceptron (MLP) classifier outputs as posterior probabilities is a well-established practice in machine learning and is supported in the literature. However, several authors point out that MLP outputs are very poor estimates of the posterior probabilities. This is demonstrated for classifiers with and without nonlinear output activation. Achieving this reliability depends on key factors such as model complexity, sufficient training data availability, and optimization techniques' effectiveness. In practice, these requirements are not met, resulting in suboptimal probability estimates. Our approach introduces an innovative method based on the softmax output. The method aim to refine MLP discriminants into more …