48440 research outputs found
Sort by
Development of Iron Chelated Silk Fibroin Microfibers as an Injectable, Magnetically Aligning Nerve Guidance Architecture
No full textAn alarming disparity persists between the incidence and prevalence of traumatic spinal cord injury (SCI): while only 18,000 new cases are reported annually, an estimated 300,000 individuals nationwide are currently living with chronic, unresolved injuries from spinal cord trauma. With minimal spontaneous neuroregeneration and no clinically available cure, the least likely prognosis for SCI patients, attained in only 0.5% of cases, is complete neurological recovery. Functionally integrative neuro-recovery is a multifaceted outcome, requiring synergistic efforts of both innate neuroprotective mechanisms and therapeutic neuroregenerative interventions. To this end, lesion-bridging biomaterial scaffolds, especially those with aligned guidance architectures, are a promising strategy for stimulating neurite outgrowth and organizing axonal advancement to reconnect disrupted neural pathways. Unfortunately, intralesional surgical implantation of these preformed scaffolds exacerbates aggressive secondary injury cascades, compromising innate neuroprotective mechanisms that stabilize the injury site. Though injectable biomaterials have been explored as a minimally invasive alternative, they generally lack the hierarchical organization critical to directed neurite outgrowth, preserving innately established neuroprotective mechanisms at the cost of neuroregenerative support. Aiming to combine the neuroprotective benefits of injectable biomaterials and the neuroregenerative benefits of hierarchically organized pre-formed scaffolds, this dissertation reports the development and characterization of an injectable, magnetically aligning nerve guidance architecture that can be incorporated into otherwise unstructured in situ crosslinking hydrogels. By leveraging the unique metal binding capacity of silk fibroin, passive chelation of ferric iron ions is explored herein as an alternative, magnetic nanoparticle (MNP)-free approach to functionalizing this innately fibrous, naturally occurring, FDA-approved biomaterial. Characterization of ferric iron chelated silk fibroin microfibers (Fe3+-mSF) revealed not only increased magnetization potential compared to nascent, ‘no iron added’ mSF, but also illustrated increased unidirectional alignment uniformity when exposed to an external magnetic field of clinically relevant strength (400 mT). When incorporated as an architectural component in otherwise unstructured hyaluronic acid-based hydrogels, Fe3+-mSF did not disrupt the syringeability, critical gelation time, viscoelastic properties, or low swelling profile of the hydrogel-only system, indicating preservation of mechanical functionalities essential to the hydrogel’s performance as a minimally invasive therapy. Fe3+-mSF cytocompatibility was demonstrated both as an independent biomaterial (2D culture) and as an architectural component of a collagen composite hydrogel (3D scaffold). Notably, a key cytoskeletal biomarker of axonal extension and elongation, TUBB3 (β-tubulin III), seemed to be upregulated in cells seeded in aligned, iron-modified Fe3+-mSF/collagen hydrogel composites by comparison to cells seeded in collagen-only scaffolds. Finally, in a proof-of-principle study, magnetically actuated and gelation-preserved alignment of Fe3+-mSF/hydrogels was achieved on the patient bed of an MRI machine with only the instrument’s static magnetic field, with no obvious aggregation or gross displacement of the fibers within the hydrogel. Given the applicability of MRI-guided delivery procedures in intralesional scaffold placement, alignment without gross displacement of Fe3+-mSF in the instrument’s magnetic field is promising for the translational relevance of this magneto-responsive, in situ aligning guidance architecture in neuroprotective, neuroregenerative SCI scaffold design. To our knowledge, this work is the first to investigate the magneto-responsive properties, minimally invasive delivery, and neuroregenerative potential of Fe3+-mSF, demonstrating a novel, MNP-free alternative approach in the design of in situ aligning nerve guidance architectures
Long-Horizon Planning and Control of Dynamic Whole-Body Locomotion
No full textRobust long‑horizon loco‑manipulation on legged robots demands trajectory optimizers that are simultaneously fast, physics‑faithful, and able to enforce hard constraints. Classical pseudospectral collocation promises spectral (exponential) accuracy on coarse grids—essential for planning over long time horizons---but its stage complexity and lack of an embedded feedback policy have prevented real‑time deployment. On the other hand, existing work in Differential Dynamic Programming exhibits remarkable linear time complexity, but suffers from poor conditioning over long time horizons when using forward dynamics and has great difficulty with handling arbitrary inequality constraints. This paper closes that gap by introducing a novel dual-layer optimal‑control solver that (i) reduces Radau IIA collocation to a fully decoupled block diagonal system based on a lifted Newton step using a novel lower‑triangular Jacobian approximation, (ii) adopts a condensed inverse‑dynamics formulation that preserves coarse‑grid fidelity while eliminating state variables, (iii) handles hard inequality constraints via an active‑set null‑space Riccati recursion, and (iv) yields a stabilizing whole‑body feedback controller "for free" from the same Riccati factors. The resulting solver unifies spectral accuracy, linear scalability, rigorous constraint handling, and closed‑loop robustness, all of which are key ingredients for real‑time, long‑horizon model‑predictive control of agile legged platforms
FATE: A Fourier Accelerated Tensor Engine
No full textThe growing demand for matrix multiplication in artificial intelligence must be met with increased tensor computing efficiency and bandwidth improvements. While digital hardware accelerators gained popularity, improving AI compute throughput, the potential of analog circuits remains largely untapped. In this Thesis, we introduce a novel Fourier-Accelerated CMOS-based Tensor Engine (FATE) that aims to optimize the high complexity of matrix multiplication with computation and bandwidth efficiencies leveraging the Fourier Dot Product. We produce and simulate a transistor level version of FATE including op amps, filters, and Digital to Analog converters. Our design reduces the computational complexity of matrix multiplication from the traditional O(N^3) to O(N^3/f), where f is a number of frequency carriers up to N. Our circuit dramatically reduces the bandwidth required for moving vectors by encoding a vector as a summed sine series routed via a physical wire. The test circuit, designed with a 180 nm standard CMOS process, achieves a strong dot product linearity with an R^2 value of 0.940. By efficiently encoding vectors and computing dot products, the circuit demonstrates competitive energy use on the 180 nm process node. Despite the nonlinearities imparted by the analog components, extensive testing on standard machine learning models showed minimal (less than 1%) performance degradation, with some models even demonstrating improved accuracy up to 10% over control models. This work highlights the untapped potential of analog circuits in modern AI, offering a highly efficient solution to a critical bottleneck in AI computation
Suggestive Audio Balance Tool
No full textAudio engineering is both a highly technical and uniquely artistic practice. Within this field, perhaps the most intense, heat-of-the-moment job is front of house engineering for live concerts. One of the duties of a front of house engineer is to balance the levels of the on-stage performers: each audio signal must be loud enough, but not too loud compared to other signals. Engineers, artists, and listeners each have individual preferences for this balance, so there is no single ground-truth for a mixture of music signals. Yet, each genre and style of music has some unquantifiable bounds for how present each voice or instrument should be in a mix, and artists want confidence that their live performance will be mixed within certain bounds. Thus, there is a need for an interdisciplinary Suggestive Audio Balance Tool (SABT) that can detect egregious level imbalances in a real-time audio mix based on provided examples of acceptable mixes. This thesis proposes an architecture for the SABT that incorporates Music Information Retrieval (MIR) datasets and audio features, classical machine learning and deep learning techniques, and statistical approaches. The result is a tool that can detect whether a stem is imbalanced with respect to other stems in a mix, based on a provided dataset of acceptably mixed songs
Machine Learning-Enabled Physical Layer Authentication via Radio Frequency Fingerprinting in Vehicle-to-Vehicle Networks
No full textAs technology evolves, new approaches toward wireless communication must be considered to support low-latency applications. Radio Frequency Fingerprinting (RFFP) can be used as a method of authentication to reduce latency for time-critical applications such as wireless vehicle communication. This dissertation presents a practical implementation of RFFP to generate novel models and datasets that improve existing RFFP models and implementation of RFFP-based authentication to reduce authentication latency in networks that require fast situational awareness, such as Vehicle-to-Vehicle networks. This dissertation implements a novel RF-hardware testbed to emulate wireless communication transmitters that generate datasets to train machine learning RFFP models. The performance of each radio is analyzed across various channels and at different ambient temperatures to measure each factor's impact on the RFFP in a controlled environment. The models are deployed in a real-time communication system to test the RFFP authentication, which shows that the methods and models used in this dissertation improve RF classification and generalization accuracy across various channel environments and persist over multiple days. This research provides several key contributions. The testbed developed is a novel method for generating datasets and characterizing wireless transmitters. The models presented show improved generalization accuracy that spans multiple days and various channel environments, reducing the need for retraining. Finally, implementing RFFP ML-based authenticators reduces latency while maintaining high accuracy on a wireless network
Improvement to Quality of Experience in Cloud-Based Game Streaming with Jitter Buffer Management
No full textCloud-based game streaming has the potential to deliver high-quality gaming experiences anywhere by streaming game frames as video from servers to clients. However, frame jitter from bandwidth and delay variability can substantially degrade the user experience. While traditional streaming systems use playout buffer to smooth over frame variability, playout buffer algorithm in cloud-based game streaming remains underexplored. This thesis investigates the effectiveness of playout buffer algorithms for cloud gaming through the implementation of a custom game streaming platform and an automated Flappy Bird-style game. We evaluated two buffer strategies, the E-Policy and the Queue Monitoring (QM) under controlled jitter conditions. By applying existing Quality of Experience (QoE) models, we measure how the critical metrics of delay and interrupts influence users’ perceived gaming quality. Our results suggest that the E- Policy reduces playback interruptions but causes higher delay, which lowers overall QoE, especially when network jitter is high. On the other hand, the QM algorithm with higher decay values keeps a better balance between delay and playback smoothness, leading to higher combined QoE across different network conditions
Statistical Observations on Water Quality Pre and Post Lead Service Line Replacements
No full textLead service lines are a common water infrastructure problem throughout the United States. Lead service lines can corrode over time and release lead into the water supply, causing public health problems for customers. Federal legislation set by the U.S. Environmental Protection Agency requires water systems to replace lead service lines when water lead levels are elevated. However, replacing lead service lines disrupts corrosive scales on lead pipes and can increase water lead levels post-replacement. This research investigates lead concentrations and water quality in consumer taps before and after the replacement of lead service lines through statistical analysis using analysis of variance (ANOVA), correlation tests, and linear mixed effects modeling (LMEM). The ANOVA results indicate meaningful differences between the means of groups, showing higher mean lead levels post partial replacements compared to post full replacements, as well as higher mean lead levels short term after replacement compared to long term. While the correlation and LMEM results were not statistically significant, the LMEM showed that higher pH lowers the probability that lead level is above the action limit and higher temperature increases the probability that lead level is above the action limit. These results can help inform decisions regarding water quality monitoring needs and potential health risks from lead as LSL replacements occur throughout the U.S
Music Video Project Introducing SuZhou City With a Focus on 2D Shaders
No full textThis project is a short music video that transforms fully modeled Suzhou-style architecture into a living Chinese ink-wash scroll. Through Blender’s real-time EEVEE renderer, a reusable node group—two-color gradient, stretched noise for vertical streaks, a muted warm tint, and a solidified outline mesh—transforms all 3-D surfaces into brush-like strokes, but also keeps iterations snappy. Lyric animation and a fine rice-paper grain are overlaid in After Effects, combining music, motion graphics and painterly visualizations. The toolkit originates from user-friendly tutorials and illustrates how procedural approaches that are accessible to beginners may marry traditional look with modern real-time tools in a single, efficient pipeline
Toward Microelectronics Assurance using Impedance Sensing
No full textWith the globalization of electronic systems fabrication, driven by reduced costs and faster time-to-market, different steps of design, fabrication, and packaging are no longer completed under the same roof. Hence, systems are exposed to adversarial threats through different stages of the supply chain, from manufacturing and assembling to deployment in the field. Examples include eavesdropping on communicated data and obtaining backdoor access through spy chip implants or hardware Trojans. Physical attacks such as side-channel analysis (SCA) and fault injection (FI) require tampering with the system at various abstraction levels, from the printed circuit board (PCB) to the package and integrated circuit (IC). The use of counterfeit or low-quality components for illegal profit further threatens system performance and reliability. Additionally, the growing use of flip-chip technologies and inadequate IC backside protection increased the vulnerability to physical attacks from the chip's backside. As embedded electronics play a critical role in applications ranging from smartphones to autonomous vehicles and critical infrastructure, robust countermeasures are essential to ensure the integrity of microelectronic systems. Beyond financial losses, such attacks pose serious safety risks, particularly in domains such as medical devices, where failures can result in injury or loss of life. Despite significant advances in addressing cybersecurity threats within the supply chain, verifying the integrity of operational hardware remains a persistent challenge. Current physical inspection techniques are often time-consuming, destructive, costly, and incompatible with legacy systems. Furthermore, there is a lack of a comprehensive framework covering different abstraction layers in the system from the board level down to the IC and package level. In this dissertation, we first review the state-of-the-art hardware verification methods in the literature and analyze their strengths and limitations. Inspired by methods known from the field of signal and power integrity, we present a comprehensive and systematic framework for system-level hardware tamper and counterfeit detection by demonstrating how the impedance characterization of the system's power distribution network (PDN) at different frequency bands can detect various classes of tamper events at different abstraction layers from board level down to the package and chip level. The proposed solution in this dissertation exploits the PDN impedance and reflection frequency response to detect tamper events, making the verification generic, non-invasive, and applicable to virtually all electronic systems. We conduct extensive experiments on counterfeit, aged, and tampered devices, including impedance measurements for various classes of printed circuit board (PCB), chip (e.g., hardware Trojans), and package-level tamper events required for conducting different side-channel, IC backside, or fault attacks. Conventional and on-chip network analyzers are leveraged to generate hardware signatures to characterize the system's impedance profile at different abstraction layers. We explain how embedded network analyzers, without any modifications to the system, can be deployed to extract the frequency response of the PDN. The analysis of these frequency responses reveals different classes of tamper events from board to chip level, and those required to expose the IC backside silicon. Different statistical metrics, such as the difference of means (DOM), Wasserstein Distance (WD), and dynamic time warping (DTW) metric, are used to distinguish between genuine and tampered samples and mitigate the impact of manufacturing process variations and environmental conditions
Steps Toward Sign Language-Centric User Interfaces Guidelines Through Culturally Aligned Research Practices
No full textResearchers have made efforts to localize online content; to provide culturally and linguistically aligned content taking into account aspects such as users’ languages, cultures, and ways of interaction. However, most of these efforts have not considered the Deaf culture which is a distinct culture with a shared language (e.g. American Sign Language in the United States), conventions, and experiences of deafness; that likely shape Deaf individuals’ perceptions, preferences, and visual attention of the user interface elements. Many efforts have been made to translate and increase the video quality of signed content and embed it onto text-based resources to enhance content accessibility. However, limited work has looked at videos beyond translation. De Galdo and Nielsen suggested that translation only is not enough to localize content and that understanding the cultural nuances and adapting the usability testing methods to the target culture are vital. Within this scope, and from the Deaf culture and “sign language-centric” design perspectives, this dissertation centers on: understanding Deaf individuals’ preferences and perceptions of customized video elements, styles, and layouts in and out of context; investigating user interfaces scanning patterns among deaf ASL-signers as a step to improve interface layout; and reporting best research practices when conducting studies with and for the Deaf Community. Investigating these aspects helps understand how Deaf individuals perceive and prefer these nontraditional video-based elements and how the context might impact their perceptions and preferences. Additionally, understanding the scanning pattern of static visual elements is an essential step to have a baseline for understanding how future signed content can be placed. Finally, adapting the research methods to align with Deaf culture, ensures the reliability of the conducted research. Therefore, beyond content translation, this work contributes to the field of human- computer interaction and accessibility by making a step toward design guidelines for sign language-centric user interfaces, taking into account the cultural perspective in the design and research practices