Highly sensitive very low frequency sensing coils are a fundamental component of many radio frequency receiver systems, with practical applications ranging from ultra-low field MRI to underwater signal detection. However, the traditional cascaded coil design has begun to reach its limitation in regard to sensitivity. Due to ohmic resistance, the voltage drops throughout a coil, introducing parasitic capacitance between each turn. Parasitic capacitance lowers the inductance of the coil, directly limiting coil sensitivity. This project aims to test a novel design known as the equal-potential coupled (EPC) coil. By splitting a coil into segments and applying an equal voltage to each segment, all segments are equipotential. This reduces parasitic capacitance, improving coil sensitivity. Using Ansys HFSS and CST Studio, we iteratively modeled and analyzed various coil designs. Maximizing the Q factor leads to improved coil sensitivity, and was therefore the focus of our coil comparisons. Initially, we hypothesized that the EPC coil’s Q factor would be approximately 4x that of the cascaded coil. The simulation results were inconclusive, with the plotted Q factors of the cascaded coil, parallel coil, and EPC coil being very similar. A small decrease in the inductance suggests that proximity effect was present, but very weak due to high spacing between coil turns. Future work will revolve around modeling more compact coils in order to observe the full impact of capacitance on the inductance of different coil geometries.

Highly sensitive very low frequency sensing coils are a fundamental component of many radio frequency receiver systems, with practical applications ranging from ultra-low field MRI to underwater signal detection. However, the traditional cascaded coil design has begun to reach its limitation in regard to sensitivity. Due to ohmic resistance, the voltage drops throughout a coil, introducing parasitic capacitance between each turn. Parasitic capacitance lowers the inductance of the coil, directly limiting coil sensitivity. This project aims to test a novel design known as the equal-potential coupled (EPC) coil. By splitting a coil into segments and applying an equal voltage to each segment, all segments are equipotential. This reduces parasitic capacitance, improving coil sensitivity. Using Ansys HFSS and CST Studio, we iteratively modeled and analyzed various coil designs. Maximizing the Q factor leads to improved coil sensitivity, and was therefore the focus of our coil comparisons. Initially, we hypothesized that the EPC coil’s Q factor would be approximately 4x that of the cascaded coil. The simulation results were inconclusive, with the plotted Q factors of the cascaded coil, parallel coil, and EPC coil being very similar. A small decrease in the inductance suggests that proximity effect was present, but very weak due to high spacing between coil turns. Future work will revolve around modeling more compact coils in order to observe the full impact of capacitance on the inductance of different coil geometries.

Brain computer interfaces (BCIs) are a rapidly growing field with the aim of creating machines that collect brain signals and translate them into desired actions. Applications of such work are widespread, from clinical work with paralyzed and amputated patients to creating smoother systems for virtual reality users. However, current state-of-the-art techniques require invasive placements of electrodes to record neural activity, which demands much and incurs significant risk for a patient. My project focuses on using non-invasive surface electromyography (sEMG) signals that measure activity from motor neurons in the forearm. We use deep neural networks to decode these signals for applications such as real-time natural typing and cursor control, with goals of a functional prosthetic hand for amputated patients. The primary challenge is to train neural networks to correctly decode actions. This involves resolving meaningful neural signals in the presence of noise due to several factors, including interference from skin and interfering signals from other muscles in the hand. Furthermore, due to the physiological variability of humans, making such a network robust across patients provides a challenging task. We use convolutional neural networks to extract spatial-temporal information from the signal. For downstream tasks such as natural typing, we use a character-level language model in combination. We achieve 95% decoding accuracy in classifying 1 of 5 fingers, and can successfully play Pac-man and are working towards proficient typing. This work demonstrates that non-invasive sEMG can be used to play games and likely allow users to type naturally.

Retroreflectors, known for their widespread use in road signs and safety gear, serve a critical function in enhancing visibility. In this context, we explore their application in visible light communication (VLC), aiming to leverage their reflective properties for information transfer. Expanding on this concept, we devise mechanisms that encode distinct signal patterns generated by human interactions, thereby enabling self-sustaining smart sensing capabilities and creating dynamic interfaces for physical environments. By turning away from the reliance on conventional systems such as electronics or external power sources, our approach embeds relevant data seamlessly into the environment. To evaluate this, we developed a mobile app that captures light intensities over time at a specified location within the camera view and investigated diverse fabrication methods related to retroreflective materials. At a small scale, 3D-printed remote controller mechanisms such as buttons, rotating knobs, and sliding switches validated our concept. At a medium scale, vinyl-cut retroreflective barcodes encoded information for indoor surfaces. At a large scale, CNC-machined arrays of triangular foam prisms used for city signage dynamically altered conveyed information based on viewing perspectives. Results demonstrate that our smart retroreflector design can enrich people’s interaction with their surroundings, promoting efficient self-sustaining information distribution.