Theory and application of a data-driven approach to compressive spectrometry in the assessment of neurophotic stimulation

Recent advances in biology and neuroscience have elucidated pathways in which visible light initiates a signaling process, via the eye, responsible for activating direct and indirect biological responses. While the full impact of unnatural light patterns on human biology remains poorly understood, controlled laboratory studies, and a handful of field studies, have strongly indicated that exposure to light (and darkness) outside of the natural light-dark cycle is a factor in increased disease prevalence and general malaise related to anomalous wake and sleep behavior enabled by modern lighting. Consequently, there exists a mismatch between our evolved needs – based on the natural day-night cycle – and the needs driven by modern lifestyle enabled by technology – which typically include an over exposure to artificial light at night and restricted access to natural light during the day. In the same way that our nutritional diet has evolved with modern convenience, so has our spectral diet. From a neurobiological perspective, we can think of discrete instances of spectral composition and intensity of light received over the day as micronutrients, that when concatenated over a 24h period, compose a diet. Relating spectral diet information to neurobiological outcomes (i.e., "neurophotic" effects), is a central aim of multidisciplinary research teams. Unfortunately, collecting spectral information in the field is challenging due to constraints associated with existing encoders. The central aim of this thesis is to circumvent these constraints by approaching spectral sensing from the angle of compressed sensing and information theory by developing SpecRA: an adaptive reconstruction algorithm. In order to demonstrate the efficacy of SpecRA in practice, we develop a novel compact spectrometer \textit{Spectrace}, in tandem and test its performance against existing state-of-the-art devices. The approach taken herein allows us to maximize the amount of information recovered from cheaper, low-fidelity, filter-array sensors by pushing beyond the limits of conventional signal reconstruction though the exploitation of regularities in the natural world. As a first application of this research, we investigate the diversity of so-called spectral diets in a limited subset of commuters with the aim of elucidating structures and properties of modern spectral diets in populations with shared behaviors, in order to better anticipate their prevalence and potential impact on health and well-being. Indeed, by applying this methodology to other domains, we could also develop more efficient sensing methods of other types of chaotic signals thereby mitigating constraints and enabling greater opportunity for data collection across a diversity of physical phenomena.