Intellectual Merit: This BRIGE project is aimed at the investigation of signal processing theory and methods for blindly calibrating sensors on a massive scale, after they are deployed, and as their calibrations change over time. The massive deployment of sensors is exciting for the prospect of what may be learned from the data; a critical challenge in using those data is knowing their quality and reliability. In deployments of even a hundred sensors, it becomes infeasible to hand-calibrate each one in order to maintain confidence in the sensor output. Thus blind calibration, i.e. calibration without the need for controlled stimulus or high fidelity ground-truth data, is of critical importance. This proposal focuses both on theory of blind calibration and the concrete practical application to air quality sensing. Realistic sensing environments are considered, where phenomena are non-stationary, and data are streaming, with corruptions and missing values. A calibration dataset will be collected and shared with the community. The two major theoretical contributions will be (1) extending theory of blind calibration to models which capture the great variety exhibited by environmental phenomena, and (2) online modeling for the practical scenario where the phenomenon of interest is non-stationary.
Broader Impacts: Environmental sensing is an important contemporary application of statistical signal processing that attracts a great deal of interest from people with diverse backgrounds. This application gets people involved in the technology, the climate, and their local community; therefore it has potential to truly broaden participation in engineering. The syllabus for Digital Signal Processing at the University of Michigan, a central course in all signal processing curricula, will be expanded to include this application and useful fundamentals like spatial models, auto-regressive models, and matrix decomposition. Additionally, signal processing for environmental sensing in particular has the potential to attract students who may typically go into science or environmental policy, but who have a strong interest in mathematics as well. Interaction with the Marian Sarah Parker Scholars program at Michigan will expose outstanding young female scientists to the wide variety of possibilities with signal processing. Beyond the classroom, exploring one's environment using sensing has the potential to make a very positive social impact. The projects developed for the Parker scholars and other Michigan educational programs will build an infrastructure for future interactions with all grade levels and the greater public.