In order to build a self-sustainable wireless sensor network, powering sensor nodes with energy harvesting devices becomes a natural and feasible solution, thanks to the recent progress on energy harvesting technology. However, wireless sensor networks usually need to collect and transmit vast amounts of data. Utilizing the random, non-uniform, and scarce harvested energy intelligently to meet the energy demand caused by 'big data' is extremely challenging. The real-time latency requirement for data delivery makes the problem even more complicated. This project overcomes these challenges through strategically allocating the stochastic energy resources to collect and transmit the most important sensor data, providing a reliable pipeline for data collection, transmission and analysis in energy harvesting sensor networks. The project is expected to have direct impact on the design and wide deployment of energy harvesting wireless sensor networks, with applications in radio spectrum management, environment monitoring, healthcare, surveillance, disaster relief, etc. Furthermore, the proposed work has widespread potential applications far beyond energy harvesting wireless sensor networks, such as automatic residential energy consumption scheduling in smart grid, micro-grid planning with renewable energy inputs, information extraction from astronomical amounts of data collected from large-scale participatory sensing, etc. The proposed project integrates a research agenda with a strong educational component and will serve the following educational purposes: 1) the PI will use the proposed project to stimulate and maintain undergraduate students' interests in engineering via undergraduate research, senior project design and project based learning; 2) the requested funding will be used to train graduate students via curriculum development, proper research mentoring and industry collaboration; and 3) the proposed project will be used to attract potential engineering students through various outreach activities.
The goal of this project is to construct a new paradigm of sensing and transmission schemes in data-intensive energy harvesting wireless sensor networks to intelligently utilize the random, non-uniform, and scarce harvested energy with analytically provable sensing, transmission and inference performance guarantees. Two different but closely coupled approaches are proposed to achieve this goal. One is an energy-driven approach and the other is a data-driven approach. For the energy-driven approach, the statistics of the energy harvesting process are exploited to construct online sensing and transmission schemes. Two main tasks addressed in this research thrust are objective-oriented cooperative sensing scheduling policies to cope with the non-uniform energy supply in large-scale sensor networks, and delay-constrained data transmission schemes to meet the real-time latency requirement. The data-driven approach utilizes the characteristics of the underlying sensing phenomena to adaptively and strategically allocate scarce energy resources for the collection and transmission of the most important sensor data. Two specific tasks in this research thrust include a Gaussian process based framework to systematically utilize the spatial-temporal correlations in sensing fields, and adaptive sensing strategies that exploit the structured sparsity of underlying sensing signals, both under the stochastic energy constraints at sensors. The project promises to build intelligent energy harvesting wireless sensor networks with superb sensing, transmission and inference performances on a solid analytical foundation. The delay-constrained information theoretic analysis for energy harvesting communications will integrate a new set of analytical tools from renewal theory with tools in information theory, and create synergies between them. The proposed data-driven adaptive sensing approach will develop synergies between stochastic queueing control and high-dimensional data analysis. The interdisciplinary nature of the research allows us to utilize techniques from stochastic queueing control, renewal theory, information theory, machine learning and is expected to advance the understanding of those areas.
|Effective start/end date||7/1/16 → 7/31/21|
- National Science Foundation: $481,386.00