Research‎ > ‎

Compressive Sensing

People

Faculty:
Justin Romberg, Mani Srivastava, Jim Xu
Post-docs: Ashwin Lall
Students: Haiquan (Chuck) Zhao, Nan Hua, Shikhar Suri

Motivation


Efficient data collection is of paramount importance for body-area sensors. In future deployments, these sensors may be used to continuously monitor an individual's physiological data for prolonged periods -- in the order of months -- without being recharged. In light of these stringent constraints, it is crucial to make data collection as energy-efficient as possible.

Since much of the data collected will be redundant or predictable (e.g., during periods of sleep), it is important to conserve the local energy resources for interesting events -- ones which deviate significantly from the norm. Additionally, we expect several of the measurement modalities to be correlated (e.g., heart and respiratory rates), which may allow the reconstruction of some signals from a subset of the rest. This correlation structure may also allow us to turn off most of the sensors and let the remaining sentry sensors alert the others when interesting events are detected.

Besides collecting the long-term physiological data, an important aspect of this project is to monitor the subjects for unusual activity so that responses can be elicited from them (e.g., "Did you recently experience a panic attack?"). Reliably and efficiently detecting such events will be a key component of future field studies.

Plans


We have identified five research tasks that we are going to tackle in the next year. We aim
  • to develop a smart sampling algorithm for measuring and estimating the average heart rate during a 2-minute window, which is important for detecting and forecasting the stress episodes of the test subjects, without assuming that the sampling hardware is equipped with compressive sampling capability;
  •  to develop smart sampling algorithms for measuring and estimating other sensing modalities (e.g., skin conductance) that do not need the aforementioned compressive sampling capabilities;
  • to develop better algorithms for measuring other sensing modalities when compressive sampling capabilities are available;  
  • to exploit the correlation structure between different sensing modalities so as to be able to turn off various sensors and still reconstruct their values from the other measurements; and  
  • to develop techniques to automatically check the fidelity of data and distinguish corrupt data from novel events.

Latest Work


Studies will begin in the first quarter of 2010.  Please return to this page for updates on our progress.