A main challenge in sensor networks is to derive a precise model of the sensors and environment.  This interface between the physical and virtual worlds determines everything about an application from sensor placement to noise filters to algorithmic design but is very difficult to model correctly, especially when an application must be deployed quickly with new sensors and in a new or complex environment.

In this talk, we present a technique called Statistical Emulation with which we can quickly collect an empirical profile of our sensors and environment and use that profile directly in simulation, allowing an algorithm to be evaluated, tuned, and redesigned based on real-world data rather than a theoretical model.  This technique can also be used to incorporate only certain aspects of the physical world into a simulation, allowing the algorithm designer to systematically identify which aspects of the environment are causing errors or problems with the algorithm.

We demonstrate this technique through the task of sensor localization.  We systematically identify subtle aspects of our real world data that cause large errors in localization algorithms from the literature. Through this analysis and by capturing an accurate representation of the real world in simulation, we achieve the first successful demonstration of ranging-based localization of a sensor field using radio signal strength (RSS), which has been notoriously unpredictable and difficult to achieve due to complex characteristics of RSS and the environment.  We conclude by demonstrating how Statistical Emulation solves similar problems in other sensor network applications like tracking.

--Academic Bio---------------------------------
Kamin Whitehouse has earned a M.S and is currently a Ph.D. candidate in Computer Science at UC Berkeley, where he is working on wireless sensor networks with Prof. David Culler.  His research has also extended to several industrial research labs, including Microsoft Research, PARC, Intel Research, and AT&T Bell Labs.  Kamin graduated from Rutgers University with a dual degree and triple major, including a B.S in Electrical and Computer Engineering as well as a B.A. in Cognitive Science and Philosophy.  He has been awarded the NSF, NDSEG, GOF, and Siebel fellowships for his work, and was awarded Carr, AFCEA, Bell Atlantic, and Xerox scholarships as an undergraduate.

Thursday, April 13, 2006

Wu & Chen Auditorium

101 Levine Hall

3:00pm - 4:15 pm