Location-Centric Distributed Computation and Signal Processing in
Microsensor Networks
Sponsor: DARPA SensIT
Program
Microsensor networks are envisioned to provide a seamless link between
the
physical world and the global information infrastructure. The link is
established via a dense deployment of cheap, small sensing devices that
serve as
a virtual skin for the surface of an object or a geographic region.
The
devices are able to sense in various modes but have limited computation
and
communication capability.
The challenges in the design and operation of microsensor networks are
three-fold:
- Simple, flexible programming abstraction: Each
sensor
device by itself may not be able to provide useful information without
collaboration with other devices. At the same time, due to the large ad
hoc
nature of sensor networks, it is a formidable challenge for a programmer
to
develop efficient distributed algorithms and implementations without a
simple,
but flexible, programming model.
- Power and bandwidth efficient distributed signal processing:
Each
device is likely to have very limited power and bandwidth capabilities
to
communicate with other devices.Therefore, any distributed computation on
the
sensor network must be very efficient in utilizing the limited power and
bandwidth budget of the sensor devices.
- Robustness to sensor device failures: Due to the harsh
conditions
in which sensor devices may be deployed, and the way in which the
devices may
be deployed, one can expect a significant fraction of the devices to be
either
non-operational or malfunctioning. Therefore, the underlying distributed
algorithms must be robust with respect to a large number of device
failures.
The overall goal of this project is to develop novel solutions
to
overcome these three challenges. Our strategy is based on three key
premises:
- Distributed computations in a microsensor network require
collaboration
among devices in the same geographic region as opposed to
collaboration
among a specified set of devices. That is, computations in microsensor
networks are location-centric and not node-centric.
- Different tasks require sensor signal processing at different levels
of
accuracy and, hence, have different power and bandwidth requirements.
For
example, reliable detection may be accomplished at a lower level of
accuracy
whereas reliable classification would typically require a higher
level
of accuracy. This motivates multi-resolution spatio-temporal
processing
of sensor signals.
- A built-in self test methodology is essential in microsensor
networks so that most faulty sensor devices can be identified and
disregarded
during distributed computation and signal processing, thereby resulting
in
more efficient and robust algorithms.
Last updated on . Please direct your comments concerning this website to sensit@ece.wisc.edu.