Ambient Intelligence Through Agile Agents - Ambient Intelligence for Scientific Discovery - page 302

dumb sensors wired directly to a centralized processing unit. WSNs differ in that
the individual sensing nodes have the capacity to perform processing and sens-
ing on the same device, as well as wirelessly communicating with other nodes
[10]. Because of their ability to communicate, they can collectively monitor a
much larger area with more accuracy than any one device alone could. Localiza-
tion of environmental perception allows for much greater scalability and reduced
deployment costs.
The primary issue at the heart of WSN research is that individual sensor
nodes are resource constrained in all aspects of their operation. Currently the
most commonly available sensor nodes use 8-bit processors, have 128 kilobytes
of memory, use 10-100 kb/s radios and are approximately priced at 100 per

unit, though there are more powerful (and hence expensive) devices available.
Examples of these devices include Crossbow Motes, Ember Corporation, VTT
Soapboxes, Modules and Sensoria's sGate and netGate products. All vendors
are engaged in a race to produce devices with comparable specifications at less
than

e

1 per unit, which would make wide-scale commercial deployment possible.
Expected uses include environmental monitoring, provision of low-profile secu-
rity systems, a broad spectrum of scientific data collection task and Ambient
Intelligence applications.
Ordinarily the computation resources of a node would be sucient to perform
many of the tasks that WSN nodes would be expected to do, such as signal
processing, in-network packet routing and collaborative calibration. However,
a more fundamental problem exists in the form of power limitations. Thus far
there have been no sensor devices created that can generate enough energy to
run from ambient sources (e.g. vibrational energy [8]), and so they must all rely
on capacitors or batteries. Thus there is a trade-off between any activities the
node might wish to perform and the lifetime of the node. With anything other
than radioisotope-based batteries [20], maximum expected lifetimes are a few
years at best. In practical terms, this means that the nodes spend most of their
time in a low-power sleep mode, which further reduces their computing capacity.
Much of the research into WSNs focuses upon power-ecient methods of
running the network. In particular, ecient use of the radio is critical. Nodes
usually transmit summary statistics of data rather than the raw data itself, and
as the data packet is routed through the network its contents are often modified
to include additional data from each relaying node. Algorithms that limit the
scope of their input to nodes that are close neighbors of the node running the
algorithm are particularly beneficial [31], since they limit the degree of relaying
transmissions in the network.
There are a number of other critical areas that must be dealt with. One of the
chief properties of WSNs is that they are often used where human supervision
and maintenance is dicult or impossible. Even in cases where the network is
easily accessible, the number of nodes deployed (which can run into the thou-
sands) may prevent any practical degree of manual intervention. This means that
an ability to auto-configure and autonomous maintenance is a definite advan-
tage in a WSN. A network of nodes that can decide what their routing topology,

e