Database Reference
In-Depth Information
2.2 Power Issues in Sensor Collection and
Transmission
Since sensor nodes have limited battery power, the issue of data col-
lection and transmission is a challenging one for sensor networks. Sensor
nodes have limited battery power, as a result of which it is necessary to
collect and transmit the data on a limited basis. This means that it is
often necessary to collect only a subset of the data for mining purposes.
A classic technique which is often used in order to reduce the amount
of collected data is
sensor selection
. In data driven sensor selection,
goal-oriented techniques are used in order to reduce the amount of data
collected for mining purposes [19, 2-1, 45, 61]. In most of these methods,
the idea is to use the massive redundancy across the different sensors in
order to reduce the total data which is collected. For example, in an
environmental monitoring applications, two sensors which are located
physically close together may often have very similar readings. In such
cases, the correlations across the different sensors are leveraged in or-
der to select a small number of sensors from which the data is collected.
The values on the other sensors can be predicted from this small set. We
note that this small set may vary over time, as different sensors may be
inoperative at different times, or the correlations among the data may
change. Methods for power-ecient and dynamic sensor selection are
discussed in [1]. Another technique which is often used in order to re-
duce the power transmission costs is a method referred to as
in network
processing. We will discuss this technique in the next subsection.
2.3 In-Network Processing
Since sensor networks may use hundreds and thousands of nodes over
a large area, and all the data from the different nodes may be need to
be fused, this can incur significant communication costs, of all the raw
data is directly transmitted to a central server for processing. While
such a naive solution is easy to implement, its energy costs may be too
large to make it practical for very large scale sensor networks which are
distributed over wide regions. Since the cost of transmission is higher
than computation, it is usually advantageous to organize the sensors into
clusters. In such an environment, the data gathered by the sensors is
pro-
cessed within the network
and only aggregated information is returned to
the central location. This responsibility is typically provided to certain
nodes in the network, which are referred to as
aggregators
.Numerous
functions can be designed for such nodes, and the corresponding data
sent may depend upon the relevant aggregate queries which are posed
at the central server. Thus, the underlying assumption is that such em-
