Database Reference
In-Depth Information
objective is to extend the WSN lifetime, while fulfilling the application
requirements (collecting the required data, or answering the queries).
There are two main ideas that researchers have explored: first, data
are correlated (both across time and over space), and second, several ap-
plications accept small errors in the data values they operate on. These
ideas have led to the development of a multitude of techniques that trade
accuracy for time performance and energy savings.
In this study, we review the efforts of the research community with
respect to the problems of real-time collection of the sensed data, and
real-time processing of these data series in the context of a WSN.Fur-
thermore, we examine the interplay between such data management
techniques and network protocols.
We note that the aim of this study is not an exhaustive enumeration
and discussion of all the related works, but rather, the description of
prominent research problems that have been studied so far with regards
to the sensor data processing and analysis, as well as of promising future
research directions.
2. Data Collection
The availability and use of sensor networks have generated a lot of
research interest. A major part of this effort has concentrated on how to
collect the sensed data at the sink (where they will be further processed
and analyzed), using the least amount of energy
1
possible. The challenge
arises from the special characteristics of WSNs and the nature of the
data they produce, namely: limited resources, intermittent connections,
and spatio-temporal correlation of the sensed values [60, 56, 101].
Several frameworks for the ecient execution of queries and collection
of data in a sensor network have been developed in the last years [60,
59, 103]. The focus in these works was to propose data processing and
optimization methods geared specifically toward sensor networks (we de-
scribe those in detail later on). The early studies described in-network
aggregation techniques for reducing the amount of data transmitted by
the nodes, while subsequent research focused on model-driven [32] and
data-driven [87] data acquisition techniques. Other works have proposed
techniques that take into account missing values, outliers, and intermit-
tent connections [44, 30, 101, 88].
A different approach is based on Kalman filters [51], with the same
goal of reducing the required communication among nodes and the sink.
1
Given that radio communication in
WSNs
is much more expensive than CPU processing,
this translates to reducing communication and data transfer.
