Database Reference
In-Depth Information
ena simultaneously). ECLUN also tries to uniformly distribute the en-
ergy usage among the nodes, resulting in a longer lifetime for the entire
sensor network, since the variance of the lifetime of individual nodes is
minimized.
A more recent study [4] focuses on the problem of identifying func-
tional dependencies among sensor data streams, in order to determine a
small number of sensors from which data are actively collected. The rest
of the sensors collect data at lower rates, with the purpose of detecting
changes in the discovered dependencies and taking actions to reorganize
the sensor data collection process. The dependencies identified in this
work are based on regression analysis that takes into account possible
lags among the streams.
The above studies use different ways of calculating the correlation
among the sensor streams in the network. For this part of the problem,
other techniques for identifying correlations in multiple data streams [107,
114, 72, 26, 82] could be used as well. The work by Aggarwal et al. [3] de-
scribes a method that additionally considers and exploits domain-specific
knowledge on the information network of the sensors (i.e., relating to
links among the sensors). Another approach for the same problem has
proposed a technique for selecting sensors that is based on feedback on
the utility of the selected sensors [43].
2.3 Data Series Summarization
Many sensor network applications in diverse domains produce volumi-
nous amounts of data series, such as in meteorology (e.g., temperature
measurements [1]), oceanography (e.g., water level measurements [90]),
and other domains. The sheer number and size of the data series we need
to manipulate in many of the real-world applications mentioned above
dictates in several cases the need for a more compact representation of
data series than the raw data itself, and a plethora of representations
have been proposed to that effect
4
.
Even though most data series representations treat every point of
the data series equally, there exist WSN applications for which the time
position of a point makes a difference in the fidelity of its approximation.
Then we would represent the most recent data with low error, and would
4
Several techniques have been proposed in the literature for the approximation of
data series, including
Discrete Fourier Transform (DFT)
[76, 36],
Discrete Cosine Trans-
form (DCT)
,
Piecewise Aggregate Approximation (PAA)
[106],
Discrete Wavelet Transform
(DWT)
[75, 21],
Adaptive Piecewise Constant Approximation (APCA)
[20, 58],
Piecewise
Linear Approximation (PLA)
[54],
Piecewise Quadratic Approximation (PQA)
[48], and
others. Most of them are amenable to incremental, online operation.
