Data Reconstruction with Spatial and Temporal
Correlation in Wireless Sensor Networks
Yayun Zhang
College of Mathematics and
Computer Sciences
Fuzhou University,China
n140327044@fzu.edu.cn
Hongju Cheng
*
College of Mathematics and
Computer Sciences
Fuzhou University,China
cscheng@fzu.edu.cn
Dongyin Chen
College of Mathematics and
Computer Sciences
Fuzhou University,China
n140327035@fzu.edu.cn
ABSTRACT
Data loss is a common phenomenon in the practical wireless
sensor networks due to several observations, such as the node or
link failure, the expensive cost of re-transmission, and so on. How
to reconstruct the lost data is one important issue for many
applications which concern heavily on the data gathering issues.
Most of the current data reconstruction solutions are based on the
spatial correlation and they adopt interpolation methods to
approximate the data reconstruction process, which leads to bad
performance especially in case that the number of data loss
increases in the network. In this paper, we consider the spatial and
temporal correlation simultaneously, and propose a novel data
reconstruction algorithm to improve the data error ratio in the
wireless sensor networks. The proposed algorithm can utilize the
spatial correlation by using the curved face reconstruction, and it
is also helpful to improve the accuracy of data reconstruction by
exploring the temporal correlation among the sensory data.
Extensive experiments demonstrate that the proposed data
reconstruction algorithm can significantly reduce the fitting data
error ratio compared with related works especially in case that
data loss is serious in the network.
CCS Concepts
• General and reference➝Estimation • Information systems➝
Sensor networks.
Keywords
Wireless sensor networks; spatial correlation; temporal correlation;
data reconstruction.
1. INTRODUCTION
A wireless sensor network generally consists of a large number of
nodes which are equipped with the sensing devices, micro-
processor, limited memory and wireless communications [1].
They can collect information from the local environment and send
the sensory data to the application center, i.e., the sink. However,
data loss might occur when data is gathered to the sink due to the
following observations. Firstly, the trend for cheaper sensor nodes
might result in increased probability of node failure. In this case,
the sensor node collect cannot obtain all the original data
especially when it works in a periodical manner. Secondly, the
congestion in the wireless communications, such as the exposed
terminal problem and hidden terminal problem, shows that the
data transmission is unreliable, which means that the sink will not
receive all the sensory data sent by nodes in the network. Thirdly,
data re-transmission is a little expensive in the wireless sensor
networks because the nodes are generally budged with un-
rechargeable battery. For example, the data loss ratio is about 23%
in the practical collected data set for Intel Indoor project [2]. How
to reconstruct the lost data is one important issue in the wireless
sensor networks.
Similar to the definition in [3], we synthesize several types of data
loss patterns by analyzing the data set in Intel Indoor project [2],
i.e., element random loss, block random loss and element
sequence loss. The element random loss describes the case that
data is lost randomly during any given collection period. In this
case, the probability of data loss for different nodes is independent.
Figure 1 demonstrates an example, in which the hollow circles
denote nodes whose data is lost (meaning that the sink does not
receive the data due to node/link failure), and the black solid
circles denote nodes whose data is correctly sent to the sink. The
block random loss is used to illustrate the case that a number of
nodes in the same block have lost all their data simultaneously at a
random time. As shown in Figure 2, nodes numbered from 1 to 6
are in the same block, and it is called block random loss if they
lose the data simultaneously (represented by hollow circles to
show the data loss). The element sequence loss shows that the
collected data for one single node is lost during the continuous
time duration. The sequential data loss might mean node failure in
the network when it is detected.
Most of the applications in the wireless sensor networks depend
on data analysis by gathering information from nodes in the
networks to the sink [4, 5]. The presented data loss may influence
the efficiency or feasibility of data analysis in these applications.
Some works aim at providing the approximate query in which
data accuracy is not the key issue and thus the application/sink
only needs to know the approximate data in the network. However,
data loss is not considered and solved in these works, which
means approximate query will encounter with the problems if too
much data is lost.
Most of the current data reconstruction algorithms, such as the
traditional KNN [6], the inverse-distance weighting [7] and the
Delaunay Triangulation [8], depend on the spatial data correlation
in the sensor networks. However, the physical phenomenon
observed in the network is both spatial and temporal correlated
generally. The different data observed by adjacent nodes is similar
and it has the same variation tendency. At the same time, the
value of sequential sensing data for one node is generally
continuous when the collection duration is small enough. Some
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee. Request permissions
from Permissions@acm.org.
MSCC'16, July 05 2016, Paderborn, Germany
© 2016 ACM. ISBN 978-1-4503-4343-5/16/07…$15.00
DOI: http://dx.doi.org/10.1145/2940353.2940360
40