Received: 27 March 2020 Revised: 6 August 2020 Accepted: 9 September 2020

DOI: 10.1002/dac.4623

RESEARCH ARTICLE

A hybrid adaptive spatiotemporal compression scheme in

cluster-based wireless sensor networks

Samia Al Fallah | Jihane El Aasri | Mounir Arioua | Ahmed El Oualkadi

Laboratory of Information and

Communication Technologies, National
Summary
School of Applied Sciences, Tangier, Energy efficiency and data reliability are the crucial issues to be explored in
Morocco
wireless sensor networks (WSNs). A large amount of power consumption is
Correspondence depleted by each sensor node to transmit its sensed data which typically
Samia Al Fallah, National School of induces high temporal and spatial dependencies. Therefore, efficient compres-
Applied Sciences of Tangier, Route
sion approaches are designed to exploit the data correlation in both space and
Ziaten,Tangier BP 1818, Morocco.
Email: [email protected] time so as to discard information redundancies with accurate recovery. In this
paper, we propose a hybrid and adaptive spatiotemporal compression
approach named Distributed Temporal Source Coding, designed to extend the
lifespan of cluster based WSNs. The proposed scheme takes advantage of the
performed Distributed Source Coding as an effective spatial compression
algorithm and adaptively exploits the Lightweight Temporal Compression as a
promising solution to reduce the temporally correlated information. The
obtained results have revealed that the proposed algorithm outperforms other
compression methods, promotes the network battery life, and ensures compel-
ling data accuracy. The proposed scheme has succeeded to reduce the gap
between the energy efficiency and data reliability in WSNs.

KEYWORDS
data reliability, DSC, energy efficiency, LTC, spatiotemporal compression, WSNs

1 | INTRODUCTION

During the last years, wireless sensor networks (WSNs) have experienced a large diffusion owing to their unique char-
acteristics, low-cost, adaptivity, and autonomy.1 This promising technology is used to measure environmental parame-
ters in various fields such as health care, meteorology, industry, military, smart grid and entertainment.2,3 Typical WSN
is composed of numerous, intelligent, cheap and small devices with four basic units: sensing, processing, transmitting
and power.4,5 With limited energy and simple computation capability, these sensors are randomly deployed in an inter-
esting area in order to collect the suited information and communicate it to the base station (BS). Energy constraint is
the major issue to be considered in improving WSNs performance.6,7 Several research works have been interested in
developing efficient and high performance energy harvesting systems for WSN environment.8,9 Some of these works
have been focused on designing efficient routing protocols to increase the network energy saving.10,11 Other works have
been appealed to data fusion algorithms to conserve the network energy by gathering and combining data from multi-
ple sources to achieve inferences.12 Practically, sampling, processing, and transmitting are the three operations mainly
responsible for energy consumption in WSNs. Any approach that directly or indirectly reduces the depleted power of
these operations, with respect to application requirements, is considered as compression.13 Since the communication

Int J Commun Syst. 2020;33:e4623. wileyonlinelibrary.com/journal/dac © 2020 John Wiley & Sons, Ltd. 1 of 17
https://doi.org/10.1002/dac.4623
2 of 17 AL FALLAH ET AL.

resources consume a large amount of energy in each sensor node, various research efforts have been focused on reduc-
ing the transmitted signal by designing adequate compression and aggregation techniques, so as to prolong the network
lifespan. Razzaque et al13 provide a comprehensive overview of existing compression approaches in the literature, with
a comparative analysis on the basis of the key performance metrics notably the energy efficiency, the computational
complexity, and distortion. This work presents also a logical classification of these techniques, considering the compres-
sion of sampling, data, and communication. Furthermore, the research work in Hooshmand et al14 adopts another clas-
sification of data compression considering temporal, spatial, and spatio-temporal methods. Temporal approaches
exploit the incurred temporal correlation in the sensed time series to discard redundant information,15 while the spatial
approaches are designed to eliminate repetitious information gathered from sensor nodes exploiting the spatial correla-
tion induced by the density of the network.16 Spatio-temporal algorithms combine these two approaches to reduce effi-
ciently the data transmission with accurate recovery. As the sensed time series of each node presents generally a
temporal dependency, and the observed signal of neighboring sensors at the same time slot is highly related for moni-
toring the same area, the collected data are highly correlated temporally and spatially.17 Effectively, lossy compression
approaches take advantage of these redundancies in space and time to discard additional information without affecting
the data reliability.14 Some of these methods have proven remarkable results in terms of energy saving owing to their
efficient data reduction but at the cost of some information loss. Other methods have been proposed to prolong the net-
work battery life by exploiting the data correlation in only one dimension (time or space).18 In this work, we propose a
hybrid and adaptive spatiotemporal compression approach Distributed Temporal Source Coding (DTSC), typically
applied in a clustered WSN. The proposed method presents a combination of Distributed Source Coding (DSC) as an
energy efficient and accurate spatial compression19,20 and Lightweight Temporal Compression (LTC) as a performed
temporal compression on the first hand21,22 and temporal correlation measuring tool on the other hand. DTSC is used
to compress the network data in double dimension to alleviate the communication burden which can significantly
reduce the overall power consumption. In addition, LTC is used as a temporal compression only when the temporal
correlation occurs. Hence, the information is retrieved in the BS with minimum error tolerance, which will reduce the
data distortion and promote the signal accuracy. As a result, the proposed scheme minimizes effectively the compro-
mise between the energy expenditure and information reliability in WSNs. The rest of the paper is organized as follows:
Section 2 reviews different contributions concerning the used compression schemes. Section 3 presents the system
model including the signal model, the network routing architecture, and the energy model for processing. The proposed
compression approach DTSC is detailed in Section 4 providing its operating system analysis. The obtained simulations
and results are discussed in Section 5. Section 6 concludes the paper with our main findings.

2 | R E LA T E D WOR KS

Spatial and temporal compression approaches have been explored in various WSNs applications in order to remove
information redundancies and promote the network lifetime. In this context, several methods have been carried
out to examine spatial as well as temporal features of sensor data. DSC has proven its efficiency to compress spa-
tially correlated signals with minimum information loss and without internode communication.23,24 DSC theory
was firstly studied and proposed in Slepian and Wolf,23 where authors stated that as long as joint decoding is per-
formed, the limit of a disjoint compression is the joint entropy of the sources. This limit is the same as when the
joint encoding of the sources is executed.23 Several research works and extensions have been produced from this
significant result, such as the lossy version of DSC with distortion constraint, where authors considered the use of
one source as side information at the decoder.24 It has been shown that for a particular correlation, where source
and side information are jointly Gaussian, there is no performance loss due to the absence of side information at
the encoder. A practical DSC method named DIstributed Source Coding Using Syndromes (DISCUS) has been pro-
posed in Pradhan and Ramchandran,25 considering the related problem of compressing a source, which is spatially
correlated with another source that is available only at the decoder. DISCUS approach aims to quantize the source
using a codebook designed for the marginal probability density function of the sources. This codebook is par-
titioned for the fictitious channel between the side information and the quantized source. The encoder, instead of
sending the index of the quantized codeword, sends only the index of the corresponding coset. The decoder
recovers the quantized codeword by decoding the side information in the given coset of the channel code. In order
to enhance the signal accuracy, Cheng et al26 proposed a hierarchical DSC with an optimal transmission schedul-
ing approach exploiting the spatial dynamics of the signal to ensure efficient decoding accuracy. A Decoding
AL FALLAH ET AL. 3 of 17

Delay-based DSC is introduced in Aktas et al,27 reducing the energy expenditure of sensor nodes by maximizing
the compression rate via the propagation delay between sensor readings. All these proposed approaches take
advantage of the signal spatial correlation to promote DSC efficiency. Furthermore, several temporal compression
methods have been proposed in the literature specifically transformation-based (DCT,28 WT,29 and FFT30), linear,21
polynomial,31 and autoregressive approaches.32 Among these methods, linear algorithms have been proven their
effectiveness to compress the sensed time series with minimum computational cost.15 Notably, LTC is a Piecewise
Linear Approximation (PLA) method that has been shown a significant performance in terms of energy saving,
owing to its lightweight algorithm.22 In addition, LTC takes advantage of the incurred temporal correlation to thor-
oughly compress the collected data with low processing energy for different network topologies (multi-hop33 and
cluster-based WSNs34). Therefore, designing an adequate compression scheme that exploits the data correlation in
both space and time is an appealing solution to reduce the energy expenditure and extend the network lifetime
with high data accuracy.35 Xie et al36 have proposed an improved spatio-temporal correlation algorithm based on
compressive sensing theory. This study has proven remarkable results in terms of energy expenditure and data
accuracy since it combines the performance of cluster-based architectures with compressive sensing algorithm to
accurately recover the original data. Furthermore, an energy efficient data gathering scheme has been proposed in
Zhou et al37 exploiting the correlation in space and time to enhance the energy efficiency of clustered sensor networks.
In this proposed scheme, dual prediction is considered in the intracluster transmission to reduce the temporal
redundancies, and a hybrid compressive sensing is used in the intercluster transmission to reduce the spatial redun-
dancy. In this instance, we have been focused in this work on designing a hybrid and adaptive compression approach
exploiting the remarkable performance of DSC as a spatial compression technique and the energy effectiveness of LTC
as temporal compression scheme so as to improve the network lifetime. The use of a simple temporal compression
scheme could be an intriguing tool to enhance DSC performance especially by exploiting the temporal dynamics of the
collected samples.

3 | SYSTEM MODEL

3.1 | Signal model

In this paper work, we consider in the first simulation synthetic signals that are already proposed in Zordan et al38 and
used in Hooshmand et al14 and Rossi et al.16 These signals have been approved and verified against real-world datasets,
providing the generation of varying signal in space and time with tunable correlation parameters. In this simulation,
the Gaussian function is used as a spatial correlation function39 and is given by the following:


d2
ρs ðdÞ = exp − 2 , ð1Þ
γα

where α is a scaling factor that depends on the field size and γ is a free parameter used to control the spatial
correlation.
For the temporal correlation, we use a temporal correlation coefficient ρt ðΔtÞ = ρ 2 ½0, 1 (this amounts to assume an
exponential correlation function). Synthetic signals allow to numerically assess the performance of compression tech-
niques by regulating meticulously the correlation degree in both space and time.
The second simulation uses real-world datasets, gathered from Willmott et al,40 which contains time series of Ter-
restrial Air Temperature gridded monthly. The first 100 measurements for a time period between 1967 and 2017 have
been selected for this study. The use of real datasets appraises practically the performance of the proposed approach in
terms of energy saving and distortion of reconstructed information.
For both synthetic and real-world signals, the collected data are organized as follows. Time t = 1,2,3,... is slotted
according to a fixed time slot duration δ. A transmission round is defined as a period of T time slots duration:
round = ½1, 2, :::, t,:::, T. We consider a network of N nodes. Each sensor collects one sample S in each time slot until the
end of round. Then, the sensed data are communicated to the BS conforming to the used routing protocol. The overall
network data received by the BS in each transmission round can be modeled as a matrix of N line and T column (N ×
T) given by the following:
4 of 17 AL FALLAH ET AL.

0 1
S11 … S1T
B C
Data = @ … Sij … A, ð2Þ
SN1 … SNT

where Sij is the sensed sample of the node i at the time slot j.

3.2 | Network routing architecture

Clustering is one of the important routing methods to extend the network lifetime in WSNs.41 In this study, we consider
a cluster-based WSN architecture, where N sensor nodes are randomly deployed in an area D. These sensors are
grouped to form a number of clusters in which a Cluster-Head (CH) is elected according to Low Energy Adaptive Clus-
tering Hierarchy (LEACH) routing protocol.42 Each node collects its sensed data as a time series of T samples and trans-
mits it to the corresponding CH at the end of the transmission round. Then, CH gathers the incoming data from all the
Cluster-Members (CMs) and communicates it to the BS. LEACH is an effective clustering algorithm which adopts a
randomized rotation to select the CH nodes and evenly distribute the energy load among the sensor nodes in the
network.42

3.3 | Energy model

A simple energy model is considered in this study in order to compute the overall power consumption in the networks.
The total consumed energy in each sensor is assumed as the sum of the processing energy and transmission energy,15
given by the following equation:

E sensor = E proc + E trans , ð3Þ

where Eproc is the energy for processing to accomplish the compression task and Etrans is the energy cost needed for
transmission. The energy for decompression at the BS is not considered in this study owing to its limitless energy
resources.
The processing energy is computed by recording the number of operations such as addition, multiplication, subtrac-
tion, comparison, and division. According to the used microcontroller, these operations are translated to a number of
clock cycles corresponding to each type of calculation for the selected CPU.43 In this study, we have selected MSP430
microcontroller for each sensor node in the network, using 16 floating point package of calculation.44 The energy con-
sumed by a clock cycle for the selected CPU is E 0 = 0:726 nJ.
Table 1 shows the number of CPU cycles needed for each type of calculation.
A first-order radio energy model is considered for this study to measure the radio hardware energy dissipation in
each sensor node whether it is CH or CM.42 This model considers the energy dissipated in the transmitter and the
receiver to transmit an L-bits message over a distance d considering free space and multipath propagation model.
The free space propagation model assumes the ideal propagation condition which supposes only one clear
line-of-sight path between the transmitter and the receiver. For multipath models, the signal is transmitted via multiple
paths to reach the receiving antenna. Therefore, these paths cause interference in various ways such as dataloss and

TABLE 1 CPU cycles needed for each type of calculation

Operation Clock cycle

Addition X + Y 184
Subtraction X − Y 177
Multiplication X × Y 395
Division X/Y 405
Comparison X<=>Y 37
AL FALLAH ET AL. 5 of 17

distortion. This study takes into consideration free space and multipath propagation models in which the received
power depends generally on the distance between the transmitter and the receiver.42 If the internode distance is lower
than a crossover distance dcrossover, the free space model is used (d2 attenuation) according to Friss transmit power equa-
tion defined as follows42:

Pt Gt Gr λ2
Pr ðdÞ = : ð4Þ
ð4πdÞ2 L

Otherwise, if internode distance d is greater than dcrossover, the two-ray ground propagation model is used according
to the following equation:

Pt Gt Gr h2t h2r
Pr ðdÞ = , ð5Þ
d4 L

where Pr(d) is the transmit power as a function of internode distance d, Gt and Gr are the gains of transmitting and
receiving antennas, respectively, λ is the signal wavelength, and L≥ 1 is the system loss factor not related to propaga-
tion.42 For these two scenarios, the crossover distance is defined as follows:
pffiffiffi
4π Lht hr
dcrossover = : ð6Þ
λ

In this work, the IEEE 802.15.4 standard has been used with a frequency of 2.4 GHz. In addition, omnidirectional
antennas have been used with the following parameters: Gt = Gr = 1 dBi, hr = ht = 1 m, and lossless system (L = 1).
According to these parameters, the obtained crossover distance is equal to 100 m.
According to the first-order radio energy model as shown in Figure 1, the transmission energy can be modeled as
follows: The transmitter consumes energy to run the radio electronics (ETX − elec) and the power amplifier (ETX − elec),
while the receiver dissipates energy to run the radio electronics (ERX − elec). The energy dissipated to send L bits packet
over a distance d is given by the following:

E TX ðL, dÞ = E TX − elec ðLÞ + E TX − amp ðL, dÞ: ð7Þ

Considering the free space and multipath communications, Equation (7) is formulated as follows:
(
LðE elec + ϵf s :d2 Þ if d < dcrossover
E TX ðL, dÞ = ð8Þ
LðE elec + ϵm :d4 Þ if d≥dcrossover ,

where Eelec is the energy per bit required to transmit and receive electronics to process the information. The amplifier
parameters ϵfs and ϵm present the energy per bit required in the transmit amplifier to send L bits message with adequate

FIGURE 1 Radio energy model

6 of 17 AL FALLAH ET AL.

SNR over a distance d2 for free space propagation model and d4 for multipath propagation model, respectively. At the
receiver side (CH), energy is dissipated in running the radio electronics E(RX − elec), which is given by the following:

E RX − elec = E elec :L: ð9Þ

3.4 | Performance metrics

3.4.1 | Compression ratio

The Compression Ratio (CR) is an essential evaluation parameter in data compression. It practically describes the
impact of the compression method on reducing the volume of the sensed data. It is defined as the ratio of compressed
to the original data bit-length given by the following:

V olume of compressed data

CR = : ð10Þ
volume of raw data

Since it indicates the reduction of communication energy expenditure, CR is a said to be the key factor of selecting
well-performed compression algorithms in view of lowering energy consumption in data transmission.

3.4.2 | Distortion

In the case of lossy compression methods, the recovered information at the receiver side is only an approximation of
the original data. Therefore, the distortion metric represents the information loss occurred in the reconstructed data
compared to the initial signal given by the following:

D = j^
x ðiÞ −xðiÞj, ð11Þ

where x(i) is an element of a given time series and x^ðiÞ represents its compression version. The accuracy requirement of
the data compression method depends on WSNs applications. Hence, more the signal is correlated (in space or time),
more the data reliability is guaranteed.
In the case of lossy compression, there is a trade-off between the achieved CR and the distortion of reconstructed
signal. Effectively, reducing the amount of transmitted data is advantageous in terms of energy saving. However, it may
affect the reliability of the received information. Therefore, the aim of each lossy compression approach is to find a
compromise between these two parameters. Efficient algorithms ensure high compression capabilities without
impacting significantly the signal representation accuracy.

4 | S PA T I O - T E M P O R A L CO MPRESSION TECHNIQUES

4.1 | Spatial compression technique: DSC

DSC compression approach is based on both encoding the sensed data from different sensors and decoding it jointly by
exploiting the occurring spatial correlation. This idea is useful for clustering topologies when CMs compress the col-
lected information and transmit it to the corresponding CH, which will be the side information at the BS. Indeed, CH
sends its uncompressed signal in addition of the compressed data from CMs, so that the entire information can be
retrieved at the sink which contains already the correlation function of the entire dataset.
The DSC procedure starts with a quantization phase, where each node uniformly quantizes its sensed reading xt
according to a fixed quantization step Δ, in order to obtain certain number of information levels Nl. The uniform quan-
tization is said to be useful as it can be implemented readily and simply in the hardware. Then bins are formed by
grouping the quantized symbols x qt conforming to Ungerboeck tree-based binning scheme proposed in Ungerboeck45
AL FALLAH ET AL. 7 of 17

and used in Hooshmand et al14 as shown in the example of Figure 2. This procedure has proven its effectiveness to
transmit information with high efficiency and band-limited channels, as it is based on the parity check for each symbol
instead of the older technique of applying it to the bit stream.45
We consider n the number of bits required to represent the quantized symbol. In Figure 2, we use four bins (A1, A2,
A3, A4), that can be represented by k = 2 bits. The quantized symbols are divided according to two important conditions:
The symbols in the same bin must have maximum distance assuring an even distribution of the symbols in the bins.
Likewise, this distance should be maintained for all bins. Afterwards, each CM transmits its obtained bins to the
corresponding CH, which will be side information at the BS to reconstruct the compressed data. Therefore, CHs simply
quantize its sensed information yqt in addition of the received bins from CMs to be transmitted to the sink. The original
information is recovered by firstly reconstructing CH signal from its quantized symbols yqt . Then, the bins are mapped
to the corresponding symbols that have minimum distance with respect to the side information received from
CH. Hence, the retrieved symbol is given by the following:

x^qt = argminxqt 2A ðjx qt −yqt jÞ: ð12Þ

The influence of the spatial correlation is significant in this reconstruction procedure, since the estimated quantized
signal x^qt will be equal to the original x qt when x qt −yqt < 2k − 2 Δ. Therefore, more CH and CMs signals are close to each
other, more the data reliability is ensured. The use of DSC as a spatial compression scheme provides efficient perfor-
mance in terms of energy saving and data accuracy. In fact, DSC reduces significantly the amount of transmitted data,
maintaining a fixed and small compression ratio cr = k/n. In addition, the cluster-based routing architecture guarantees
high spatial correlation in each cluster, which makes the reconstructed signal significantly accurate.
In order to assess the performance of DSC algorithm in terms of compression ratio as a function of the error bound,
we calculate the network compression ratio which is given by the following:

ðN −N CH Þ∗k + N CH ∗L
CRDSC − net = , ð13Þ
N:L

where NCH is the number of CH in the network and k is the bin identifier (binID) bit-length. From Equation (13), it
appears that CR is directly affected by the number of CH in the network. As a result, more the number of CH in the
network is increased, more the received data are accurate. In our performed simulation, the number of CH is tuned to
provide different clustering topologies in order to assess DSC performance in terms of data reliability.

4.2 | Temporal compression technique: LTC

LTC is a simple linear approximation technique, which has proven its performance in terms of resources saving.21,22 It
is designed to approximate a given time series x(n) with a sequence of line segments through a low approximation
tolerance ε.

F I G U R E 2 Binning-based
quantization procedure
8 of 17 AL FALLAH ET AL.

As shown in Figure 3, the algorithm operates by approximating multiple successive readings by a single line, in a
way that this segment should meet the error tolerance ε for all points. Once this occurs, only the first and last reading
in the segment are transmitted instead of all data points.
Previous works have been proven that LTC necessitates a small computational cost due its lightweight
algorithm, which requires a few number of operations during the processing.15,21 In addition, LTC becomes more
energy efficient when the data values change slowly over time, which makes its performance highly related to the tem-
poral correlation.

4.3 | Proposed approach: DTSC

The main goal of the proposed scheme DTSC is to prolong energy lifetime as well as to improve data reliability of real-
word datasets. This is achieved by exploiting the efficiency of spatial (DSC) and temporal (LTC) compression schemes
for various correlation parameters. We consider the network matrix data previously mentioned in Equation (2), where
each sensor node collects a time series of T samples in the transmission round. The proposed algorithm operates
according to the following procedure. At the CM level, we consider a time series of T time slots [x(1), …, x(t), …, x(T)]
that represents a line vector of the signal matrix data. In a cluster-based architecture, each sensor node compresses its
first t data samples [x(1), …, x(t)] via LTC algorithm in order to check if the temporal correlation occurs. If the compres-
sion ratio is lower than 1 (temporal correlation is important), LTC is applied for the entire time series. The compressed
data are quantized to form bins before to be sent to CH according to DSC algorithm. If the compression ratio is equal to
1 (no temporal correlation), quantization and binning are applied to the original data without performing LTC proce-
dure as shown in Algorithm 1.

FIGURE 3 Example of LTC compression process

AL FALLAH ET AL. 9 of 17

Equation (3) can be modeled as follows:

E CM = ELTC + E q + E b + Etrans , ð14Þ

where ELTC is the energy needed to compress data via LTC, Eq is the energy for quantization, and Eb is the energy for
binning. As previously mentioned, the temporal correlation is verified in a training phase, by applying LTC to the first
t samples. Thus, the energy required to compress data via LTC is given by the following:


E 0 :N cc ðxð1Þ, …, xðtÞÞ if CR < 1
E LTC = ð15Þ
E0 :N cc ðxð1Þ, …,xðTÞÞ if CR = 1,

where E0 is the energy consumed per a clock cycle according to the used microcontroller, Ncc is the number of clock
cycles needed for calculation to process LTC algorithm, and CR is the compression ratio. The energy consumed to pro-
cess the quantization task is given by the following:

E q = E q0 :T:CR, ð16Þ

where Eq0 is the energy needed to quantize one data sample and T is the number of samples in the time series. The
energy for binning can be modeled by the following equation:

E b = E b0 :T:CR, ð17Þ

where Eb0 is the energy needed for binning of one data sample. The energy for transmission is expressed as follows:

E trans = E TX :k:T:CR, ð18Þ

where ETX is the energy for transmission of one bit and k is the number of bits to present the binID. If the temporal cor-
relation occurs, the total energy depleted in the CM is given by the following:

E CM = ðE q0 + Eb0 + E TX :kÞT:CR + E 0 :N cc ðxð1Þ, …,xðTÞÞ: ð19Þ

• At the CH level

At the end of the transmission round, CH collects the bins from the CMs, when its original data are only quantized
in order to be side information at the BS (Algorithm 2).

Energy is consumed at CH level in the reception of CMs data, in the quantization of CH information, and the trans-
mission of the entire data to the sink. The energy dissipated in each CH is given by the following:
Xm   Xm  
E CH = E RX :K:T i=1
CRðiÞ + E q 0
:T + E TX K:T i=1
CRðiÞ +l , ð20Þ

where ERX is the reception energy of 1 bit, m is the number of CMs, and l is the packet size of the quantized data in CH.

• At the BS level

After receiving all the network packets from CHs, the sink firstly decodes the collected bins from the CMs and the
quantized information from CH using DSC decoding. According to the dimension of the obtained time series
10 of 17 AL FALLAH ET AL.

(dimension < T), LTC decompression is applied in order to retrieve the entire network data matrix (Algorithm 3).

In order to assess DTSC performance in terms of distortion as a function of CR, we calculate the network CR of the
adopted scheme as follows:

CRDTSC − net = CRDSC − net :CRLTC − net , ð21Þ

where CRLTC − net is the mean of all sensors CR through LTC algorithm. From Equation (13), CRDTSC − net can be
reformulated as follows:

ðN −N CH Þ∗k + N CH ∗L
CRDTSC − net = :CRLTC − net : ð22Þ
N:L

The proposed approach DTSC is considered as a hybrid approach owing to the combination between temporal and
spatial compression algorithms (LTC and DSC). In addition, this approach performs adaptively according to the tempo-
ral correlation occurance, since LTC decides if it will compress its entire time series conforming to the result of the first
set of samples (x(0) to x(t)). This step is essential to avoid depleting energy in compressing data when the temporal cor-
relation is very low. Previous studies have proven the efficiency of DSC algorithm in terms of energy expenditure and
data accuracy for high levels of spatial correlation.14,16 The objective of the proposed scheme is to enhance its perfor-
mance by exploiting the advantage of temporal correlation to prolong the network lifetime with minimum distortion.

5 | S I MULATIONS A ND R ESULTS

In this section, we compare the performances of the proposed spatio-temporal compression scheme DTSC to these of
temporal and spatial schemes LTC and DSC, respectively. To this aim, we consider a network of N = 50 nodes deployed
randomly in the area D (100 × 100 m). BS is placed at the center of WSN area. We assume that each sensor node collects
its data samples according to the signal model in Section 3. Each reading is represented by n = 16 bits, while the binID
in DSC algorithm takes k = 4 bits of memory. For the routing topology, we use a cluster-based architecture, where the
number of clusters Nc is varied between 5 and N in steps of 5. For the selected compression schemes, the total network
energy is assessed as a function of the network lifespan represented by the transmission rounds. The signal accuracy is
evaluated in terms of the network CR for each compression scheme. Table 2 shows the simulation parameters used in
this study.
Figures 4–7 show the performance of the selected compression schemes in terms of energy consumption as a func-
tion of transmission rounds for different correlation parameters. In Figure 4, the signal is uncorrelated in space and
time. The energy performance of the proposed scheme DTSC and DSC are almost identical with a slight difference due
to the energy consumed by DTSC to test the temporal correlation occurrence in sensor nodes. Since the collected signal
changes significantly over time, LTC algorithm should be discarded as its energy dissipation in this case is higher than
that incurred in sending all the data uncompressed. For temporally correlated signal (Figure 5), DTSC algorithm
reaches 739 rounds and outperforms DSC (557 rounds) and LTC (604 rounds) in terms of energy saving. Owing to its
adaptability to compress data in two levels, DTSC firstly compresses the collected readings through LTC, then the
AL FALLAH ET AL. 11 of 17

TABLE 2 Simulation parameters

Parameter Value
ϵfs 2.59 × 10−10 J/bit/m2
ϵmp 2.52 × 10−14 J/bit/m4
Initial node energy 500 mJ
Eelec 50 nJ/bit
Number of nodes 50
Frequency 2.4 GHz
Bit rate 250 kbps
Antennas gain 1 dB
Time slot in the round 100
dcrossover 100 m
Reading bit length 16 bits
binID 4 bits

F I G U R E 4 Total system energy versus transmission rounds for

uncorrelated signal in space and time (ρs =0.001, ρt = 0)

F I G U R E 5 Total system energy versus transmission rounds for

temporally correlated and spatially uncorrelated signal (ρs = 0:1,ρt = 0:8)

F I G U R E 6 Total system energy versus transmission rounds for

spatially correlated and temporally uncorrelated signal (ρs = 5,ρt = 0)
12 of 17 AL FALLAH ET AL.

obtained data are quantized to form bins. The energy for transmission is essentially reduced owing to the small number
of bins to be transmitted to CH. In Figure 6, the LTC provides unfavorable performance (same as in Figure 4) due to its
energy depletion for processing without reducing the amount of collected information. DSC performance is unaffected
by the incurred spatial correlation, since its operating algorithm transforms each sensor reading into a binID with a
fixed bit length independently of the correlation parameters. As mentioned in the previous section, DTSC compresses
the first t samples of the sensed time series through LTC algorithm to verify the presence of the temporal correlation,
which is not the case in this figure. Hence, LTC procedure is abandoned, and the entire time series remains
uncompressed. Afterwards, DSC procedure is applied to obtain the binIDs of the collected datasets. In this case, DTSC
consumes a part of its energy in testing the temporal correlation through LTC algorithm and then performs as DSC
algorithm. For this reason, DTSC is also unaffected by the spatial correlation in terms of energy saving. For highly cor-
related signal (Figure 7), the proposed algorithm DTSC outperforms other compression schemes owing to its efficiency
to exploit the temporal correlation through LTC algorithm on the first hand and to reduce the amount of transmitted
data via DSC binning procedure on the other hand.

F I G U R E 7 Total system energy versus transmission rounds for

highly correlated signal in space and time (ρs = 5, ρt = 0:5)

F I G U R E 8 Error bound versus compression ratio for uncorrelated

signal in space and time (ρs = 0:001,ρt = 0)

F I G U R E 9 Error bound versus compression ratio for temporally

correlated and spatially uncorrelated signal (ρs = 0:1, ρt = 0:8)
AL FALLAH ET AL. 13 of 17

Figures 8–11 show the data accuracy of selected compression schemes varying the compression ratio as a function
of the error bound for different correlation parameters. For uncorrelated signal (Figure 8), DTSC distortion is slightly
higher than DSC for CR > 0.57. Otherwise, DTSC becomes more stable in terms of distortion and reaches efficient CR
even though LTC and DSC errors are important for CR ≤ 0.57. For temporally correlated signal (Figure 9), DTSC keeps
the same behavior as Figure 11 even though the distortion of LTC is low. For spatially correlated signal (Figure 10),
DTSC performs between LTC and DSC and is still reaching efficient ratio of CR (0,1). Same performance is obtained
when the signal is correlated in both space and time (Figure 11) with a slight difference due to the use of LTC as the
temporal correlation is important.
Since DTSC uses LTC as a first compression step before applying DSC algorithm, it was expected that DTSC distor-
tion would be the sum of LTC and DSC distortions, which is not the case for all these figures. Effectively, these surpris-
ing results have proven that DTSC error has been compensated exploiting the obtained distortion of LTC and DSC.
Indeed, in each sensor node, the compressed time series through LTC algorithm involves some information loss ε,
which will be introduced to the quantization and binning procedure via DSC scheme. Thus, DSC decoder retrieves the

F I G U R E 1 0 Error bound versus compression ratio for spatially

correlated and temporally uncorrelated signal (ρs = 5,ρt = 0)

F I G U R E 1 1 Error bound versus compression ratio for highly

correlated signal in space and time (ρs = 5, ρt = 0:5)

F I G U R E 1 2 DTSC Total System Energy versus transmission

rounds for various correlation parameters
14 of 17 AL FALLAH ET AL.

quantized data samples from the received bins according to Equation (12) with respect to the side information data. As
a result, the LTC error ε minimizes the gap between the compressed data samples via LTC and side information.
Figures 12 and 13 show DTSC performance, respectively, in terms of energy consumption and error bound.
From Figure 12, it appears that DTSC is highly affected by the temporal correlation in terms of energy saving. As
LTC, DTSC network lifespan is prolonged only when the temporal correlation is important, owing to the use of LTC as
a first step to discard information redundancies in sensor nodes time series. However, DTSC keeps the same behavior
of DSC when temporal correlation is low, even if the spatial correlation is increased. This expected result emanates
from the fact that DSC compression procedure is maintained with the same CR and energy for processing regardless of
correlation parameters values. DTSC exploits the energy performance of DSC for all correlation parameters and the effi-
ciency of LTC for high temporal correlation.
When we compare DTSC performance in terms of data accuracy (Figure 13), DTSC performance is significantly
influenced by the spatial correlation in terms of information reliability. As DSC, DTSC ensures efficient data accuracy
only when the spatial correlation occurs owing to the performed binning procedure that guarantees effective represen-
tation of collected data with minimum bit-length. However, the distortion increases with the increase of temporal

F I G U R E 1 3 DTSC Error bound versus compression ratio for

various correlation parameters

F I G U R E 1 4 Total System Energy versus transmission rounds for

real-world datasets

FIGURE 15 Error bound versus compression ratio for real-world

datasets
AL FALLAH ET AL. 15 of 17

correlation, while the use of LTC is important providing additional loss of information, which explains the occurred
data distortion as long as the temporal correlation is significant.
Figures 14 and 15 show, respectively, the energy performance and data reliability of selected compression schemes
using real-world datasets. From these results, the proposed approach DTSC outperforms other compression in terms of
extending the network lifetime, taking advantage of LTC and DSC in compressing the collected datasets that are highly
correlated in space and time. Moreover, DTSC has proven its efficiency in terms of data accuracy and has succeeded to
reduce the trade-off between energy saving and distortion.

6 | C ON C L U S I ON

In this paper, we have proposed a novel hybrid and adaptive spatio-temporal compression scheme (DTSC) that com-
bines the efficiency of temporal (LTC) and spatial (DSC) approaches. Extensive simulations assess the performance of
the proposed technique in a cluster-based WSN using synthetic signals with tunable correlation parameters and real-
world datasets. The obtained results have revealed that DTSC outperforms other compression schemes and promotes
the network lifetime owing to its adaptivity to discard redundant information in two steps: DTSC reduces the collected
time series in each sensor through LTC algorithm especially by exploiting the occurring temporal correlation. Then, the
obtained data samples are quantized and transformed via DSC binning procedure. Furthermore, DTSC ensures efficient
data accuracy through its capability to compensate the incurred error at each compression step. The proposed approach
has succeeded to reduce the gap between energy efficiency and data reliability in WSNs.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.

ORCID
Samia Al Fallah https://orcid.org/0000-0002-1561-3833

R EF E RE N C E S
1. Yong-Min L, Shu-Ci W, Xiao-Hong N. The architecture and characteristics of wireless sensor network. In: 2009 International Conference
on Computer Technology and Development, Vol. 1; 2009:561–565. https://doi.org/10.1109/ICCTD.2009.44
2. Aldeer MMN. A summary survey on recent applications of wireless sensor networks. In: 2013 IEEE Student Conference on Research
and Developement. IEEE; 2013; Putrajaya, Malaysia:485–490. https://doi.org/10.1109/SCOReD.2013.7002637
3. Wang J, Yang Y, Wang T, Sherratt RS, Zhang J. Big data service architecture: A survey. J Internet Technol. 2020;21(2):393–405.
4. Yadav S, Yadav RS. A review on energy efficient protocols in wireless sensor networks. Wireless Netw. 2016;22(1):335–350.
5. Curry RM, Smith JC. A survey of optimization algorithms for wireless sensor network lifetime maximization. Comput Ind Eng. 2016;
101:145–166.
6. Damaso A, Freitas D, Rosa N, Silva B, Maciel P. Evaluating the power consumption of wireless sensor network applications using
models. Sens. 2013;13(3):3473–3500.
7. Mostafavi S, Hakami V. A new rank-order clustering algorithm for prolonging the lifetime of wireless sensor networks. Int J Commun
Syst. 2020;33(7):e4313. https://doi.org/10.1002/dac.4313
8. Shaikh FK, Zeadally S. Energy harvesting in wireless sensor networks: a comprehensive review. Renew Sustain Energy Rev. 2016;55:
1041–1054.
9. ELkamel R, Cherif A. Energy-efficient routing protocol to improve energy consumption in wireless sensors networks. Int J Commun
Syst. 2017;30(17):e3360. https://doi.org/10.1002/dac.3360
10. Wang J, Gao Y, Yin X, Li F, Kim H-J, Lloret J. An enhanced PEGASIS algorithm with mobile sink support for wireless sensor networks.
Wirel Commun Mob Comput. 2018;2018:1–9.
11. Wang J, Gao Y, Zhou C, Simon Sherratt R, Wang L. Optimal coverage multi-path scheduling scheme with multiple mobile sinks for
WSNs. Comput Mater Contin. 2020;62(2):695–711.
12. Wang J, Gao Y, Liu W, Sangaiah AK, Kim H-J. An intelligent data gathering schema with data fusion supported for mobile sink in wire-
less sensor networks. Int J of Distributed Sens Netw. 2019;15(3):1550147719839581.
13. Razzaque MA, Bleakley C, Dobson S. Compression in wireless sensor networks: a survey and comparative evaluation. ACM Trans Sen
Netw. 2013;10(1):1–44.
14. Hooshmand M, Rossi M, Zordan D, Zorzi M. Covariogram-based compressive sensing for environmental wireless sensor networks. IEEE
Sensors J. 2016;16(6):1716–1729.
15. Zordan D, Martinez B, Vilajosana I, Rossi M. On the performance of lossy compression schemes for energy constrained sensor network-
ing. ACM Trans Sens Netw. 2014;11(1):1–34.
16 of 17 AL FALLAH ET AL.

16. Rossi M, Hooshmand M, Zordan D, Zorzi M. Evaluating the gap between compressive sensing and distributed source coding in WSN.
In: 2015 International Conference on Computing, Networking and Communications (ICNC). IEEE; 2015; Garden Grove, CA, USA:
911–917. https://doi.org/10.1109/ICCNC.2015.7069468
17. Parker D, Stojanovic M, Yu C. Exploiting temporal and spatial correlation in wireless sensor networks. In: 2013 Asilomar Conference on
Signals, Systems and Computers. IEEE; 2013; Pacific Grove, CA, USA:442–446. https://doi.org/10.1109/ACSSC.2013.6810315
18. Xiong Z, Liveris AD, Cheng S. Compression via channel coding—distributed source coding for sensor networks. IEEE Sig Process Mag.
2004;21(5):80–94. https://doi.org/10.1109/msp.2004.1328091
19. Hua G, Chen CW. Distributed source coding in wireless sensor networks. In: Second International Conference on Quality of Service in
Heterogeneous Wired/Wireless Networks (QSHINE'05). IEEE; 2005; Orlando, FL, USA:6–6. https://doi.org/10.1109/QSHINE.2005.19
20. Aljohani AJ, Ng SX, Hanzo L. Distributed source coding and its applications in Relaying-Based transmission. IEEE Access. 2016;4:
1940–1970.
21. Schoellhammer T, Greenstein B, Osterweil E, Wimbrow M, Estrin D. Lightweight temporal compression of microclimate datasets [Wire-
less Sensor Networks]. In: 29Th Annual IEEE International Conference on Local Computer Networks; 2004; Tampa, FL, USA, USA:
516–524. https://doi.org/10.1109/LCN.2004.72
22. Fallah SA, Arioua M, El Oualkadi A, El Asri J. On the performance of piecewise linear approximation techniques in WSNs. In: 2018
International Conference on Advanced Communication Technologies and Networking (Commnet). IEEE; 2018; Marrakech:1–6. https://
doi.org/10.1109/COMMNET.2018.8360262
23. Slepian D, Wolf J. Noiseless coding of correlated information sources. IEEE Trans Inform Theory. 1973;19(4):471–480.
24. Wyner A, Ziv J. The rate-distortion function for source coding with side information at the decoder. IEEE Trans Inform Theory. 1976;22
(1):1–10.
25. Pradhan SS, Ramchandran K. Distributed source coding using syndromes (DISCUS): design and construction. IEEE Trans Inform The-
ory. 2003;49(3):626–643.
26. Cheng S-T, Shih J-S, Chou C-L, Horng G-J, Wang C-H. Hierarchical distributed source coding scheme and optimal transmission sched-
uling for wireless sensor networks. Wireless Pers Commun. 2013;70(2):847–868.
27. Aktas M, Kuscu M, Dinc E, Akan OB. D-DSC: Decoding delay-based distributed source coding for internet of sensing things. PLOS
ONE. 2018;13(3):e0193154.
28. Ahmed N, Natarajan T, Rao KR. Discrete cosine transform. IEEE Trans Comput. 1974;C-23(1):90–93. https://doi.org/10.1109/T-C.1974.
223784
29. Shen G, Ortega A. Transform-based distributed data gathering. IEEE Trans Signal Process. 2010;58(7):3802–3815. https://doi.org/10.
1109/TSP.2010.2047640
30. Narang B, Kaur EA, Singh DD. Comparison of DWT and DFT for energy efficiency in underwater wireless sensor networks; 2015.
31. Ozdemir S, Xiao Yang. Polynomial regression based secure data aggregation for wireless sensor networks. In: 2011 IEEE Global Tele-
communications Conference - GLOBECOM, Vol. 2011. IEEE; 2011; Houston, TX, USA:1–5. https://doi.org/10.1109/GLOCOM.2011.
6133924
32. Lu J, Valois F, Dohler M, Wu M-Y. Optimized data aggregation in WSNs Using Adaptive ARMA. In: 2010 Fourth International Confer-
ence on Sensor Technologies and Applications. IEEE; 2010; Venice, Italy:115–120. https://doi.org/10.1109/SENSORCOMM.2010.25
33. Al Fallah S, Arioua M, El Oualkadi A, El Asri J. PLA compression schemes assessment in multi-hop wireless sensor networks. Proc
Comput Sci. 2018;130:279–286. https://doi.org/10.1016/j.procs.2018.04.040
34. Aasri JE, Fallah SA, Arioua M, Oualkadi AE, Zekriti A. Vertical and horizontal compression scheme assessment in cluster-based WSNs.
In: 2018 IEEE 5th International Congress on Information Science and Technology (CiSt); 2018:650–655. https://doi.org/10.1109/CIST.
2018.8596476
35. He B, Li Y, Huang H, Tang H. Spatial–temporal compression and recovery in a wireless sensor network in an underground tunnel envi-
ronment. Knowl Inf Syst. 2014;41(2):449–465.
36. Xie X, Wang J, Hu F, Jiang N, Ge S. An improved spatial-temporal correlation algorithm of wsns based on compressed sensing. In: 2017
IEEE International Conference on Computational Science and Engineering (CSE) and IEEE International Conference on Embedded
and Ubiquitous Computing (EUC). IEEE; 2017; Guangzhou, China:159–164. https://doi.org/10.1109/CSE-EUC.2017.36
37. Zhou Y, Yang L, Yang L, Ni M. Novel energy-efficient data gathering scheme exploiting spatial-temporal correlation for wireless sensor
networks. Wireless Comm Mobile Comput. 2019;7:1–10.
38. Zordan D, Quer G, Zorzi M, Rossi M. Modeling and generation of space-time correlated signals for sensor network fields. In: 2011 IEEE
Global Telecommunications Conference - GLOBECOM, Vol. 2011. IEEE; 2011; Houston, TX, USA:1–6. https://doi.org/10.1109/
GLOCOM.2011.6133891
39. Abrahamsen P. A review of Gaussian random fields and correlation functions. Oslo: Norsk Regnesentral/Norwegian Computing Center;
1997.
40. Willmott CJ, Matsuura K, Legates DR. Global air temperature and precipitation: Regridded monthly and annual climatologies. Tech.
Rep., Newark,USA, Dept. Geography, Univ. Delaware; 2012.
41. Wang J, Gao Y, Liu W, Wu W, Lim SJ. An asynchronous clustering and mobile data gathering schema based on timer mechanism in
wireless sensor networks. Comput Mater Contin. 2019;58(3):711–725.
42. Heinzelman WB, Chandrakasan AP, Balakrishnan H. An application-specific protocol architecture for wireless microsensor networks.
IEEE Trans Wireless Commun. 2002;1(4):660–670. https://doi.org/10.1109/TWC.2002.804190
AL FALLAH ET AL. 17 of 17

43. Zordan D, Martinez B, Vilajosana I, Rossi M. To Compress or Not To Compress: Processing vs Transmission Tradeoffs for Energy Con-
strained Sensor Networking. arXiv:12062129 [cs]. http://arxiv.org/abs/1206.2129; 2012.
44. Bierl L. MSP430 Family Mixed-Signal Microcontroller Application Reports. 22.
45. Ungerboeck G. Channel coding with multilevel/phase signals. IEEE Trans Inform Theory. 1982;28(1):55–67.

How to cite this article: Al Fallah S, El Aasri J, Arioua M, El Oualkadi A. A hybrid adaptive spatiotemporal
compression scheme in cluster-based wireless sensor networks. Int J Commun Syst. 2020;33:e4623. https://doi.
org/10.1002/dac.4623