Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Computers and Electronics in Agriculture 127 (2016) 553–560

Contents lists available at ScienceDirect

Computers and Electronics in Agriculture


journal homepage: www.elsevier.com/locate/compag

Original papers

New empirical path loss model for wireless sensor networks in mango
greenhouses
Auda Raheemah a,b, Naseer Sabri a, M.S. Salim c,⇑, Phaklen Ehkan a, R. Badlishah Ahmad a
a
Computer and Communication Eng. School, University Malaysia Perlis, Malaysia
b
Al-Muthanna University, Iraq
c
Laser and Optical Electronics Eng. School, Alnahrain University, Iraq

a r t i c l e i n f o a b s t r a c t

Article history: Signal propagation losses in protected environments are investigated using wireless sensor networks
Received 29 October 2015 (WSNs) based on the IEEE 802.15.4 standard of operating frequency, 2.425 GHz. In this research, various
Received in revised form 12 July 2016 empirical measurements were conducted to examine the effects of each part of a tree on path loss using
Accepted 13 July 2016
different transceiver heights. A new linear path loss regression curve-fitting model (LRCFM) was derived
Available online 25 July 2016
based on the regression technique of computing the total path loss inside the greenhouse environment.
The greatest vegetation effects appear within 1.5 m tree height; in this research, this height was adopted
Keywords:
to study and analyse vegetation models in a mango greenhouse. This research proves that path loss pre-
Wireless sensor network
IEEE802.15.4
diction based on free space path loss (FSPL) and two-ray (2-Ray) propagation models is inaccurate in pre-
Path loss modelling dicting loss in certain environments, as these approaches are simplistic and optimistic. Thus, most known
Mango greenhouse foliage models used in conjugation with FSPL and 2-Ray are inaccurate in predicting the total path loss in
a greenhouse environment. The analytical and empirical results prove that the new derived model, the
LRCFM, is the best candidate compared to other foliage models. The mean absolute percentage error
(MAPE) of the total path loss based on the new LRCFM model was 2.7% compared to the 10.69% of the
well-known models.
Ó 2016 Elsevier B.V. All rights reserved.

1. Introduction ZigBee was developed as an open, global standard to achieve


cost-effectiveness and less power consumption in air network
With the invention of small, cost-effective transceiver modules communication. ZigBee operates on IEEE 802.15.4 physical radio
and controllers, wireless sensor networks (WSNs) gained interest specifications and operates in unlicensed bands, including
from researchers and hardware and software developers for a vari- 2.4 GHz, 900 MHz and 868 MHz (Morais et al., 2008).
ety of applications. Currently, the WSN application field includes Newly emerging WSN technology has extended into a variety of
home monitoring and controlling, health monitoring, factories multidisciplinary applications. Farming and agriculture is one of
and industries, environmental monitoring and control and preci- the industries that has lately redirected its concentration to wire-
sion agriculture, military and space applications (Lewis, 2004; less network sensors, looking to use this low-price technology to
Sabri et al., 2012). WSN developers adopt different communication enhance the yield standard and thus production (Morais et al.,
technologies based on the IEEE 802.15.4 standard, which allows for 2008). Several test points equipped with wireless nodes of various
a communication range of a few hundred metres and offers WSN sensor types are used to collect the local climate and set fertilisa-
applications easy-going latency and throughput necessities. The tion and irrigation parametres within diverse divisions of the
IEEE 802.15.4 standard is less complex and more cost-effective greenhouse to provide information for a successful greenhouse
standard for WSNs, with low power consumption and a data rate computerisation system. However, cabling sensor nodes affect
of 250 Kbps. These features are suitable for adequate yet cheap the measurement system in terms of its expenditure and suscepti-
mobile devices or spatially embedded devices. One of the most bility. Furthermore, cable test points are inflexible in repositioning
promising WSN technology protocols is the ZigBee standard. once they have been mounted. Thus, WSNs are an attractive and
cost-effective alternative for building a more effective measure-
ment and control systems.
⇑ Corresponding author.
Temperature, humidity, light and carbon dioxide are the most
E-mail addresses: naseersabri@yahoo.com (N. Sabri), muhsabri1967@yahoo.com
(M.S. Salim).
significant features in determining the productivity and quality

http://dx.doi.org/10.1016/j.compag.2016.07.011
0168-1699/Ó 2016 Elsevier B.V. All rights reserved.
554 A. Raheemah et al. / Computers and Electronics in Agriculture 127 (2016) 553–560

of plants. Hence, careful observation of these ecological parametres two transceivers of wireless sensor nodes since it does not take
can offer valuable information regarding their effects on plant reflections, obstructions or other effects between transmitter and
growth and hence the best strategy for optimum harvest produc- the receiver into account. Meanwhile, the 2-Ray model considers
tivity (Sabri et al., 2012). Finding the optimal greenhouse climate two waves from the sender to listener units: a straight wave and
can enable adjustments for the improvement of productivity and ground-reflected wave. The 2-Ray model assumes that the separa-
the achievement of remarkable energy savings (Savage et al., tion between the transmitter antenna and the receiver antenna is
2003). However, real-time signal propagation is highly affected much larger than the antennas’ heights (Rappaport, 2002). In addi-
by shadowing and multipath and attenuation factors; therefore, tion, it often represents the ground as flat, which is a perfect con-
wave propagation analysis is complex. In addition, the agricultural ducting surface. However, different grounds have different
application of WSNs is faced with challenges such as node posi- properties that govern the reflection of an incident wave
tioning for mesh networks over wide areas and achieving consis- (Rappaport, 2002). Therefore, such a simplistic approach results
tent communication quality over yield canopies. in predictions that are not precise in almost all real-world
WSN technology can be used in various environments, such as scenarios.
orchards, bare fields and vineyards, from simple to complex topog- Many empirical path loss models have been proposed for differ-
raphy and in different weather circumstances, all of which affect ent outdoor environment deployment scenarios of WSNs. For char-
radio link quality and performance (Sabri et al., 2012; Andrade- acterising near-ground RF propagation, Meng et al. (2009)
Sanchez et al., 2007; Scott et al., 2006; Zhou et al., 2004). In various performed path loss modelling in tropical plantations over a large
surroundings and circumstances, node gaps, topography, yield distance in the forest. In Balachander et al. (2013), the researchers
growth, transceiver elevation, and other common factors all have presented measurements of path loss in agricultural fields and gar-
an effect on the communication link quality and power budget. dens to specify the variables of the empirical model. Further, the
To determine the behaviour of the propagated waves, an accurate authors in Correia et al. (2013) presented an empirical model as
model of path loss and propagation must be adopted. However, the a reference to deploy WSNs in vineyard environments. In
propagation models normally used in wireless communication AlSayyari et al. (2014a, 2014b, 2014c, 2014d), the authors pre-
might not precisely describe WSNs (Mestre et al., 2010). WSN sented four empirical path loss models based on the experimental
nodes are spatially located, as they are usually located near the RF measurements of WSNs deployed in dense trees, concrete sur-
earth’s surface; thus, the main rays between sender–receiver nodes faces, artificial turfs and sand terrain environments, respectively.
are often absent, although WSN nodes have a short-distance distri- In addition, the authors in Otero et al. (2014) presented RF mea-
bution. Therefore, WSN propagation waves may face obstacles like surements and an empirical path loss model for WSN deployment
fences, trees, dense foliage and buildings (Sabri et al., 2012, 2013). in sparse tree and long grass environments.
Currently, there are two dominant channel modelling The present research studies the propagated signals affected by
approaches: theoretical and empirical (Rappaport, 2002). While imposed foliage loss in order to derive a new model for the predic-
theoretical path loss models depend on the knowledge of the phys- tion of path loss inside greenhouse environments based on the
ical laws of the wireless channel, such as the electrical properties of regression technique of realistic measurements. Propagation loss
the ground, empirical path loss models are based on actual radio is modelled by a combination of foliage effects, the effects caused
frequency (RF) measurements of wireless channels. The advan- by reflections (from the ground or the tree canopy), diffraction
tages of empirical path loss models over theoretical path loss mod- and scattering of the travelling wave.
els include their ease of implementation and their ability to include
all environment-related factors that affect the propagation of radio
waves in practice (Rappaport, 2002). In WSNs, accurate RF models 2. Materials and methods
are expected to help in achieving proper evaluation and optimizing
network performance during the deployment planning process 2.1. Large-scale propagation models
(Otero et al., 2010), to improve the power efficiency of the sensor
nodes (Sawant et al., 2007) and to make the localization and target Radio signals passing through any environment are affected by
detection applications that depend on the knowledge of received diffraction, reflection and scattering, which all results in an atten-
signal levels more reliable (Akyildiz and Vuran, 2010). uation in the RSS. The attenuation of the propagated signal
In WSNs, two simplistic path loss models are categorized as between the transceiver antenna can be used to compute the RSS
large-scale propagation path loss models: free space path loss values (Rappaport, 2002). Models used to predict the RSS between
(FSPL) and two-ray (2-Ray) path loss models (Sabri et al., 2013; two transceivers are called large-scale propagation models since
Otero et al., 2009), which are used in most of the published work. the spatial distance of the transmitted signal is large.
Both of these models are based on simplistic approaches and are
considered very optimistic in near-ground propagation scenarios, 2.1.1. FSPL model
as in various outdoor applications of WSNs. In such environments, Path loss (PL) in radio wave propagation can be computed using
signal propagation depends heavily on the type of terrain and the the FSPL model, as shown in Eq. (1), which is considered a lower
objects that exist in that terrain. Many works have reported differ- bound for path loss estimation:
ences in received signal strength (RSS) values in various deploy- PLFSPL ¼ 27:56 þ 20 logðdÞ þ 20 logðf Þ ð1Þ
ment environments (Balachander et al., 2013; AlSayyari et al.,
2014a, 2014b, 2014c; Otero et al., 2014; Sabri et al., 2011). where f is the frequency in MHz and d is the separation space of the
Most of the published work concerning outdoor environments transmitter and receiver antennas in metres.
depends on FSPL and 2-Ray models (Pande et al., 2012; Willis
and Kikkert, 2007; Neto et al., 2010). The FSPL model supposes that 2.1.2. Ground reflection model
transceiver antennas, both transmitters and receivers, use line-of- When the RF propagates near the ground with an LOS, the plane
sight (LOS) communication, with no obstructions or reflections of earth (PE) wave propagation model can be used to define the path
any type. In real situations, there are almost always obstructions loss instead of the FSPL model. This model includes the effects of
in or near the propagation path or the surface from which radio ground ray reflection and the LOS ray, which is given by Eq. (2):
waves can be reflected (Mao et al., 2007). Therefore, the FSPL
model is considered very optimistic for predicting the RSS between PLPE ðdBÞ ¼ 40 logðdÞ  20 logðhT Þ  20 logðhR Þ; ð2Þ
A. Raheemah et al. / Computers and Electronics in Agriculture 127 (2016) 553–560 555

where d is the distance between the transmitter and receiver anten- Using the COST 235 model, measurements were performed on
nas in metres and hT and hR are the elevations of the transceiver in-leaf and out-of-leaf trees in two seasons. The COST 235 model
antennas in metres. The separation distance (d) in this model is is valid for the 200 MHz–95 GHz frequency range. Frequency in
assumed to be much larger than hT and hR. the COST 235 and ITU-R models are displayed in MHz, while the
depth of the trees is displayed in metres.
2.1.3. Log-distance model In general, the excess foliage-induced loss can be symbolised by
In most practical applications, the RSS for the same transmis- the following expression (Rogers et al., 2002):
sion distance is different (Liao and Sarabandi, 2005). Hence, varia- B C
tion in path loss due to location can be modelled using Eq. (3): PLVeg ¼ A  f  d : ð7Þ
  The three parameters in (7), A, B and C, can be empirically com-
d
PdB ðdÞ ¼ Pðd0 Þ þ 10a log10 ; ð3Þ puted. The A value is determined based on the foliage type, and the
d0
B and C values represent the frequency and distance dependencies,
where P(d0) is the path loss in dB at the reference (d0) and a is the respectively. In Al-Nuaimi and Stephens (1998), Al-Nuaimi and
path loss exponent that represents the rate of the path loss value Stephens adopted the least-square error method to optimise the
increase as a function of distance. value estimation of the numerical parameters using numerous sets
The log-distance model (3) considers the impact of vegetation, of measurands. The collected datasets used 11.2 GHz and 20 GHz
foliage, etc., on loss through different values of the path loss expo- frequencies during two foliation states, in-leaf and out-of-leaf.
nent and intercept. The log-normal distance expression indicates Thus, they derived the fitted ITU-R (FITU-R) model presented in
that the average signal power decreases logarithmically with the Eq. (8):
distance (Rappaport, 2002). 8 2 0:18 0:59
>
< 37  10  f d no leaf
2.2. Foliage models PLFITUR ðdBÞ ¼ 39  10  f
2 0:39
d
0:25
in leaf : ð8Þ
>
:
The signal propagation components in the foliage intermediate
impinge extra excess loss, such as that caused by direct and
2.3. Total path loss (PLtot)
reflected waves. Researchers are currently establishing methods
for including these factors by introducing empirical foliage mod-
When a wireless signal arrives at the receiver (Rx), it has already
elling to prevent excess loss for various geometries and frequencies
suffered attenuation along the propagation path. This attenuation
(Seybold and John, 2005; Rappaport, 1996; COST 235, 1996;
influences the received power (Pr), which can be expressed as a
Al-Nuaimi and Stephens, 1998). The well-known empirical models
function of the transmitted power (Pt), transmitter and receiver
are listed below.
antenna gains (Gt, Gr) and the total path loss (PLtot), as in Eq. (9):
2.2.1. Weissberger’s modified exponential decay model (Seybold and PLtot ¼ ðPt þ Gt þ Gr Þ  Pr : ð9Þ
John, 2005)
The total path loss can be divided into the path loss caused by
Weissberger’s modified exponential decay model is applicable
wave spreading, the path loss caused by free space and the path
when a ray path is blocked by dense, dry, in-leaf trees found in
loss caused by the existing obstacles within the broadcasting path,
temperate climates. It is applicable in situations where propaga-
as in Eq. (10):
tion is likely to occur through a grove of trees rather than through
diffraction over a canopy of trees. It is given by Eq. (4): PLtot ¼ PLfree space þ PLv eg ; ð10Þ
( 0:284 0:588
1:33  f d 14 m < d 6 400 m where PLtot is the total path loss, PLfree space is the path loss in free
PLW ðdBÞ ¼ 0:284
; ð4Þ space and PLveg is the path loss caused by obstacles.
0:45  f d 0 m 6 d < 14 m

where d is the depth of the foliage in metres and f is the frequency 2.4. Measurement campaign
in GHz. The frequency range over which this model is valid between
230 MHz and 95 GHz. All the propagation models were created in MATLAB (V.
R2013b). These models were applied to the mango greenhouse
2.2.2. ITU recommendation (ITU-R) (Rappaport, 1996) used in the experiment, which is described in Section 2.2. MATLAB
The ITU-R model was developed from measurements conducted was also used for all data processing and figure generation.
mainly at ultra-high frequency (UHF) and is proposed for cases
where either the transmitter or the receiver antenna is near to a 2.4.1. Experimental site
small grove of trees so that the majority of the signal propagates The collected measurements were taken at a mango greenhouse
through the trees. This model is commonly used for frequencies located in the Perlis, the northern state of Malaysia at the border of
between 200 MHz and 95 GHz and is depicted in Eq. (5), where d Thailand. The dimensions of the greenhouse are 50 m  10 m 
is the depth of the foliage in metres: 5 m. The greenhouse is composed of three lanes with 13 mango
0:3 0:6
trees in each row. The trees were 5 years old (Fig. 1). The geomet-
PLITUR ðdBÞ ¼ 0:2f d ð5Þ rical characteristics of the trees were as follows: a mean maximum
height of 2 m (r = 0.21); a main trunk height of 1 m (r = 0.12) and
2.2.3. The COST 235 model (COST 235, 1996) a mean trunk diameter of 0.16 m (r = 0.005). The separation dis-
The COST 235 model was proposed based on measurements tance between each tree in the same row was approximately
made on millimetre-wave frequencies (9.6–57.6 GHz) through a 3.2 m, and the separation distance between each row was 2.2 m.
small grove of trees. It is presented in Eq. (6):
2.4.2. Measurement set-up
(
26:6  d
0:5
out of leaf Measurements were performed using wireless nodes developed
PLCOST235 ðdBÞ ¼ 0:009 0:26
: ð6Þ by the NXP Company. This RF module adopts the ZigBee protocol
15:6  f d in leaf
stack based on the IEEE 802.15.4 standard (Yu et al., 2009). The
556 A. Raheemah et al. / Computers and Electronics in Agriculture 127 (2016) 553–560

(0.5 m, 1.0 m, 1.5 m, 2 m, 2.5 m, 3 m and 3.5 m). The receiver nodes
(Rx) were positioned at each individual tree at the same height
(Tx = Rx).
The empirical results were computed with respect to the energy
of the RSS indicator (RSSI), which offers an assessment of the
strength of the radio wave in dB units. To predict the energy
received, 500 packets were transmitted at an RF data rate of
250 Kbps. The receiving node was connected to a laptop, where
all the RSSI readings were recorded directly to a link quality indi-
cator (LQI). This process is depicted in Fig. 4.
Eq. (11) was used to compute the values of the RSS in dB based
on the LQI measurement (ZigBee development kit, 2013). Then, the
average value of the RSS measurement was used to calculate the
total path loss, which was considered independent, i.e., existing
outside the parameters of the transmitter and receiver system. It
is a positive quantity that represents signal attenuation, and it is
used to measure degradation in the strength of the signal as a func-
tion of distance. For the sake of simplicity, the focus of the discus-
Fig. 1. Mango greenhouse.
sion for the rest of this paper is placed on path loss rather than on
the RSS. The model is depicted in Eq. (11):
Table 1
 
LQI
Set-up of JN5148 wireless nodes. RSSðdBÞ ¼  Rsen  30 ; ð11Þ
2:5
Channel number 15
Transmit power 0 dBm where Rsen is the receiver sensitivity and is equal to (95 dBm) in
Frequency operation 2.245 GHz our experiment. Generally, the relationship between path loss,
Power level 24
transmitted power and received power can be expressed by Eq.
Power mode 0 dB
Retry amount 3 (9). The total path loss (PLtot) value at each measured point is given
Data rate 250 bps by Eq. (12), as the power transmitted (Pt) is 0 dB:
Receiver sensitivity 95 dBm
Antenna type Omni-directional PLtot ðdBÞ ¼ RSSðdBÞ: ð12Þ

3.1. Comparison between the actual measurements and the path loss
predicted by large-scale propagation models

Fig. 5 represents the actual RF measurements in the mango


greenhouse at seven different transceiver elevations (0.5 m,
1.0 m, 1.5 m, 2 m, 2.5 m, 3 m and 3.5 m). The receiver (Rx) was
located at each individual tree at the same height (Tx = Rx).
As shown in Fig. 5, the greatest vegetation effects appear within
1.5 m of vegetation depth, where the path loss is 86.6 dB within a
3.6 m distance and shows an increasing value until 108.6 dB at a
21.5 m distance. This height was adopted in this research to study
and analyse the vegetation models in the mango greenhouse using
WSNs.
Table 2 represents the path loss predicted by the large-scale
propagation models (FSPL and 2-Ray) at a height of 1.5 m for both
the transmitter and receiver antennas (Tx = Rx = 1.5 m).
As shown in Fig. 6, large differences exist between the path loss
predicted by the large-scale propagation models (FSPL and 2-Ray)
and the empirical measurements.
Fig. 2. Wireless sensor network experimental set-up.
The differences in path loss prediction can be calculated using
absolute percentage error (APE) and mean absolute percentage
error (MAPE). The APE represents the difference between the
set-up of the WSN modules are shown in Table 1. The nodes used
approximate and actual values as a percentage of the actual value.
in this work consisted of two modules of wireless nodes, one acting
The APE is given by Eq. (13):
as the coordinates (the receiver node) and the other as the end
 
device (the transmitter node), as shown in Fig. 2. X i;model  X i;actual
APE ¼ abs  100%; ð13Þ
X i;actual

3. Results and discussion where Xi,model is the path loss value predicted by the models and
Xi,actual is the actual path loss value gained by empirical
The mean geometry of the trees in the mango greenhouse is measurements.
shown in Fig. 3. Measurands were collected to approximate the The MAPE represents the average value of the APE. The MAPE
reduced energy of RF propagation within a row of trees in the measures the prediction accuracy of an expectation method in
mango greenhouse. The transmitter nodes (Tx) were positioned statistics and is considered a good method to measure the accuracy
facing the first tree in each row at seven different elevations of propagation models. Table 3 represents the APE of the path loss
A. Raheemah et al. / Computers and Electronics in Agriculture 127 (2016) 553–560 557

Direct Wave Ground-reflected Wave


Fig. 3. Average tree geometry.

Fig. 4. Jennic wireless network node serial data.

Table 2
Path loss predicted by FSPL and 2-Ray models in dB.

Large-scale path loss Distance


models
3.6 m 7.4 m 10.8 m 14.4 m 18 m 21.5 m
FSPL 51.27 57.53 60.81 63.31 65.25 66.79
2-Ray 15.21 27.73 34.29 39.29 43.17 46.25

predicted by large-scale propagation models (FSPL and 2-Ray) and


that of the empirical measurements.
The above comparison between widely used large-scale propa-
gation models (FSPL and 2-Ray) and the empirical measurements
produces large MAPEs (38.20% and 65.74%). The large MAPE
between FSPL, 2-Ray models and the empirical measurements
Fig. 5. Empirical path loss measurements. indicates their unsuitability as a path loss prediction tool inside
558 A. Raheemah et al. / Computers and Electronics in Agriculture 127 (2016) 553–560

Fig. 6. Path loss predicted by large-scale propagation models and empirical Fig. 7. Path loss predicted by combining vegetation models with the FSPL model.
measurements.

Table 3
APE and MAPE of large-scale propagation models and empirical measurements (%).

Model Distance
3.6 m 7.4 m 10.8 m 14.4 m 18 m 21.5 m
FSPL 40.80 38.67 37.57 37.32 36.90 38.50
2-Ray 82.44 70.44 64.79 61.10 58.25 57.41

the greenhouse environment, as both models are based on simplis-


tic approaches. The greenhouse environment is composed of vari-
ous factors, such as reflection, diffraction and scattering, which are
not considered precisely by these models. In addition, the 2-Ray Fig. 8. Path loss predicted by combining vegetation models with the 2-Ray model.
model considers only main and reflected rays, whereas the green-
house environment involves multipath rays.
Table 6
APE and MAPE of large-scale propagation models combined with the COST 235 model
3.2. Comparison between the actual measurements and the path loss (%).

predicted by large-scale propagation models combined with Model Distance


vegetation models 3.6 m 7.4 m 10.8 m 14.4 m 18 m 21.5 m
FSPL + COST 235 17.37 12.58 9.84 8.51 7.08 8.76
Tables 4 and 5 show the PLtot computation based on Eq. (10),
2-Ray + COST 235 59.01 44.35 37.07 32.29 28.43 27.67
which represents the total path loss computed by combining the
path loss predicted by the FSPL and 2-Ray models with foliage
models. The transceivers were 1.5 m high, and with different sep-
As shown in Figs. 7 and 8, large differences exist between the
aration distances between Tx and Rx were used.
path loss predicted by the large-scale propagation models com-
The comparison of total path loss gained in Tables 4 and 5 with
bined with the vegetation models and the empirical results. The
the empirical total path loss in Fig. 5 shows that the COST 235
APE and the MAPE for both are shown in Table 6.
model had the nearest result to the empirical test, as shown in
The above comparison between the results of (FSPL + COST 235)
Figs. 7 and 8.
and (2-Ray + COST 235) and the empirical measurements produces
large MAPEs (10.69% and 38.14%, respectively). (FSPL + COST 235)
Table 4 has better results than (2-Ray + COST 235), but it is still not
Total path loss by (PLFSPL + PLVeg) (dB). suitable as a path loss prediction tool inside the greenhouse
Model Distance
environment.

3.6 m 7.4 m 10.8 m 14.4 m 18 m 21.5 m


3.3. New model optimisation
WES + FSPL 53.35 61.81 67.06 71.52 74.61 77.18
ITU-R + FSPL 55.74 64.42 69.45 73.58 76.99 79.85
FITU-R + FSPL 62.50 70.97 75.59 79.19 82.03 84.34 While the combination of the COST 235 model with the FSPL
COST 235 + FSPL 71.56 82.00 87.81 92.41 96.08 99.09 model gives a good prediction of vegetation loss in the greenhouse
environment, there is still a significant error between measured
and predicted path loss, as shown in Fig. 7. In order to improve
Table 5
path loss prediction inside the mango greenhouse, a new model
Total path loss by (PL2-Ray + PLVeg) (dB). is derived and evaluated. Fig. 9 represents the actual total path loss
and the linear regression line of the measurements.
Model Distance
The relation of path loss to the logarithmic distance measurand
3.6 m 7.4 m 10.8 m 14.4 m 18 m 21.5 m is modelled using linear regression methods. Linear regression
WES + 2-Ray 17.29 32.01 40.54 47.50 52.53 56.64 attempts to model the mutual effects of two variables by fitting a
ITU-R + 2-Ray 19.68 34.61 42.93 49.56 54.91 59.31 linear equation to the empirical or observed data. It works by find-
FITU-R + 2-Ray 26.43 41.17 49.07 55.17 59.95 63.80
ing the optimal fitting straight line over the measurand. Therefore,
COST 235 + 2-Ray 35.50 52.20 61.29 68.39 74.00 78.55
a simple linear equation in the form of (Y = a + bX) is used, where -
A. Raheemah et al. / Computers and Electronics in Agriculture 127 (2016) 553–560 559

4. Conclusion

This research proves that large-scale propagation models (FSPL


and 2-Ray) are inaccurate when used with foliage models to pre-
dict total path loss in vegetation environments. The MAPE of these
models diverged by 10.69% and 38.14% from the actual measure-
ments, respectively.
An analysis of the measured data shows that the path loss
model based on COST 235 and FSPL models had the best perfor-
mance compared to the empirical measurements versus other foli-
age models. However, this path loss model is still not optimal, as
the MAPE was 10.69%. Therefore, this study presented a new
empirical path loss model, the LRCFM, for the mango greenhouse.
Fig. 9. Actual path loss and linear regression line. The LRCFM parametres are derived from empirical experiments
in which the model was used.
To determine the performance of the LRCFM model, the MAPE
X is the descriptive variable and Y is reliant on the variable. The line was computed. The result was 2.75% compared to the 10.69% of
slope is defined by b, and a is the intercept, i.e., the value of y the FSPL model combined with the COST 235 model. Based on the-
when x = 0. oretical and empirical results, the LRCFM model is the most effec-
Various software packages can be used to find linear regression tive total path loss model for representing the vegetation loss in
equations based on observed data. In this research, the authors this environment.
used MATLAB (V.R2013b) and Microsoft Excel, which has a built-
in function to find the linear regression. Both software programs
gave the same linear regression equation. Eq. (14) represents the References
linear regression equation for the empirical measurements:
Akyildiz, I.F., Vuran, M.C., 2010. Wireless Sensor Networks, Localization. Wiley.
Al-Nuaimi, M.O., Stephens, R.B.L., 1998. Measurements and prediction model
y ¼ 27x þ 71: ð14Þ optimization for signal attenuation in vegetation media at centimeter wave
frequencies. IEE Proceedings - Microwaves, Antennas and Propagation 145 (3),
To model path loss for RF propagation in the mango greenhouse, 201–206.
the log-normal distance model in Eq. (3) was correlated with Eq. AlSayyari, A., Kostanic, I., Otero, C.E., 2014a. An empirical path loss model for
(14), which results in the path loss equation and parametres in wireless sensor network deployment in dense-tree environment. 2014 IEEE
World Forum on Internet of Things (WF-IoT) 2014, 218–223.
Eq. (15): AlSayyari, A., Kostanic, I., Otero, C.E., 2014b. An empirical path loss model for
wireless sensor network deployment in a concrete surface environment. In:
PLLRCFM ¼ 71 þ 27 log10 ðdÞ; ð15Þ 2015 IEEE 16th Annual Wireless and Microwave Technology Conference
(WAMICON), pp. 1–6.
AlSayyari, Kostanic, I., Otero, C.E., 2014c. An empirical path loss model for wireless
sensor network deployment in an artificial turf environment. In: 2014 IEEE 11th
3.4. Verifying the performance of the new proposed model International Conference on Networking, Sensing and Control (ICNSC), pp. 637–
642.
Alsayyari, Abdulaziz, Melbourne, F.L., Kostanic, I., Otero, Carlos, Almeer,
Table 7 represents the actual measurements and the PLtot Mohammed, 2014d. An empirical path loss model for wireless sensor
gained from the new linear regression curve-fitting model network deployment in a sand terrain environment. In: 2014 IEEE World
Forum on Internet of Things (WF-IoT), pp. 218–223.
(LRCFM).
Andrade-Sanchez, P., Pierce, F.J., Elliot, T.V., 2007. Performance Assessment of
Table 8 shows the APE and MAPE between the PLtot gained by Wireless Sensor Networks in Agricultural Settings, vol. 7. In: 2007 ASABE
the actual measurements and the PLtot predicted by the new Annual International Meeting, Technical Papers.
proposed model in Eq. (15). A comparison between the two PLtot Balachander, D.T., Rao, Rama, Mahesh, G., 2013. RF propagation investigations in
agricultural fields and gardens for wireless sensor communications. In: 2013
values results in a small MAPE of 2.75%. IEEE Conference on Information & Communication Technologies (ICT), pp. 755–
Thus, the MAPE indicates that the new proposed model is a 759.
suitable path loss prediction tool in the mango greenhouse Balachander, D., Rama Rao, T., Mahesh, G., 2013. RF propagation investigations in
agricultural fields and gardens for wireless sensor communications. In: 2013
environment. IEEE Conference on Information & Communication Technologies (ICT), pp. 755–
759.
Correia, Felipe P., Alencar, Marcelo S., Carvalho, Fabrício B.S., Lopes, Waslon T.A.,
Table 7 Leal, Braulio G., 2013. Propagation analysis in Precision Agriculture
PLtot by empirical measurements and the PLLRCFM model (dB). environment using XBee devices. In: Microwave & Optoelectronics
Conference, pp. 1–5.
Model Distance COST 235, 1996. Radio propagation effects on next-generation fixed-service
terrestrial telecommunication systems, Luxembourg (Final Rep.).
3.6 m 7.4 m 10.8 m 14.4 m 18 m 21.5 m
Lewis, F.L., 2004. Smart Environments: Technologies, Protocols, and Applications.
PLtot 86.6 93.8 97.4 101 103.4 108.6 Wireless Sensor Networks, New York, pp. 11–46 (Chapter 2).
PLLRCFM 79.35 94.46 98.90 102.27 104.89 106.97 Liao, D., Sarabandi, K., 2005. Near-earth wave propagation characteristics of electric
dipole in presence of vegetation or snow layer. IEEE Trans. Antennas Propag. 53,
where P(d0) = 71 m, a = 2.7 m and d0 = 1 m. 3747–3756.
Mao, G., Anderson, B.D.O., Fidan, B., 2007. Path loss exponent estimation for
wireless sensor network localization. Comput. Networks 51 (10), 2467–2483.
Meng, Song, Yu, Lee, Yee Hui, Ng, Boon Chong, 2009. Empirical near ground path
Table 8 loss modeling in a forest at VHF and UHF bands. IEEE Trans. Antennas Propag.
APE and MAPE for the actual measurements and PLtot values predicted by the new 57 (5), 1461–1468.
model (%). Mestre, P., Serodio, C., Morais, R., Azevedo, J., Melo-Pinto, P., 2010. Lecture notes in
engineering and computer science, Vegetation growth detection using wireless
Model Distance sensor networks, vol. I. In: Proceedings of the World Congress on Engineering,
pp. 802–807.
3.6 m 7.4 m 10.8 m 14.4 m 18 m 21.5 m Morais, R., Fernandes, M.A., Matos, S.G.C., Ferreira Serodio, P., Reis, M., 2008. A
PLLRCFM 6.85 3.77 2.83 1.44 0.91 2.47 ZigBee multipowered wireless acquisition device for remote sensing
applications in precision viticulture. Comput. Electron. Agric. 62 (2), 94–106.
560 A. Raheemah et al. / Computers and Electronics in Agriculture 127 (2016) 553–560

Neto, J.S., Neto, J.R., Yang, Y., Glover, I., 2010. Plausibility of practical low-cost Sabri, N., Aljunid, S.A., Ahmad, R.B., Malek, M.F., Yahya, A., Kamaruddin, R., Salim, M.
location using WSN path-loss law inversion. In: IET International Conference on S., 2012. Performance evaluation of wireless sensor network channel in
Wireless Sensor Network, 2010. IET-WSN, pp. 260–265. agricultural application. Am. J. Appl. Sci. 9 (1), 141–151.
Otero, C.E., Kostanic, I., Otero, L.D., 2009. Development of a simulator for stochastic Sabri, N., Aljunid, S.A., Ahmad, R.B., Malek, M.F., Yahya, A., Kamaruddin, R., Salim, M.
deployment of wireless sensor networks. J. Networks 4 (8), 754–762. S., 2013. Path loss analysis of WSN wave propagation in vegetation. J. Phys:
Otero, C.E., Shaw, W.H., Kostanic, I., Otero, L.D., 2010. Multiresponse optimization of Conf. Ser. 423 (1).
stochastic WSN deployment using response surface methodology and Sabri, N., Aljunid, S.A., Ahmad, R.B., Malek, M.F., Yahya, A., Kamaruddin, R., Salim, M.
desirability functions. IEEE Syst. J. 4 (1), 39–48. S., 2013. Information and control, cognitive wireless sensor actor network: an
Otero, C.E., Haber, R., Peter, A., AlSayyari, A., Kostanic, I., 2014. A wireless sensor agricultural perspective. Int. J. Innovat. Comput. 10 (2), 631–658.
networks’ analytics system for predicting performance in on-demand Savage, N., Ndzi, D., Seville, A., Vilar, E., Austin, J., 2003. Radio wave propagation
deployments. IEEE Syst. J. 9 (4), 1344–1353. through vegetation: factors influencing signal attenuation. Radio Sci. J. 38 (5),
Otero, C.E., Haber, R., Peter, A., AlSayyari, A., Kostanic, I., 2014. A wireless sensor 108.
networks analytics system for predicting performance in on-demand Sawant, R.P., Liang, Q., Popa, D.O., Lewis, F., 2007. Experimental path loss models for
deployments. IEEE Syst. J. 9 (4), 1344–1353. wireless sensor networks. IEEE Milit. Commun. Conf., 1–7
Pande, M., Choudhari, N.K., Pathak, S., 2012. Energy efficient hybrid architecture and Scott, T., Wu, K., Hoffman, D., 2006. Radio propagation patterns in wireless sensor
positioning of sensors in WSN for precision agriculture. In: Proceedings of the networks: new experimental results. In: International Conference on Wireless
CUBE International Information Technology Conference, pp. 198–203. Communications and Mobile Computing, pp. 857–862.
Rappaport, Theodore S., 1996. Prentice Hall PTR, New Jersey, vol. 2 (Chapter 3). pp. Seybold, John S., 2005. Introduction to RF Propagation. John Wiley & Sons.
70–73. Willis, S., Kikkert, C., 2007. Radio propagation model for long-range wireless sensor
Rappaport, T.S., 2002. Mobile Radio Propagation: Large-Scale Path Loss,, .. Wireless networks. In: 2007 6th International Conference on Information,
Communications: Principle and Practice, second ed. Prentice Hall. Communications & Signal Processing, pp. 1–5.
Rogers, Neil C., et al., 2002. Radio Agency. A generic model of 1–60 GHz radio Yu, Chengbo, Cui, Yanzhe, Zhang, Lian, Yang, Shuqiang, 2009. ZigBee wireless sensor
propagation through vegetation-final report. network in environmental monitoring applications. In: 5th International
Sabri, N., Aljunid, S.A., Ahmad, R.B., Malek, M.F., Yahya, A., Salim, M.S., Kamaruddin, Conference on Wireless Communications, Networking and Mobile Computing.
R., 2011. Towards smart wireless sensor actor networks: design factors and IEEE, pp. 1–5.
applications. ISIEA2011, IEEE Symposium on Industrial Electronics and Zhou, G., He, T., Krishnamurthy, S., Stankovic, J.A., 2004. Impact Radio Irregularity
Applications, vol. 1, pp. 704–708. on Wireless Sensor Networks. In: MobiSys: Proceedings of the 2nd International
Sabri, N., Aljunid, S.A., Ahmad, R.B., Malek, M.F.A., Kamaruddin, R., Salim, M.S., 2012. Conference on Mobile Systems, Applications, and Services, pp. 125–138.
In: Antennas and Propagation Conference (LAPC), Wireless Sensor Network ZigBee development kit. <http://www.jennic.com/files/support_files/JN-UG-3062-
Wave Propagation in Vegetation: Review and Simulation, 2012 Loughborough, JN5148-EK010-User-Guide.pdf> (last accessed 26th August 2013).
pp. 1–4.

You might also like