Telecommunication Systems (2022) 79:309–321

[Link]

Localization of isotropic and anisotropic wireless sensor networks in

2D and 3D fields
Soumya J. Bhat1 · K. V. Santhosh1

Accepted: 12 November 2021 / Published online: 6 December 2021

© The Author(s) 2021

Abstract
Internet of Things (IoT) has changed the way people live by transforming everything into smart systems. Wireless Sensor
Network (WSN) forms an important part of IoT. This is a network of sensor nodes that is used in a vast range of applications.
WSN is formed by the random deployment of sensor nodes in various fields of interest. The practical fields of deployment can be
2D or 3D, isotropic or anisotropic depending on the application. The localization algorithms must provide accurate localization
irrespective of the type of field. In this paper, we have reported a localization algorithm called Range Reduction Based
Localization (RRBL). This algorithm utilizes the properties of hop-based and centroid methods to improve the localization
accuracy in various types of fields. In this algorithm, the location unknown nodes identify the close-by neighboring nodes
within a predefined threshold and localize themselves by identifying and reducing the probable range of existence from
these neighboring nodes. The nodes which do not have enough neighbors are localized using the least squares method. The
algorithm is tested in various irregular and heterogeneous conditions. The results are compared with a few state-of-the-art
hop-based and centroid-based localization techniques. RRBL has shown an improvement in localization accuracy of 28% at
10% reference node ratio and 26% at 20% reference node ratio when compared with other localization algorithms.

Keywords Anisotropic network · Heterogeneous network · Internet of things · Irregular fields · Range-free localization ·
Wireless sensor network

1 Introduction Most of these applications require the location of sensor

nodes to be known. This is necessary to understand the loca-
The growth of the Internet of Things (IoT) has revolution- tion of observed events and to act on them [9]. Consider the
ized the digital world. IoT has changed the way people live case of the military surveillance application. Here, WSN is
by transforming everything into smart systems [1]. Wireless formed by dropping sensor nodes from an aerial vehicle like
Sensor Network (WSN) is an important part of IoT [2]. This a helicopter or drone in the region of interest. These nodes
is a network of multiple sensor nodes used to monitor the track enemy movements using the attached sensors and send
events of the physical world, digitize them and send infor- the sensed information to a base station located in a remote
mation to the systems or end-users [3]. WSN is used in a place. The data at the base station is further analyzed to take
number of applications of smart cities, healthcare systems, necessary actions. Similarly, in the applications of intelligent
intelligent transportation systems, disaster management, mil- transportation systems, WSN is formed by deploying sensor
itary surveillance, etc. [4–8]. nodes on either side of highways. Sensors attached to nodes
monitor traffic congestions and send this information to the
traffic control system for automatic controlling of traffic. In
B K. V. Santhosh all the above-discussed applications, the collected informa-
[Link]@[Link] tion is useful only when the location of the occurrence of
Soumya J. Bhat events is known. By knowing the accurate location of enemy
[Link]@[Link] vehicles or traffic congestion, necessary actions can be taken
1 Department of Instrumentation and Control Engineering, automatically. But, in these applications, the location infor-
Center for Cyber Physical System, Manipal Institute of mation of sensor nodes is unknown as they are deployed
Technology, Manipal Academy of Higher Education, randomly. Placing the sensors in pre-determined locations
Manipal, India

123
310 S. J. Bhat, K. V. Santhosh

is not always possible, especially in large-scale applications as RSS, Time of Arrival (TOA), Angle of Arrival (AOA)
with harsh and unreachable areas of interest [10]. to estimate the distance between nodes [25–27]. Range-free
One way to get the location of randomly placed sensor localization algorithms use only connectivity information
nodes in large networks is by using satellite navigation sys- among nodes to estimate the distance between them. These
tems like Global Positioning System (GPS). But, equipping types of algorithms are more suitable for large-scale WSN
every sensor node with these systems increases the energy because they do not need any additional hardware on sensor
consumption, size, and cost of the nodes. Also, location nodes to measure the radio signal characteristics [23]. Hence,
estimations of these systems are not accurate in the indoor they can be readily used in existing sensor nodes. Several
environment and dense urban areas [11–13]. Hence, to over- range-free localization algorithms are available in the litera-
come this problem, several localization methods are derived ture. Most popular among these are hop-based techniques and
to estimate the location of sensor nodes in a WSN. centroid-based techniques [28]. Hop-based techniques make
The localization algorithms use few location-aware sen- use of hop lengths to reference nodes from sensor nodes for
sor nodes that are called reference nodes and apply different location estimation, whereas centroid-based techniques use
methods to estimate the proximity between location unknown connectivity to immediate neighbor nodes for location esti-
nodes and location-aware reference nodes [14]. The obtained mation.
proximity measures are further processed using different The localization techniques are used in different applica-
algorithms such as convex algorithm [15], optimized distance tions which require WSN to be deployed in different types of
[16], swarm intelligence algorithms [17], Hessian regular- fields. Some applications like air quality monitoring systems
ization regression [18], self localization protocol [19], deep require the deployment of sensor nodes in 2D plain fields, and
neural network [20], sine cosine algorithm [21]. A convex some applications like military surveillance and intelligent
algorithm based localization method is developed in [15] to transportation systems demand the deployment of nodes in
estimate the locations of unknown nodes in WSN used in anisotropic and irregular shaped 3-Dimensional (3D) fields.
real time monitoring of uranium tailings reservoir. Here, the The localization of sensor nodes in different types of fields
distances between the unknown nodes and reference nodes is a challenge [29]. The boundary condition of WSN in 3D
are obtained using the time difference of arrival method and fields becomes more complicated and the connectivity of the
then, the convex location algorithm is used to reduce the network can be more diverse compared to the 2D scenario
region of the unknown nodes and to estimate the coordinates [30]. Also, the anisotropy of fields results in the shortest path
of unknown nodes. An optimized distance range free local- between nodes to be bent, causing inaccurate localization.
ization algorithm reported in [16] limits the use of hop size Hence, it is important to analyze the impact of field charac-
and hop count to approximate nearly a straight line distance teristics on localization algorithms.
between a known and an unknown node without addi- In this paper, we have reported a range-free localiza-
tional hardware and without increasing extra communication. tion algorithm called Range Reduction Based Localization
Then, linear optimization is used for the location estimation (RRBL) for localization of nodes in 2D and 3D, isotropic
of nodes. A squared error norm of residual fitting error matrix and anisotropic fields. This algorithm uses the characteristics
cost function for mixed source localization is reported in of hop-based and centroid methods for efficient localization
[17]. The proposed cost function is optimized using particle of nodes under different scenarios. Performance is evalu-
swarm optimization, whale optimization, and grey wolf opti- ated in fields of various shapes such as 2D square, 2D C
mization algorithms to obtain accurate localization. Another shaped, 3D cube, 3D C shaped, and 3D mountain terrain.
localization algorithm is reported in [18] using Received Sig- The effect of field characteristics on localization algorithms
nal Strength (RSS) measurements. The locations of nodes is analyzed by evaluating the error introduced to localization
are then estimated using Hessian regularization regression. results when the field of deployment is changed from 2D to
Self localization protocol based localization method is a 3D and isotropic to anisotropic.
distributed and collaborative positioning algorithm that is The rest of the paper is organized as follows. In Sect. 2,
developed by exploiting the connectivity information among some of the existing localization algorithms are reviewed.
neighbors at one hop [19]. A deep neural network based Network model and network topology are discussed in
localization method is reported in [20] which explores the Sect. 3. A brief description of different range-free local-
correlation between the RSS data and the node placement to ization algorithms is given in Sect. 4. A novel localization
estimate accurate locations. A localization algorithm consid- algorithm called RRBL is discussed in Sect. 5 and the results
ering the sensor motion effect using time delay and Doppler are analyzed in Sect. 6. Finally, Sect. 7 provides the conclu-
frequency shift measurements are reported in [22]. These sion and proposed future work.
localization methods can be classified as range-based local-
ization and range-free localization [23,24]. Range-based
localization algorithms use radio signal characteristics such

123
Localization of isotropic and anisotropic wireless sensor networks... 311

2 Background and related work

There are several localization algorithms reported in the last

two decades. A few of the early range-free localization algo-
rithms are Centroid, DV-Hop, Approximate Point In Triangle
and Multidimensional Scaling [31–34]. These localization
algorithms were developed considering ideal network condi-
tions. Many enhancements were reported on these algorithms
Fig. 1 Communication range a DOI = 0 b DOI = 0.05
for improving localization accuracy. A multi-hop localization
algorithm with multisource information was reported in [23].
This is an error-compensation based node distance estimation denoted by N1 , N2 ,..., N N . Among these, K sensor nodes
that uses the information provided by the adjacent areas and are assumed to know their locations and are used as refer-
nodes. A localization algorithm based on the weighted cen- ence nodes. Other N-K nodes are location unaware nodes and
troid localization technique, where the positions of unknown their locations need to be estimated with the help of reference
nodes were calculated using the fuzzy logic method was nodes. These sensor nodes communicate with each other to
reported in [2]. An improved DV-Hop localization algorithm form an ad-hoc network.
based on hybrid chaotic strategy was reported in [35]. Jian Radio Irregularity Model (RIM):
[36] reported a new DV-Hop localization algorithm based The communication range of a node depends on the
on half-measure weighted centroid. A new hop-based local- transmission power of the sensor node and the environ-
ization algorithm based on the probability property of the ment of communication. Signals transmitted from nodes are
random distribution of wireless sensor network was reported attenuated by environmental factors. The transmitted signals
in [37]. This algorithm reduced the error accumulation by from nodes undergo reflection, refraction, and attenuation
control flooding and improved the positioning accuracy by by the environment during signal propagation. Propagated
using the jump distance correction method. A weighted cen- radio signals exhibit a continuous variation with incremental
troid DV-Hop algorithm was reported in [5]. Here, factors changes in direction. This characteristic of the radio signal
such as the number of anchors, communication radius, and is modeled using RIM, which was established based on data
nearest anchor were considered to determine the location of from real sensor devices [39]. This model uses a parame-
unknown nodes. A modified DV-Hop algorithm using Teach- ter called Degree of Irregularity (DOI) which is defined as
ing Learning Based Optimization was reported in [25]. Here, the maximum path loss percentage variation per unit degree
location errors caused by collinear anchor nodes are reduced change in the direction of radio propagation [39]. Signal
using the concept of collinearity. Further, Teaching Learning propagation in WSN is represented using the RIM model
Based Optimization is used to enhance localization accu- as follows:
racy. These localization algorithms were developed with the
motivation of improving localization accuracy. But, these Received signal strength = Transmission power -
algorithms have assumed ideal fields for the deployment of
DOI adjusted Path loss + Fading (1)
sensor nodes. Localization algorithms were evaluated in 2D
square experimental areas.
where DOI adjusted path loss = Path loss × K θ . K θ is a
Many applications demand the deployment of sensor
coefficient to represent the difference in path loss in different
nodes at different types of fields. Realistic fields of inter-
directions.
est are not always a plain field. Though a few algorithms
We have
like DisLoc [30], MSVR Based [38] have considered 3D
anisotropic fields for the analysis of localization algorithms, 
1 if θ = 0.
most of the developed algorithms have assumed ideal field Kθ =
dimensions. In this paper, we have introduced a localization K θ−1 ± Rand × D O I if θ ∈ {1, 2, 3, · · · , 359}.
algorithm that can be used to localize nodes in any type of (2)
field.
Rand in Eq. (2) is a random number which is found to fit
Weibull distribution [39]. When the DOI is at 0, the commu-
3 Network model nication range is a perfect circle/sphere. Let C R − 0i denote
the communication range of node i at DOI=0. As the DOI
We have considered a WSN comprising of N static sen- value increases, the communication range becomes more and
sor nodes randomly distributed in various fields of interest. more irregular. The communication range of a node at dif-
Every sensor node is assumed to have a unique identification ferent DOI is shown in Fig. 1.

123
312 S. J. Bhat, K. V. Santhosh

Heterogeneity Network is considered to be heterogeneous 4.1 Hop-based localization methods

with sensor nodes having different transmission powers. In
practical cases, the transmission power of any two sensor These methods make use of a few location-known reference
nodes can differ due to their different battery status, hardware nodes, the number of hops to reference nodes from sensor
differences and different specifications followed by manu- nodes, and average hop length to estimate their locations in
facturer [39,40]. This results in different coverage ranges of the network.
nodes. DV-Hop is a classic hop-based localization algorithm that
Network Topology Different applications of WSN demand has attracted more attention from researchers due to its sim-
deployment of nodes in different fields. For example, in smart plicity, stability, feasibility and less hardware requirement
city applications, sensor nodes need to be deployed in var- [25]. DV-Hop is a range-free localization algorithm that esti-
ious industrial or commercial buildings of different shapes mates locations of unknown nodes by making use of few
and sizes. In military surveillance applications or disaster location-aware reference nodes [32,43,44]. In the first step
management applications like forest fire detection, the field of this algorithm, all reference nodes broadcast their location
of deployment can be a forest area with multiple hills, val- information to their neighbor nodes with the hop count set
leys, and lakes. In all the above cases, the field of interest to one. Every neighbor node on receiving this information
where WSN needs to be formed is irregular and 3D. But stores reference node location and hop count to reference
most of the existing research has restricted the analysis of node. Then they increment the hop count value and broad-
localization algorithms to regular shaped 2D fields. cast reference node location and new hop count information
To ensure that the reported localization algorithm works to their neighbor nodes. This process is continued till every
smoothly in different types of fields, we have considered node has a minimum hop count value to every other reference
multiple realistic fields. A square-shaped field with multiple node and there are no new neighbor nodes to be updated with
hills and valleys is considered to represent the application of this information.
military surveillance. Here, nodes are deployed on the sur- In the second step, reference nodes measure the average
face of mountain terrain as shown in Fig. 2a. To represent per-hop distance as the ratio of the sum of distances with
the application of a smart city where nodes are deployed in other reference nodes to the total number of hop counts. In
buildings of different shapes, as shown in Fig. 2b, a 3D-C the next step, using the average hop distance and number
shaped deployment field is considered. Another 3D deploy- of hop counts to reference nodes, every node measures its
ment field of cubic shape is considered to represent isotropic approximate distance to reference nodes.
3D fields which is shown in Fig. 2c. In the last stage, coordinates of unknown target nodes are
Localization algorithms are also tested in 2D isotropic and estimated using the least-squares solution [45].
anisotropic fields. A 2D square field and a C-shaped field are Least squares solution: Let there be K reference nodes
considered to represent the deployment of nodes in plain with coordinates (xi , yi , z i ) where i = 1, 2, ..., k. Let
ground. This is shown in Fig. 3. (xu , yu , z u ) be the location of a location unknown sensor
Anisotropy and irregularities in the fields affect the node which needs to be estimated. Let dui be the estimated
distance measurements between nodes and thus degrade distance from i th reference node to this location unknown
the accuracy of localization algorithms [41,42]. Therefore, node at (xu , yu , z u ). The distance equations between k ref-
the localization algorithms are evaluated in these fields to erence nodes and location unknown node can be written as
eliminate the influence of field elements on localization accu- shown here.
racies.

(xu − x1 )2 + (yu − y1 )2 + (z u − z 1 )2 = du1

2

4 Range-free localization methods (xu − x2 )2 + (yu − y2 )2 + (z u − z 2 )2 = du2

2

.
Range-free localization methods are the simplest localization
techniques that are most suitable for WSN with limited bat- .
tery power and computational resources. Unlike range-based .
methods which make use of RSSI, TOA, AOA measure- (xu − xk )2 + (yu − yk )2 + (z u − z k )2 = duk
2
ments for distance estimations, range-free methods make use
of only connectivity information among nodes. There are This can be represented in matrix form as
mainly hop-based localization methods and centroid-based
localization methods. In this section, we have discussed a AX = B (3)
few state-of-the-art hop-based and centroid-based localiza-
tion techniques. where

123
Localization of isotropic and anisotropic wireless sensor networks... 313

Fig. 2 Deployment of location unknown sensor nodes and reference nodes in a mountain terrain b 3D C shaped field c 3D cubic field

Fig. 3 Deployment of location

unknown sensor nodes and
reference nodes in a 2D square
field b 2D C shaped field

and

⎡ ⎤
⎡ ⎤ 2 − d 2 − (x 2 − x 2 ) − (y 2 − y 2 ) − (z 2 − z 2 )
du1
−2(x1 − xk ) −2(y1 − yk ) −2(z 1 − z k ) ⎢
uk 1 k
.
1 k 1 k
⎥
⎢ . ⎥ ⎢ ⎥
⎢ ⎥ ⎢
b=⎢ . ⎥
⎢ ⎥, ⎥
A=⎢ . ⎥ ⎣ . ⎦
⎣ . ⎦ 2
du(k−1) − duk
2 − (x 2 − x 2 ) − (y 2 − y 2 ) − (z 2 − z 2)
k−1 k k−1 k k−1 k
−2(xk−1 − xk ) −2(yk−1 − yk ) −2(z k−1 − z k )
⎡ ⎤ The least squared solution can be obtained as in Eq. (4).
xu
X = ⎣ yu ⎦
zu X̂ = (A T A)−1 A T b (4)

123
314 S. J. Bhat, K. V. Santhosh

Fig. 4 Hop-based localization Fig. 5 Centroid-based localization

Hence, Ns receives the location information from reference

DV-maxHop [46] is another hop-based localization method nodes N2 , N3 and N4 . Then, Ns estimates its location as
that makes use of a control parameter to reduce the errors in arithmetic mean of coordinates of reference nodes N2 , N3
distance estimations. In practical cases, there exist obstruc- and N4 as shown in Eq. (5).
tions in the fields which cause detoured paths among nodes.
These detoured paths alter the minimum hop counts among
nodes and hence introduce errors in distance estimations. To 4 4 4
overcome this problem, the DV-maxHop algorithm uses a xs = xi , ys = yi and z s = zi (5)
control parameter to filter out hop count information from i=2 i=2 i=2
farther reference nodes. Only the closer reference nodes are
Few other centroid-based localization methods are WCL [48]
used and the locations are estimated similar to the DV-Hop
and EWCL [5]. These algorithms use weights to give more
algorithm. Consider a WSN with 4 reference nodes denoted
priority to closer reference nodes and less priority to farther
as N1 , N2 , N3 and N4 . If the control parameter is chosen as
reference nodes. WCL uses a weight of 1/hop count. Hence a
3, sensor node Ns uses only reference nodes N1 , N2 , and N3
farther reference node will have a higher hop count and lower
which are within 3 hop lengths. The reference node N4 which
weight whereas a closer reference node will have a smaller
is at 5 hop lengths away from Ns is not used for localizing
hop count and higher weight. EWCL uses a threshold hop
Ns . This is illustrated in Fig. 4.
value and only reference nodes within this threshold hop
Weighted DV-Hop [47] is another hop-based localization
value are considered. It uses a weight value defined as a
method that uses a weight function inversely proportional to
function of hop count, average hop distance of its nearest
the hop length. Instead of filtering out farther reference nodes,
node, and communication radius.
this algorithm gives a lower weight to farther reference nodes
Most of these algorithms are developed by assuming
and a higher weight to closer reference nodes.
homogeneous WSN in isotropic 2D fields. But, in reality,
WSN can be heterogeneous and fields of deployment are 3D
4.2 Centroid-based localization methods and anisotropic. Hence, to bridge this gap, we have devel-
oped a range reduction based localization technique that
These algorithms assume that sensor nodes are uniformly sur- considers the heterogeneity of nodes and improves localiza-
rounded by reference nodes. Hence, they localize unknown tion accuracy in both 2D and 3D fields. We have also done
nodes as the centroid of neighboring reference nodes. The a comparative analysis of the behavior of these algorithms
traditional Centroid algorithm [31] consists of two steps. In when the field of deployment is changed from 2D to 3D and
the first step, reference nodes broadcast their location infor- isotropic to anisotropic.
mation to all other nodes which are in their communication
range. In the second step, every unknown node estimates
its location in the network as the arithmetic mean of the 5 RRBL algorithm
coordinates of all the received reference nodes’ location
information. As illustrated in Fig. 5, node Ns is within the In this section, we describe the steps of the RRBL algo-
communication range of reference nodes N2 , N3 and N4 . rithm. This algorithm uses the properties of both centroid

123
Localization of isotropic and anisotropic wireless sensor networks... 315

and hop-based methods to overcome the influence of field y jmin−new = max(y jmin−old , yi − (C R − 0i j × ADi ))
irregularities and heterogeneity in the network. y jmax−new = min(y jmax−old , yi + (C R − 0i j × ADi ))
Phase 1: This step is similar to DV-Hop. But, instead of z jmin−new = max(z jmin−old , z i − (C R − 0i j × ADi ))
broadcasting hop counts, every reference node broadcasts its z jmax−new = min(z jmax−old , z i + (C R − 0i j × ADi ))
expected communication radius C R−0i at DOI=0 along with
(8)
location information. Nodes can know the expected com-
munication radius from their transmission power using log
Locations for node j is estimated as follows.
normal path loss model [42]. The neighbor nodes store the
received information, add their C R − 0 to the received value (x jmin + x jmax )
and broadcast it again. This is continued till every node in the xj =
2
network has information on the location of reference nodes (y jmin + y jmax )
and the minimum value of the sum of C R − 0 required to yj =
2
reach each reference node. (z jmin + z jmax )
Next, every reference node measures the Average Distance zj = (9)
2
(AD) in the network as in Eq. (6).
For localization, phase 2 requires at least a few neighbors
k within the threshold communication range. If the neighbor
di j count is very low, locations are estimated as described in
j=1
ADi = , i = j (6) phase 3.
k
C R − 0i j Phase 3: This step uses all the neighbors and distance to
j=1
each reference node from the unknown node is measured as
in Eq. (10).
where ADi is average distance for reference node i and di j
is the distance from reference node i to j. The measured AD
dui = C R − 0ui × ADi (10)
is then broadcast to other nodes in the network.
Phase 2: Every unknown node localizes itself indepen- Using these approximated distance values, location is esti-
dently from the obtained data in phase 1. It first identifies mated from the least squares method as in Eq. (4).
the neighboring reference nodes in its close proximity using
a threshold CR-0 value. We have chosen threshold value as Pseudo code for RRBL
the maximum CR-0 in the network. If the number of neigh-
bor nodes within the threshold CR-0 is more than 3, location
is estimated using range reduction method described here. Input: WSN with randomly deployed location unknown nodes and
For an unknown node j, let the neighbors within thresh- reference nodes.
old CR-0 distance be R1 , R2 , ...Rc with location (x1 , y1 , z 1 ), Output: Localization of unknown nodes.
(x2 , y2 , z 2 ), ..., (xc , yc , z c ). 1. Nodes in the network exchange information on
Initialize the range for (x j , y j , z j ) as in Eq. (7). . location and CR-0 of reference nodes.
2. Reference nodes measure the average distance
x jmin = x1 − (C R − 01 j × AD1 ) . using Eq. (6).
3. Unknown nodes identify the nearest reference
x jmax = x1 + (C R − 01 j × AD1 )
. nodes residing within the threshold CR-0 value.
y jmin = y1 − (C R − 01 j × AD1 ) 4. If (the number of nearest nodes ≥ 3)
y jmax = y1 + (C R − 01 j × AD1 ) 5. Initialize (x j , y j , z j ) as in Eq. (7).
z jmin = z 1 − (C R − 01 j × AD1 ) 6. For i=2 to (number of nearest nodes)
7. Update (x j , y j , z j ) as in Eq. (8).
z jmax = z 1 + (C R − 01 j × AD1 ) (7)
8. End for
9. Find the final location using Eq. (9).
The initialized range are then updated using other neighbors
10. Else If (the number of nearest nodes ≤ 3)
R2 to Rc .
11. Distance to each reference node from
. unknown node is measured using Eq. (10).
For i = 2 to c,
12. Location is estimated using least squares
. method using Eq. (4).
x jmin−new = max(x jmin−old , xi − (C R − 0i j × ADi ))
13. End If
x jmax−new = min(x jmax−old , xi + (C R − 0i j × ADi ))

123
316 S. J. Bhat, K. V. Santhosh

6 Results and discussion DV-Hop

5
4
In this section, the reported localization algorithm is eval- RRBL DV-maxHop
3
uated under various scenarios. The results are compared
with other localization algorithms such as DV-Hop [32], DV- 2
maxHop [46], Weighted DV-Hop [47], Centroid [31], WCL 1
[48] and EWCL [5]. These algorithms are selected as a com- 0
bination of classic and latest hop-based and centroid-based Weighted DV-
EWCL
Hop
methods. The algorithms are evaluated in different types of
fields namely 3D cubic field, 3D C shaped field, 3D moun-
tain terrain, 2D square field, and 2D C shaped field. These 3D cubic field
fields are chosen as a combination of 2D and 3D, isotropic WCL Centroid
2D square field
and anisotropic fields to eliminate the effect of field type on
the performance of the algorithm. Fields are assumed to have (a)
a path loss factor of 4. A heterogeneous set of sensor nodes
with transmission powers varying from -5dBm to -15dBm DV-Hop
6
are deployed in these fields. Random deployment of WSN is
5
simulated using MATLAB R2018a. The localization results
RRBL 4 DV-maxHop
are evaluated using Localization Error (LE) which is defined 3
as below [49]. 2
1
N −K 0
(xi,est − xi )2 + (yi,est − yi )2 + (z i,est − z i )2
i=1 Weighted DV-
LE = EWCL
Hop
R × (N − K )
(11)

where (xi,est , yi,est , z i,est ) is the estimated co-ordinate of i th WCL Centroid

node, (xi , yi , z i ) is the actual co-ordinate of i th node and R 3D C shaped field
is the average communication radius of nodes. N and K are (b) 2D C shaped field
number of sensor nodes and reference nodes respectively.
Fig. 6 Effect of dimensionality on a Cubic field b C shaped field at
6.1 Effect of dimensionality 10% reference node ratio

The influence of dimensionality is evaluated by comparing

fields than in 2D fields. This is shown in Tables 1, 2 and 3.
the localization results in the 2D square field with 3D cubic
Here, the run time to localize a single node using DV-Hop,
field and 2D C shaped field with 3D C shaped field. For this,
EWCL, and RRBL localization algorithms is compared in
400 sensor nodes are deployed in these fields with additional
2D and 3D fields. Irrespective of the shape of the field, with
10% − 20% reference nodes. The area/volume of every field
an increase in reference node ratio, the run time of all the
is maintained at 45000m 2 /45000m 3 . A uniform node den-
localization algorithms in 3D fields increases rapidly than in
sity of 0.01/m 2 in 2D fields and 0.01/m 3 in 3D fields is
2D fields.
maintained in all the fields. Results obtained from different
localization algorithms are shown in Figs. 6 and 7.
Figures 6 and 7 show the comparison of LE for different 6.2 Effect of anisotropy
algorithms in 2D and 3D fields. The proposed RRBL algo-
rithm has shown an improvement of 29% in 3D fields and The influence of anisotropy is studied by comparing the
27% in 2D fields at 10% reference node density and 35% in localization results in anisotropic fields with isotropic fields.
3D fields and 12% in 2D fields at 20% reference node density Various localization algorithms are evaluated in 2D and 3D
when compared with EWCL. All the localization algorithms fields. Localization results in a 2D C-shaped field are com-
show improved localization accuracies in 3D fields than in pared with its corresponding 2D regular field i.e., 2D square
2D fields. This is because the coverage area of a 3D node is field. Similarly, localization results of 3D mountain terrain
higher than that of a 2D node. Hence, at common node den- field and 3D C shaped field are compared with 3D cubic
sity, 3D fields have more connectivity than 2D fields. But, field. All the algorithms show increased errors in anisotropic
due to this, the run time of the algorithms increases in 3D fields. While the error is higher in traditional algorithms, the

123
Localization of isotropic and anisotropic wireless sensor networks... 317

DV-Hop reported RRBL shows better performance in the presence of

4
field anisotropy. Results obtained from different localization
3 DV-
RRBL algorithms are shown in Fig. 8. Figure 8a shows the compar-
2 maxHop
ison of LE in 2D fields and Fig. 8b shows the comparison of
1
LE in 3D fields. RRBL has shown an improvement of 26%
0 in 2D fields and 32% in 3D fields when compared with the
Weighted
EWCL other best performing algorithm, EWCL.
DV-Hop

3D cubic field
6.3 Effect of heterogeneity
WCL Centroid 2D square field
(a) The effect of different topologies on localization algorithms
is evaluated at different levels of heterogeneity. A hetero-
DV-Hop geneous network is formed by three sets of sensor nodes
4
with different transmission powers. The transmission pow-
3 DV-
RRBL ers of these nodes are varied and the effect of deviation in
2 maxHop
sending power on the performance of the RRBL localiza-
1 tion algorithm is evaluated. Table 4 shows the transmission
0 power of different sets of sensor nodes used in the experi-
Weighted
EWCL ment.
DV-Hop
3D C shaped field
Figure 9 shows the performance of the RRBL algorithm
in terms of LE for a heterogeneous network of sensor nodes.
WCL Centroid 2D C shaped field
Increasing variance in sending power generally degrades the
(b) performance of localization algorithms. But, in this algo-
rithm, the effect of deviation in sending power on localization
Fig. 7 Effect of dimensionality on a Cubic field b C shaped field at
20% reference node ratio
algorithms is reduced. When the heterogeneity is increased
from 0 to 5, there was no degradation in the performance
of RRBL. Further increase in heterogeneity to 10 dete-
riorated the performance of the localization algorithm by
17%.

Table 1 Run time (s)

Reference node ratio 3D cubic field 3D C shaped field 2D square field 2D C shaped field
comparison of DV-Hop
algorithm 10% 0.105 0.093 0.11 0.106
20% 0.215 0.183 0.138 0.132
30% 0.416 0.423 0.236 0.199
40% 0.932 0.95 0.359 0.400

Table 2 Run time comparison

Reference node ratio 3D cubic field 3D C shaped field 2D square field 2D C shaped field
of EWCL algorithm
10% 0.108 0.100 0.108 0.108
20% 0.220 0.2 0.134 0.130
30% 0.421 0.421 0.215 0.190
40% 0.95 0.43 0.332 0.412

Table 3 Run time comparison

Reference node ratio 3D cubic field 3D C shaped field 2D square field 2D C shaped field
of RRBL algorithm
10% 0.101 0.135 0.118 0.131
20% 0.226 0.2 0.147 0.141
30% 0.349 0.32 0.21 0.226
40% 0.53 0.58 0.343 0.47

123
318 S. J. Bhat, K. V. Santhosh

DV-Hop 1
6 0.9
0.8
5 0.7
RRBL 4 DV-maxHop 0.6

LE (m)
3 0.5
0.4
2 0.3 2D C shaped field
1 0.2 3D cubic field
0 0.1 3D C shaped field
0
Weighted DV- het:0 het:5 het:10
EWCL
Hop Heterogeneity

Fig. 9 Effect of heterogeneity on RRBL

WCL Centroid 2D C shaped field

2D square field 1.6 3D cubic field
(a) 2D C shaped field
1.4 3D C shaped field
1.2
1

LE (m)
DV-Hop 0.8
3
0.6
2.5 0.4
RRBL 2 DV-maxHop 0.2
1.5 0
1 400 500 600 700 800 900
0.5 Node density
0
Fig. 10 Influence of node density on RRBL
EWCL Weighted DV-Hop

Localization algorithms behave differently in different

fields. Hence, by analyzing the results for just 2D fields, the
WCL Centroid results for 3D fields cannot be estimated. Similarly, by ana-
3D cubic field lyzing the results for isotropic fields, results for anisotropic
3D mountain terrain fields cannot be estimated. Therefore, we have tested the
(b) 3D C shaped field reported localization algorithm RRBL at various anisotropic
conditions, heterogeneity, and node densities. The results
Fig. 8 Effect of anisotropy in a 2D fields b 3D fields show improvement in the accuracy of localization.

6.4 Effect of node density 7 Conclusion

The performance of the reported algorithm is tested for vary- In the present world of IoT, the WSN localization problem
ing node densities in 2D and 3D fields. The number of nodes is gaining much importance. The accuracy of localization
in the network is increased from 400 to 900 and the reference algorithms is a concern in irregular fields. Most of the exist-
node ratio is maintained at 10%. The results are illustrated ing algorithms in literature were developed to localize nodes
in Fig. 10. An increase in node density improved the perfor- in regular fields, whereas practical applications of WSN are
mance of the algorithm. The rate of improvement was high irregular fields. To help in bridging this gap, we have reported
at lower node densities and the change in LE was reduced at a localization algorithm called RRBL. This algorithm utilizes
higher node densities. the properties of both hop-based and centroid methods to
9 improve the localization accuracy in various types of fields.

Table 4 Heterogeneous set of

Sending power Deviation in sending power Heterogeneity
sensor nodes
−10 dBm, −10 dBm, 10 dBm 0 dBm LOW
−5 dBm, −10 dBm, −15 dBm 5 dBm MEDIUM
0 dBm, −10 dBm, −20 dBm 10 dBm HIGH

123
Localization of isotropic and anisotropic wireless sensor networks... 319

In this algorithm, the nodes with a higher number of imme- 4. Chen, S., Zhang, J., Mao, Y., Xu, C., & Gu, Y. (2019). Efficient
diate neighbors are localized by identifying and reducing the distributed method for NLOS cooperative localization in WSNs.
Sensors, 19(5), 1173. [Link]
range of probable location of unknown nodes. The nodes with 5. Kaur, A., Kumar, P., & Gupta, G. P. (2019). A weighted centroid
a lower number of neighbors are localized using farther refer- localization algorithm for randomly deployed wireless sensor net-
ence nodes from the least squares method. The results show works. Journal of King Saud University-Computer and Information
improved localization accuracies when tested in various 2D Sciences, 31(1), 82–91. [Link]
007.
and 3D fields of different shapes and heterogeneity. The algo- 6. Musa, A., Gonzalez, V., & Barragan, D. (2019). A new strategy
rithm has shown an improvement of 29% in 3D fields and to optimize the sensors placement in wireless sensor networks.
27% in 2D fields at 10% reference node density and 35% in Journal of Ambient Intelligence and Humanized Computing, 10(4),
3D fields and 12% in 2D fields at 20% reference node density 1389–1399. [Link]
7. Lv, Y., Liu, W., Wang, Z., & Zhang, Z. (2020). WSN local-
when compared with other localization algorithms. As future
ization technology based on hybrid GA-PSO-BP algorithm for
work, we plan to implement this algorithm on a real WSN indoor three-dimensional space. Wireless Personal Communi-
to solve the localization problem. Also, we plan to study the cations, 114(1), 167–184. [Link]
effect of nonuniform node distribution on localization and 07357-4.
8. Phoemphon, S., So-In, C., & Leelathakul, N. (2020). A hybrid
the challenges of improving localization accuracy in those localization model using node segmentation and improved particle
scenarios. swarm optimization with obstacle-awareness for wireless sensor
networks. Expert Systems with Applications, 143, 113044. https://
Funding Open access funding provided by Manipal Academy of Higher [Link]/10.1016/[Link].2019.113044.
Education, Manipal 9. Chai, Q. W., Chu, S. C., Pan, J. S., Hu, P., & Zheng, W. M. (2020).
A parallel WOA with two communication strategies applied in
DV-Hop localization method. EURASIP Journal on Wireless Com-
Declarations munications and Networking, 2020(1), 1–10. [Link]
1186/s13638-020-01663-y.
10. Rai, S., & Varma, S. (2017). Localization in wireless sensor
Conflict of interest The authors declare that they have no conflict of networks using rigid graphs: A review. Wireless Personal Com-
interest. munications, 96(3), 4467–4484. [Link]
017-4397-7.
Open Access This article is licensed under a Creative Commons 11. Cheng, L., Li, Y., Wang, Y., Bi, Y., Feng, L., & Xue, M. (2019). A
Attribution 4.0 International License, which permits use, sharing, adap- triple-filter NLOS localization algorithm based on fuzzy C-means
tation, distribution and reproduction in any medium or format, as for wireless sensor networks. Sensors, 19(5), 1215. [Link]
long as you give appropriate credit to the original author(s) and the 10.3390/s19051215.
source, provide a link to the Creative Commons licence, and indi- 12. Zhang, X., Tepedelenlioglu, C., Banavar, M. K., Spanias, A., &
cate if changes were made. The images or other third party material Muniraju, G. (2019). Location estimation and detection in wireless
in this article are included in the article’s Creative Commons licence, sensor networks in the presence of fading. Physical Communica-
unless indicated otherwise in a credit line to the material. If material tion, 32, 62–74. [Link]
is not included in the article’s Creative Commons licence and your 13. Tan, X., Sun, Z., Wang, P., & Sun, Y. (2020). Environment-aware
intended use is not permitted by statutory regulation or exceeds the localization for wireless sensor networks using magnetic induction.
permitted use, you will need to obtain permission directly from the copy- Ad Hoc Networks, 98, 102030. [Link]
right holder. To view a copy of this licence, visit [Link] 2019.102030.
[Link]/licenses/by/4.0/. 14. Zhu, Y., Xing, S., Zhang, Y., Yan, F., & Shen, L. (2017). Local-
isation algorithm with node selection under power constraint in
software-defined sensor networks. IET Communications, 11(13),
2035–2041. [Link]
15. Yu, X. W., Huang, L. P., Yong, L. I. U., Hao, Y. U., & Ying, L. I.
(2021). Convex localization algorithm based on time difference
of arrival for WSN in uranium tailings radioactive contamina-
References tion. Wireless Personal Communications. [Link]
s11277-020-08055-x.
1. Rathore, M. M., Paul, A., Ahmad, A., & Jeon, G. (2017). IoT- 16. Kumar, S., Kumar, S., & Batra, N. (2021). Optimized distance range
based big data: From smart city towards next generation super city free localization algorithm for WSN. Wireless Personal Communi-
planning. International Journal on Semantic Web and Information cations, 117(3), 1879–1907. [Link]
Systems (IJSWIS), 13(1), 28–47. [Link] 07950-7.
2017010103. 17. Maruthi, S. P., & Panigrahi, T. (2020). Robust mixed source
2. Amri, S., Khelifi, F., Bradai, A., Rachedi, A., Kaddachi, M. L., localization in WSN using swarm intelligence algorithms. Digi-
& Atri, M. (2019). A new fuzzy logic based node localization tal Signal Processing, 98, 102651. [Link]
mechanism for wireless sensor networks. Future Generation Com- 2019.102651.
puter Systems, 93, 799–813. [Link] 18. Yadav, R. K., Verma, S., & Venkatesan, S. (2021). iHRNL: Iterative
10.023. Hessian-based manifold regularization mechanism for localization
3. Messous, S., Liouane, H., & Liouane, N. (2020). Improvement in WSN. The Journal of Supercomputing. [Link]
of DV-Hop localization algorithm for randomly deployed wireless s11227-021-03761-0.
sensor networks. Telecommunication Systems. [Link] 19. Silmi, S., Doukha, Z., & Moussaoui, S. (2021). A self-localization
1007/s11235-019-00592-6. range free protocol for wireless sensor networks. Peer-to-Peer Net-

123
320 S. J. Bhat, K. V. Santhosh

working and Applications. [Link] 36. Jian Yin, L. (2019). A new distance vector-hop localization
01155-w. algorithm based on half-measure weighted centroid. Mobile Infor-
20. Shen, Z., Zhang, T., Tagami, A., & Jin, J. (2021). When RSSI mation Systems. [Link]
encounters deep learning: An area localization scheme for perva- 37. Shi, X., Li, Y., Zhang, S., & Tian, L. (2018). Distance vector hop
sive sensing systems. Journal of Network and Computer Applica- localisation algorithm based on the limitation by the probability to
tions, 173, 102852. [Link] hops. International Journal of Sensor Networks, 27(2), 128–135.
21. Zhang, S., Fan, F., Li, W., Chu, S. C., & Pan, J. S. (2021). A 38. Anand, N., Ranjan, R., & Varma, S. (2017). MSVR based range-
parallel compact sine cosine algorithm for TDOA localization of free localization technique for 3-D sensor networks. Wireless
wireless sensor network. Telecommunication Systems. [Link] Personal Communications, 97(4), 6221–6238. [Link]
org/10.1007/s11235-021-00804-y. 1007/s11277-017-4835-6.
22. Jia, T., Ho, K. C., Wang, H., & Shen, X. (2019). Effect of sensor 39. Zhou, G., He, T., Krishnamurthy, S., & Stankovic, J. A. (2006).
motion on time delay and Doppler shift localization: Analysis and Models and solutions for radio irregularity in wireless sensor
solution. IEEE Transactions on Signal Processing, 67(22), 5881– networks. ACM Transactions on Sensor Networks (TOSN), 2(2),
5895. [Link] 221–262. [Link]
23. Wang, Z., Zhang, B., Wang, X., Jin, X., & Bai, Y. (2018). 40. Sharma, G., & Kumar, A. (2018). Fuzzy logic based 3D localization
Improvements of multihop localization algorithm for wireless sen- in wireless sensor networks using invasive weed and bacterial for-
sor networks. IEEE Systems Journal, 99, 1–12. [Link] aging optimization. Telecommunication Systems, 67(2), 149–162.
1109/JSYST.2018.2851782. [Link]
24. Sheltami, T. R., Shahra, E. Q., & Shakshuki, E. M. (2017). Perfo- 41. Xu, L., Li, Z., & Li, X. (2020). A hybrid approach using multistage
mance comparison of three localization protocols in WSN using collaborative calibration for wireless sensor network localization
Cooja. Journal of Ambient Intelligence and Humanized Comput- in 3D environments. IEEE Access, 8, 130205–130223. [Link]
ing, 8(3), 373–382. [Link] org/10.1109/ACCESS.2020.3009171.
25. Sharma, G., & Kumar, A. (2018). Modified energy-efficient range- 42. Bhat, S. J., & Venkata, S. K. (2020). An optimization based local-
free localization using teaching-learning-based optimization for ization with area minimization for heterogeneous wireless sensor
wireless sensor networks. IETE Journal of Research, 64(1), 124– networks in anisotropic fields. Computer Networks, 179, 107371.
138. [Link] [Link]
26. Ho, K. C., Lu, X., & Kovavisaruch, L. O. (2007). Source local- 43. Tseng, C. L., Liu, F. Y., Lin, C. H., & Lee, C. Y. (2017). Boundary-
ization using TDOA and FDOA measurements in the presence improved distance vector-hop localization method with multipower
of receiver location errors: Analysis and solution. IEEE Trans- correction for wireless sensor networks. Sensors and Materials,
actions on Signal Processing, 55(2), 684–696. [Link] 29(6), 675–687. [Link]
1109/TSP.2006.885744. 44. Qiao, X., Chang, F., & Ling, J. (2019). Improvement of localization
27. Chan, Y. T., & Ho, K. C. (1994). A simple and efficient estimator algorithm for wireless sensor networks based on DV-hop. Interna-
for hyperbolic location. IEEE Transactions on Signal Processing, tional Journal of Online Engineering, 15(6).
42(8), 1905–1915. [Link] 45. Jia, T., Wang, H., Shen, X., Jiang, Z., & He, K. (2018). Target
28. Kumar, S., Kumar, S., & Batra, N. (2020). Optimized distance range localization based on structured total least squares with hybrid
free localization algorithm for WSN. Wireless Personal Communi- TDOA-AOA measurements. Signal Processing, 143, 211–221.
cations. [Link] [Link]
29. Bhat, S. J., & Santhosh, K. V. (2020). Is localization of wireless 46. Shahzad, F., Sheltami, T. R., & Shakshuki, E. M. (2016). Multi-
sensor networks in irregular fields a challenge? Wireless Per- objective optimization for a reliable localization scheme in wireless
sonal Communications, 114, 2017–2042. [Link] sensor networks. Journal of Communications and Networks, 18(5),
s11277-020-07460-6. 796–805. [Link]
30. Fan, J., Hu, Y., Luan, T. H., & Dong, M. (2017). DisLoc: A con- 47. Yan, X., Sun, L., Zhou, J., & Song, A. (2018). DV-hop localisation
vex partitioning based approach for distributed 3-D localization in algorithm based on optimal weighted least square in irregular areas.
wireless sensor networks. IEEE Sensors Journal, 17(24), 8412– Electronics Letters, 54(21), 1243–1245. [Link]
8423. [Link] 2018.6512.
31. Bulusu, N., Heidemann, J., & Estrin, D. (2000). GPS-less low- 48. Zhang, B., Ji, M., & Shan, L. (2012). A weighted centroid local-
cost outdoor localization for very small devices. IEEE Personal ization algorithm based on DV-hop for wireless sensor network.
Communications, 7(5), 28–34. [Link] In Proceedings of the 2012 8th international conference on wire-
32. Niculescu, D., & Nath, B. (2003). DV based positioning in ad hoc less communications, networking and mobile computing (pp. 1-5).
networks. Telecommunication Systems, 22(1–4), 267–280. https:// IEEE. [Link]
[Link]/10.1023/A:1023403323460. 49. Kumar, S., & Lobiyal, D. K. (2013). An advanced DV-Hop
33. He, T., Huang, C., Blum, B. M., Stankovic, J. A., & Abdelzaher, T. localization algorithm for wireless sensor networks. Wireless
(2003). Range-free localization schemes for large scale sensor net- Personal Communications, 71(2), 1365–1385. [Link]
works. In Proceedings of the 9th annual international conference 1007/s11277-012-0880-3.
on Mobile computing and networking (pp. 81-95). ACM. https://
[Link]/10.1145/938985.938995
34. Shang, Y., Ruml, W., Zhang, Y., & Fromherz, M. P. (2003). Local-
Publisher’s Note Springer Nature remains neutral with regard to juris-
ization from mere connectivity. In Proceedings of the 4th ACM
dictional claims in published maps and institutional affiliations.
international symposium on Mobile ad hoc networking and com-
puting (pp. 201–212). [Link]
35. Song, L., Zhao, L., & Ye, J. (2019). DV-hop node location algorithm
based on GSO in wireless sensor networks. Journal of Sensors.
[Link]

123
Localization of isotropic and anisotropic wireless sensor networks... 321

Soumya J. Bhat received Bach- K. V. Santhosh received his bach-

elor of Engineering degree from elors and Masters from Visves-
Manipal Institute of Technology, varaya Technological University,
India and Master of Technology Belgaum in 2004 and 2006. He
from Malnad College of Engi- received his PhD from National
neering, India. She has pursued Institute of Technology, Silchar
internship at Honeywell Technol- in the year 2013. Dr. Santhosh
ogy Solutions Lab Pvt. Ltd. for KV is currently working as Asso-
1 year and worked as R&D engi- ciate Professor, in the Department
neer in Nokia Solutions and Net- of Instrumentation and Control
works India Private Limited for 9 Engineering, Manipal Institute of
years. Currently, she is a research Technology, MAHE. Dr. Santhosh
scholar at the Department of Instru- KV research interest are in the
mentation and Control Engineer- area of sensor fusion, smart sensor,
ing, Manipal Institute of Technol- and fault detection and isolation.
ogy, India. Her research interests include wireless sensor networks,
wireless communication, network planning and localization.

123