Design of a Guideline for Range-based Localization Algorithms

Evaluation using Multiple Linear Regressions

Dhouha El Houssaini

1,2

, Zaid Abdullah

1

, Sabrine Kheriji

2,3

, Kamel Besbes

Microelectronics and Instrumentation Laboratory, Faculty of Sciences of Monastir, 1002, Tunisia

tracking, health monitoring, and military supervision. Analytical modeling and analysis of the localization

system remain challenging and infeasible since it offers oversimplified results with limited reliability to the

evaluated cases. Likewise, disseminating test-beds involves a lot of effort, making the simulation phase indis-

pensable to study the WSN localization. The defined localization model needs to ensure solid and pragmatic

network assumptions during the simulation. However, most network simulators don’t meet specific criteria re-

lated to network definition, such as scalability and heterogeneity. As part of this endeavor, a guideline for eval-

uating and analyzing technical methods of range-based localization is developed. Multiple linear regression is

used to generate the different localization instances, which enables to support different and non-dependent pa-

rameters. The developed guideline for range-based localization is tested and validated for existing localization

solutions.

Localization of these installed nodes is critical and

required for different applications, like object track-

ing, process supervision, and monitoring (El Hous-

saini et al., 2020; Naguib et al., ). Besides, following

the urgent trend to include the Internet of Things (IoT)

concept, the location information remains critical and

necessary for remote control and supervision activi-

ties (Ahmed Mansoor and Irtaza, 2019). The knowl-

edge of the position of installed nodes serves during

the network activities such as the routing, data trans-

decide upon the localization technique for the net-

work assumptions and characteristics, such as the net-

of obstacles, mobile system, etc.), hardware specifi-

cation (sensing and communication range and power

module) and energy constraints. Thus, the realization

of a specific test-bed is extremely expensive and chal-

lenging. Furthermore, the implementation of real ex-

periments always takes more time than the simulation.

Therefore, simulation phase is crucial for the develop-

ment of WSNs.

advantages. NS2 is a discrete event-driven simula-

tor, mainly used for academic research in the areas

of computer networks, MANETs, and WSNs (Hogie

et al., 2006). It is built based on C++ and supports

a wide range of protocols. It also provides complete

support of communication protocol. The main lim-

itation of NS2 is its graphical interface, as it only

provides a simple reflection of the network. It does

not allow the scalability and extension of the net-

work. Furthermore, NS3 was developed based on the

main concepts of NS2 (Nsam, ). It integrates Ne-

tAnim module for the graphical simulation of a net-

work model. Additionally, different simulators in-

tegrate GUI interfaces. One example is OMNET++

(Omnet, ), which is an example of a modular discrete

event simulator written in C++ and provides a power-

ful GUI library for animation and tracing and debug-

ging support. However, their libraries remain limited

Sim has the advantage of supporting many protocols,

including a highly detailed simulation of WSNs plus

Localization features. However, the JSim develop-

ment is closed, and its scale is medium.

This paper aims to develop a generalized guide-

line, which identifies the appropriate localization

technique for the test network. It considers the def-

inition of the deployment environment (Network size

and path loss exponent) and the system assumptions,

such as the sending power and the communication

senting an overview of range-based localization tech-

niques. In section 3, the proposed evaluation guide-

line is provided. Section 4 presents some test and

evaluation results. Finally, a conclusion is provided

in Section 5.

2 RELATED WORKS

The localization techniques can be classified into two

main categories: Range-based and range-free. The

range-based technique uses the distance or range in-

formation to determine the position of a node, such as

tween two nodes (Singh and Sharma, 2015). They use

radio connectivity to communicate among the nodes

to estimate their locations, so no coordinates system

is used. The position of the target node can be ob-

tained by applying geometrics relations. Most range-

free techniques are based on the number of surround-

ing neighbors known as DV-Hop, Approximate Point-

In-Triangulation test (APIT), or centroid system. DV

hop (Hu and Li, 2013; Liu et al., 2016) estimates the

range between nodes using the hop count. Indeed, by

comparing and evaluating the performance and lim-

itations of these methods, it is necessary to find a

trade-off between centralized and distributed topolo-

ficiently simulate and evaluate the localization tech-

nique before the realization and implementation to en-

sure better network performance. Various research

efforts were performed to qualitatively classify and

review the range-based localization techniques. Us-

ing the Web of Science database, research for English

articles including the words ”localization WSN* sur-

vey” or ”Localization WSN* review” or ”localisation

WSN* survey” or ”Localisation WSN* review” (* to

include the plural) in their title, abstracts or keywords.

tion techniques into five categories, but they focus

only on the ranging technique, which is divided into

range-based and range-free methods. An overview of

the AoA, TDoA, and RSSI in the range-based meth-

ods is provided. Four range-free techniques are il-

lustrated: Centroid, APIT, DV-Hop, and Amorphous.

To compare the performance of the described tech-

eral comparative analysis of only the range-based and

range-free methods is discussed in terms of accuracy,

cost, power consumption, and additional hardware re-

quirements. Then, a comparison between RSSI, AoA,

TDoA/ToA is performed by describing their advan-

tages and disadvantages. In (Khelifi et al., 2015),

the advantages and drawbacks of the range-based and

range-free localization technique are described. Au-

thors proved that it is hard to find the best algorithm,

which dominates by all criteria. Authors in (Ismail

ing GPS, range-based methods including RSSI, ToA,

AoA, and TDoA and range-free methods such as cen-

troid, DV-Hop, Amorphous. The discussed reviews

are summarized in Table 1.

of localization accuracy. The system includes five

main blocks: Network initialization, localization in-

stance, data preparation, multiple regression module,

and evaluation (see Figure 1). In the network initial-

ization, parameters related to the network definition

are initialized.

and evaluated, the most suitable localization tech-

nique corresponding to the network initialization is

proposed in terms of localization accuracy and posi-

tion estimation error.

the most suitable localization techniques based on the

input parameters. The regression module permits to

translate the localization choice into a mathematical

model describing the relationship between the depen-

dent and independent variables. More precisely, the

regression module allows mapping the results based

on numeric inputs. The general mathematical module

of multiple linear regression is illustrated in Equation

1.

y

x

x

for each explanatory variable and is the residual value.

The input parameters of the regression module are

the localization instances of each localization tech-

nique. Distances between nodes are estimated based

on each range-based technique. Once the distances

are estimated, a simple trilateration method is carried

out to estimate the coordinates of nodes. 10000 sam-

ples for each localization technique are created based

on the experiment. 80 % of the total number of sam-

ples is defined to train the regression module to iden-

tify each localization technique. The remaining 20 %

of samples are used later to evaluate and test the re-

gression module.

4 TEST AND EVALUATION

chine learning and mathematical models. A user-

friendly, easy-to-use, and flexible guideline is pro-

posed in this work. It allows the user to define its

network parameters, like the network dimensions, the

path loss exponent, and sensor nodes characteristics.

Figure 2 illustrates the designed interface. The inter-

face allows the user to define all network assumptions

that characterize the test field. As output and based

on the localization aspect, the decision of the suitable

localization technique is made based on the accuracy.

calization error, and inaccuracy. Indeed, the proposed

system offers different graphical results to understand

the comparison results better. The first graphical com-

parison is a histogram modeling, as presented in Fig-

ure 3. The illustrated results are for the localiza-

tion error over various possible nodes collections. It

presents a comparative analysis of the nodes deployed

in the system. This is an accuracy graph, which shows

the cumulative accuracy of the system. The x-axis

shows the average Localization by plotting the error

(a)

(b)

Figure 3: Error distribution of range-based localization

technique for a network size of 15 × 20 m

2

, communica-

tion range R

In this work, the influence of the network defini-

tion is studied. In fact, during the localization phase,

nodes are placed in a specific network with specific

hardware characterization. Each parameter may di-

scenarios. It proves that the choice of the suitable

localization technique is strongly dependent on the

network definition. For example, for large-scale net-

works, the outperforms other localization techniques

(Figure 4a ), whereas the works better for small-scale

networks. The path exponent presents the effect of the

network’s scale impact, as it defines the environmen-

tal characteristic. It helps to identify if there exists

some interference and shadowing in the network and

if the signal is propagating in the line of sight. Figure

better results. In the last scenario, the communica-

tion range of the instated nodes is studied. The com-

munication range enables deciding if the network is

covered or not, which helps reduce energy consump-

tion and maintain communication between installed

nodes. If the communication range increases, the lo-

calization capabilities increase, which helps reduce

the measurement error.

Figure 4: Performance evaluation of range-based localiza-

tion techniques: (a) Influence of network size, (b) influence

of path loss exponent and (c) influence of communication

range.

Finally, the proposed guideline is validated using

scenarios validate our guideline to study the effects

based on their achieved localization accuracy. Be-

sides, the selected works implements some artificial

intelligence technique for the determination of the

nodes’ location. We selected different works based

on RSSI techniques only since its strongly dependent

of the deployment environment and sending power of

the nodes. As illustrated in Table 2, obtained results

remain compatible with those reported by selected lo-

calization algorithms in their papers. In (Weingartner

et al., 2009) authors propose an RSSI-based localiza-

information is used to estimate the position of nodes.

The estimated localization accuracy is around 75 %,

compared with the accuracy of the guideline of 93 %.

A preliminary study on the evaluation of range-

based localization techniques is proposed in this

work. An evaluation guideline is presented. It en-

ables to build a better knowledge of the performance

of the localization techniques with the definition of

the network assumptions, mainly, the network size,

communication range, and sending power. This pre-

pared the guideline results with the results of the se-

lected localization algorithms, it was found that they

had been consistent with the original results of these

algorithms in their original papers. In the future, other

localization algorithms can be incorporated into the

guideline, which can be used by other researchers.

For better assumptions considerations, it is neces-

sary to include the energetic module and deployment

JSim. Jsim home page. https://www.physiome.org/jsim/.

sensor networks: A survey on algorithms, measure-

ment techniques, applications and challenges. Journal

of sensor and actuator networks, 6(4):24.

Rahman, M. S., Park, Y., and Kim, K.-D. (2012). Rss-

based indoor localization algorithm for wireless sen-

sor network using generalized regression neural net-

work. Arabian journal for science and engineering,

37(4):1043–1053.

Saad, E., Elhosseini, M., and Haikal, A. Y. (2018). Re-

cent achievements in sensor localization algorithms.

Alexandria engineering journal, 57(4):4219–4228.

Wang, Y., Xu, X., and Tao, X. (2009). Localization in wire-

less sensor networks via support vector regression. In

2009 Third International Conference on Genetic and

Evolutionary Computing, pages 549–552. IEEE.

Weingartner, E., Vom Lehn, H., and Wehrle, K. (2009).

A performance comparison of recent network simula-

tors. In 2009 IEEE International Conference on Com-

munications, pages 1–5. IEEE.