electronics

Article
Assessment of Anchors Constellation Features in
RSSI-Based Indoor Positioning Systems for
Smart Environments
Alessandro Cidronali , Giovanni Collodi * , Matteo Lucarelli , Stefano Maddio ,
Marco Passafiume and Giuseppe Pelosi
Department of Information Engineering (DINFO), University of Florence, 50139 Florence, Italy;
alessandro.cidronali@unifi.it (A.C.); matteo.lucarelli@unifi.it (M.L.); stefano.maddio@unifi.it (S.M.);
marco.passafiume@unifi.it (M.P.); giuseppe.pelosi@unifi.it (G.P.)
* Correspondence: giovanni.collodi@unifi.it; Tel.: +39-055-2758544




Received: 8 May 2020; Accepted: 13 June 2020; Published: 21 June 2020


Abstract: In this paper, we assess the features of a rectangular constellation of four anchors on
the position estimation accuracy of a mobile tag, operating under the IEEE 802.15.4 specifications.
Each anchor implements a smart antenna with eight switched beams, which is capable to collect
Received Signal Strength Indicator (RSSI) data, exploited to estimate the mobile tag position within a
room. We also aim at suggesting a deployment criterion, providing the discussion of the best trade-off
between system complexity and positioning accuracy. The assessment validation was conducted
experimentally by implementing anchor constellations with different mesh sizes in the same room.
Mean accuracies spanning from 0.32 m to 0.7 m on a whole 7.5 m × 6 m room were found by varying
the mesh area from 1.19 m2 to 17 m2 , respectively.

Keywords: wireless positioning; smart environments; distributed networks; collaborative systems;

practical deployment; wireless communications; Internet-of-Things

1. Introduction
It is widely recognized that user localization in Global Navigation Satellite Systems (GNSS)
denied environments represents an essential feature for the capillary diffusion of the paradigm [1–3].
This concept, which has recently been extended and referred to as the “Internet of Things” (IoT), consists
of a complex system where physical objects are provided with sensors and actuators, as well as network
connectivity. This paradigm relies both on the devices low-energy demand, and their ability to reach an
information network where they can contribute to distributed processes [4–7]. These elements typically
communicate information about their status as well as the state of the surrounding environment.
In many cases it is required for the transmitted data to be associated with the context in which the mobile
agent operates to improve the information effectiveness, in particular to be linked to its localization [8].
The position needs to be seamlessly available even in those environments with no access to GNSS.
The need for localization, as well as tracking, becomes relevant not only from the network management
point of view but also to improve the capability of Wireless Sensor Network (WSN) systems in a smart
home environment [9,10]. Nevertheless, enabling or creating a localization system with an IoT network
is not a trivial task. The reason is that low cost, compactness and long lasting autonomy are mandatory
for IoT devices to enable the most widespread diffusion. This fact results in objects with very limited
hardware components, and very stringent specifications for elements such as CPU, memory and
battery, [11]. It is commonly agreed that this problem is not yet solved. In fact, a technology (or a
combination of technologies) does not seem to be available that is capable of recreating in indoor
environments the experience that GNSS offers for outdoor spaces.

Electronics 2020, 9, 1026; doi:10.3390/electronics9061026 www.mdpi.com/journal/electronics

Electronics 2020, 9, 1026 2 of 18

In this work, the problem is addressed by following the paradigm of increasing the amount of
information exchanged by a number of wireless nodes, while assuming a limited use of additional
ad-hoc hardware to that already implemented in the established wireless infrastructure. The reason is
that positioning should be offered as an application layer on top of the data networking layers, with a
minimal impact on the already existing networks.
In this context, several positioning approaches relying on wireless infrastructures based on IEEE
802.11x/802.15.4, have been presented in recent literature [12,13]. The adoption of commercial radio
transceivers makes the use of the Received Signal Strength Indicator (RSSI) a common approach.
Typically, an RSSI enables fingerprinting, which has its main drawback in its requirement of a periodical
and lengthy on-site calibration phase, to overcome environment characterization uncertainties.
Alternatively, a dense distribution of specific beacons can be used. This type of solutions is not feasible
in IoT environments, especially in typical smart environment applications, which include integrations
of positioning systems and autonomous machines. For instance, in smart home environments a robot
can use data from a positioning system to improve the context awareness of interacting objects, and not
only to move into the specific scenario [14]. This requires a localization system characterized by low
cost, low power, ease-of-installation, and ease-of-maintenance.
In [15] and [16], the authors proposed an indoor localization system based on a mesh network of
IEEE 802.15.4 compliant anchor nodes. In those works, as well as hereinafter, the mesh is associated to
the location of the positioning system anchors, which define the mesh vertexes. The technology is
based on the installation of a set of switched anchor nodes equipped with Switched Beam Antennas
(SBA) [15–17], each one capable of IEEE 802.15.4 communications. After having installed a proper
number of anchors working together as a “distributed” router, the network is able to offer wireless
connectivity to its nodes, while collecting a long set of distributed RSSI values. The various received
signal power levels obtained with this procedure are then exploited using proper algorithms to obtain
a position estimation, based only on received signals magnitude (i.e., phase-less). A high level of
accuracy is normally achieved after an on-site calibration phase (as opposed to [16]), or additional
tracking based on odometric readings (as opposed to [13,18]). Using only RSSI (as defined in the
IEEE 802.11x and 802.15.4 standards) readings as input data for positioning algorithms, this method is
suitable to any IEEE compliant wireless communication protocol, making its implementation in any
pre-existent wireless network infrastructure easily achievable (i.e., [19,20]).
In this paper we further extend the concept already presented in [15,16], by introducing and
validating a comprehensive assessment about the trade-off between system positioning complexity
and positioning accuracy in a typical office room site.
The paper is organized as it follows. In Section 2 we review the architecture of the positioning
system, including that of the mobile tag. The distributed positioning technique is reviewed in Section 3,
which includes also the analysis of the anchor constellation shape and size. The results of this analysis
are discussed in details in Section 4, also comparing simulation and experimental data related to a real
test site. A discussion about the outcomes of the work concludes the paper.

2. Localization System Architecture

2.1. Front-End Architecture

As shown in [16], the localization system is based on a constellation of anchors capable
of exchanging communication data with the target node, while simultaneously collecting the
corresponding set of RSSI.
A block diagram of the anchor system architecture is given in Figure 1 [15]. It is based on
an 8-channel SBA, a SP-8T microwave switch, and a System-on-Chip (SoC), which implements a
radio-frequency transceiver and a microcontroller unit (MCU). The hardware scheme proposed in
Figure 1 is deliberately simplified to highlight the ease of implementation of the anchor architecture.
The vast majority of Commercial Off The Shelf (COTS) SoCs are suitable to provide the required features
Electronics 2020, 9, 1026 3 of 18
Electronics 2020, 9, x FOR PEER REVIEW 3 of 18

for
thisthis applications.
applications. TheThe device
device chosen
chosen forfor
thisthis
workwork is the
is the Texas
Texas Instruments
Instruments CC2530,
CC2530, which
which includesa
includes
a2.45
2.45GHz
GHztransceiver
transceiversupporting
supportingthe theIEEE
IEEE802.15.4
802.15.4protocols
protocols and and integrates
integrates aa 80518051 MCU
MCU with with 8-kB
8-kB of
of
RAM. The additional feature required to the MCU consists of the calculation
RAM. The additional feature required to the MCU consists of the calculation and the transmission of and the transmission
of
thethe Received
Received Signal
Signal Strength
Strength Indicator
Indicator (RSSI)
(RSSI) to to
thethe localization
localization engine.
engine. In In this
this project
project thethe RSSI
RSSI is
is transmitted by the Ethernet interface. The MCU provides a set of
transmitted by the Ethernet interface. The MCU provides a set of GPIOs, used also to control the GPIOs, used also to control the
single-pole
single-pole eight-trough
eight-trough (SP-8T)
(SP-8T) RF RF switch
switch that
that sequentially
sequentially routes routes the
the transmitted
transmitted or or received
received signal
signal
to each of the antenna elements of the SBA. In this project, we adopted
to each of the antenna elements of the SBA. In this project, we adopted a HMC321, which a HMC321, which is a is
low-loss
a low-
SP-8T non-reflective
loss SP-8T switchswitch
non-reflective capable to work
capable to in the DC-8
work in theGHz DC-8 bandwidth
GHz bandwidth with lesswiththanless
2.5 than
dB losses at
2.5 dB
the operative frequency. This bandwidth is much larger than what
losses at the operative frequency. This bandwidth is much larger than what is required for this is required for this application;
this is because
application; this
this is architecture
because thisisarchitecture
extremely flexible and modular,
is extremely flexible andand modular,
the switchand boardthehas beenboard
switch used
also for other
has been usedexperiments
also for otherat experiments
different bandwidths.
at different The firmware provides
bandwidths. The firmwarethe proper synchronization
provides the proper
between the RSSI data collection and the switch settling time, which
synchronization between the RSSI data collection and the switch settling time, which is 150 is 150 ns, thus granting a bijective
ns, thus
correspondence between the receiving antenna and the obtained
granting a bijective correspondence between the receiving antenna and the obtained RSSI by RSSI by avoiding the spreading of
aavoiding
single packet through multiple
the spreading of a single antennas. An RSSImultiple
packet through collection cycle is completed
antennas. when a network
An RSSI collection cycle is
packet is exchanged with the mobile node for each of the 8
completed when a network packet is exchanged with the mobile node for each of the antennas of a single SBA. In8the case ofof
antennas thea
IEEE
single802.15.4
SBA. In radio access,
the case of thethe data
IEEE transfer
802.15.4 rate access,
radio is about the250datakbit/s with rate
transfer a packet length
is about 250 of 60 bytes.
kbit/s with
The
a packet length of 60 bytes. The entire set of RSSI data for each anchor is collected in about 13.3toms.
entire set of RSSI data for each anchor is collected in about 13.3 ms. An additional time has be
added to this value for the position algorithm execution, which is around
An additional time has to be added to this value for the position algorithm execution, which is around 10 ms depending on the
chosen algorithm on
10 ms depending andthe
thechosen
poweralgorithm
of the machineand the onpower
which of it is
theperformed.
machine on which it is performed.

Figure 1.
Figure 1. Anchor
Anchor system
system architecture.
architecture.

An entire RSSI data set is collected at the receiving receiving of 8 packets by each anchor anchor node.
node. The anchor
network is able to optimize the packet flow by synchronizing synchronizing packet readings between different
nodes, thus the mobile nodes have to transmit only 8 packets. Referring to CC2530 transceiver
anchor nodes,
datasheet, when powered at 3 V supply voltage, the current absorption during active transmission is
about 2929 mA
mAcorresponding
correspondingtotoa apeak peakpower
powerconsumption
consumption of of
87 87
mW,mW, while
while during
during radio
radio receiving
receiving the
the current
current absorption
absorption is 24ismA
24 mA
(with (with a peak
a peak powerpower consumption
consumption of 72ofmW).
72 mW).
According to
According to the
thegeneral
generalneedneedabout
aboutwhich,
which,mobile
mobilenodes
nodestransmission
transmission profile
profile must
must bebe tailored
tailored to
to maximize
maximize battery
battery life,life,
theythey
sendsend
theirtheir data
data as as a of
a burst burst of 8 repetitions
8 repetitions of data packets,
of data packets, only
only at the at the
required
required localization
localization refresh the
refresh according according the radio
radio channel channel
is sensed is free.
to be sensed to be free.
Following this,Following this,power
their average their
average powerdecreases,
consumption consumption whiledecreases, while the
the localization localization
refresh refresh time
time increases. As increases.
an example, As considering
an example,
aconsidering
burst of 8 apackets,
burst ofequal8 packets,
to 480 equal
bytesto or
48014bytes or 14 milliseconds
milliseconds of transmission
of transmission time, a time
time, a refresh refresh
of
1time of 1 leads
second second to leads to an average
an average power consumption
power consumption of 1.2 mW. of 1.2 mW.
Because the
Because theanchor
anchornodes
nodes areare
partpart of fixed
of fixed network
network infrastructure,
infrastructure, to dealtowith
deal with
user userrandom
nodes nodes
random
packet packet transmissions,
transmissions, they must be they mustready
always be always ready
to receive to so
data, receive
they aredata, so they
always are always
in receiver enabledin
receiver
state. By enabled
this, theirstate. By this,
average power their average power
consumption consumption
exactly equals peak exactly equals peak
consumption of 72 consumption
mW.
of 72 mW.
 Electronics2020,
Electronics 2020,9,9,x1026
FOR PEER REVIEW 44of
of18
18

2.2. Switched Beam Antenna Architecture

2.2. Switched Beam Antenna Architecture
The key part of the anchor system architecture is the SBA, which is based on an array of seven
The key part of the anchor system architecture is the SBA, which is based on an array of seven
antennas, arranged with one hexagon at the SBA center and six perimetric pentagons. Figure 2 shows
antennas, arranged with one hexagon at the SBA center and six perimetric pentagons. Figure 2 shows
the mechanical arrangement of the SBA: it is a hemispherical structure with a maximum diameter
the mechanical arrangement of the SBA: it is a hemispherical structure with a maximum diameter
around 146 mm, a height of 53.8 mm and with a dihedral angle of 120 degree.
around 146 mm, a height of 53.8 mm and with a dihedral angle of 120 degree.

(a) (b)

Figure
Figure2.2.The
Themechanical
mechanicalstructure
structureof
ofthe
theSwitched
SwitchedBeam
BeamAntenna;
Antenna;(a):
(a):top
topview,
view,(b):
(b):side
sideview.
view.

Thedihedral
The dihedralangle
angleisisaadesign
designparameter
parameterdepending
dependingupon uponthe theoperative
operativeconditions,
conditions,in inparticular
particular
coveragearea,
coverage area,and
andnoise,
noise,asas discussed
discussed in in [21],
[21], nevertheless
nevertheless its optimum
its optimum value
value is between
is between 100 130
100 and and
130 degree
degree for circular
for circular coverage
coverage areaarea radius
radius up up to to2.52.5m.m.The TheSBA SBAexhibits
exhibitssix sixidentical
identical perimetric
perimetric
antenna,while
antenna, whilethethe
toptop antenna
antenna differentiates
differentiates from from the previous
the previous ones as ones as it admits
it admits two radiation
two radiation modes
modes
and and corresponding
corresponding input ports,input ports,
[22]; [22]; the beams
the effective effectivearebeams
therefore are 8therefore 8 in total.
in total. Each of the Each of the
perimetric
perimetric elements is a Circularly Polarized (CP) antenna designed
elements is a Circularly Polarized (CP) antenna designed by the principle explained in [23], which by the principle explained in [23],
is
which is basically
basically a techniquea technique
capable capable
to generate to generate two propagation
two propagation modesmodes with awith a single
single feedingfeeding
signal,signal,
by
by splitting
splitting it inittwo
in two orthogonal
orthogonal andand properly
properly phasedphased components.
components. This This is obtained
is obtained by removing
by removing the
the conductor in an elliptical region centered within the disk. The
conductor in an elliptical region centered within the disk. The top antenna is instead a dual mode top antenna is instead a dual
mode circularly
circularly polarizedpolarized
antenna antenna
capable capable
to provideto provide a regular
a regular cardioid
cardioid patternpattern
withwiththe the maximum
maximum of
of radiation
radiation in the
in the boresight
boresight exploiting
exploiting its its first
first input
input pin,
pin, alongwith
along witha asecond
secondpatchpatchthatthatprovides
providesaa
toroidalpatter
toroidal patterwith
with the
the maximum
maximumradiation radiationat at45 45 degree
degree from from thethe antenna
antenna axis axis obtained
obtained using
using its its
secondpin.
second pin.
Theadoption
The adoptionofofcircular
circularpolarization
polarization is is motivated
motivated byby two twomainmain reasons:
reasons: the the
firstfirst consists
consists in the in
the need
need to reduce
to reduce anyany signal
signal fading
fading due due to to cross-polarization
cross-polarization comingfrom
coming fromaageneric
generictarget–anchor
target–anchor
orientation. The
orientation. The second one one consists
consistsininthe theinherent
inherentrobustness
robustness of of
this propagation
this propagation modemodeto multi-path
to multi-
phenomena.
path phenomena. The SBA patchpatch
The SBA antennas
antennasare capable
are capable of operating
of operating in thein2.4–2.48
the 2.4–2.48GHzGHz band.band.
InFigure
In Figure33the thepicture
pictureof of the
theprototype
prototype adoptedadopted in in this
this work
work is is reported.
reported. The Thepictures
picturesshowshowthe the
SBA assembly (Figure
SBA (Figure3a),3a),with
with thethe
toptopantenna
antenna andand 3 out3 ofouttheof6 perimetric
the 6 perimetricantennas. The topThe
antennas. antennatop
implements
antenna a number
implements of parasitic
a number elementselements
of parasitic that improve the gain the
that improve of antenna
gain of element
antenna #8, the one
element #8,with
the
a toroidal
one with apattern,
toroidalwhich
pattern,results
which particularly affected byaffected
results particularly the global byarrangement. The rear view
the global arrangement. Thevisible
rear
in Figure
view 3b shows
visible the 3b
in Figure control
shows unittheof control
the SBAunit and theof theotherSBA parts
andofthe the other
anchorpartsarchitecture shown
of the anchor
in Figure 1. shown
architecture This specific
in FigureSBA1.is This
implemented
specific SBA in each anchor nodein
is implemented byeachusing a multiplexer
anchor node by properly
using a
synchronized
multiplexer with the
properly RSSI block; this
synchronized withgrants
the RSSI that the RSSI
block; acquisition
this grants that the is related to the specific
RSSI acquisition array
is related
element
to and thus
the specific to aelement
array specificand spacethussector.
to a Thespecificmultiplexing
space sector. between the antennas operates
The multiplexing between in thea
synchronized
antennas mode
operates in for all the anchors
a synchronized modebelonging
for all theto the same constellation
anchors belonging to to theavoid
samesignal clashes.to
constellation
avoid signal clashes.
The radiation pattern of the SBA is shown in Figure 4, in the azimuthal plane (ϕ) and elevation
(θ). In particular, Figure 4a shows the measured radiation patterns for all the 8 antennas at elevation
angle θ = 90 degree. This is not the boresight direction for antennas 2–7, thus they exhibit maxima
contains the dihedron formed by these two antenna.
The global pattern results from the sequence of each cardioid pattern, and provides Spatial
Division Multiplexing
Electronics 2020, Access
9, x FOR PEER (SDMA). A single anchor node is hence capable of performing signal
REVIEW 5 of 18
direction-of-arrival estimation, [23], and is inherently suitable for an integration in more complex
anchors
Electronics
lower than constellations
9, 1026
2020,their distributed
absolute maximum across anFrom
gain. entirethe
area. 5 of 18
graphs it is possible to see that the respective
direction of maximum radiation is uniformly spread across the radial direction. In this picture,
antennas 1 and 8 are observed by their horizontal angle of view, far from their maximum, and thus
do not provide any pattern variation. Figure 4b reports the corresponding radiation patterns for each
antenna in the elevation plane; in this case all the antennas are observed at different angle of view
and thus provide gain variation. In particular, the complementarity of the antenna 1 and 8 gain
patterns is clearly observable, thus providing an additional piece of information to the localization
engine. Antenna 8 exhibits a lower gain with respect to the others because it is affected by the
presence of antenna 1; the outer parasitic elements partly compensate this effect. Because the specific
selection of the θ angle, in this graph the radiation patterns are very similar as the elevation plane
contains the dihedron formed by these two antenna.
The global pattern results from the sequence of each cardioid pattern, and provides Spatial
Division Multiplexing Access (SDMA). A single anchor node is hence capable of performing signal
direction-of-arrival estimation, [23], and is inherently suitable for an integration in more complex
anchors constellations distributed across an entire area.
(a) (b)

Switched Beam
Figure 3. Switched Beam Antenna
Antenna prototype:
prototype: (a)
(a) upper
upper view;
view; the
the top
top antenna
antenna in this realization
presents some parasitic elements to improve the radiation gain of antenna #8;
#8; (b) rear view, with the
control unit board in view.
view.

The radiation pattern of the SBA is shown in Figure 4, in the azimuthal plane (φ) and elevation (θ).
In particular, Figure 4a shows the measured radiation patterns for all the 8 antennas at elevation angle
θ = 90 degree. This is not the boresight direction for antennas 2–7, thus they exhibit maxima lower
than their absolute maximum gain. From the graphs it is possible to see that the respective direction of
maximum radiation is uniformly spread across the radial direction. In this picture, antennas 1 and
8 are observed by their horizontal angle of view, far from their maximum, and thus do not provide
any pattern variation. Figure 4b reports the corresponding radiation patterns for each antenna in
the elevation plane; in this case all the antennas are observed at different angle of view and thus
provide gain variation. In particular, the complementarity of the antenna 1 and 8 gain patterns
is clearly observable, thus
(a) providing an additional piece of information(b) to the localization engine.
Antenna 8 exhibits a lower gain with respect to the others because it is affected by the presence of
Figure 3. Switched Beam Antenna prototype: (a) upper view; the top antenna in this realization
antenna 1; the outer parasitic elements partly compensate this effect. Because the specific selection of
presents some parasitic elements to improve the radiation gain of antenna #8; (b) rear view, with the
the θ angle, in this graph the radiation patterns are very similar as the elevation plane contains the
control unit board in view.
dihedron formed by these two antenna.

(a) (b)

Figure 4. Switched Beam Array: (a) measured pattern in azimuthal plan; (b) pattern at ϕ= 70° (side
pointing perpendicular direction).

(a) (b)

Figure 4. Switched
SwitchedBeam
BeamArray:
Array:(a) (a)
measured pattern
measured in azimuthal
pattern plan;plan;
in azimuthal (b) pattern at ϕ=
(b) pattern φ =(side
at 70° 70◦
pointing
(side perpendicular
pointing direction).
perpendicular direction).
Electronics 2020, 9, 1026 6 of 18

The global pattern results from the sequence of each cardioid pattern, and provides Spatial
Division Multiplexing Access (SDMA). A single anchor node is hence capable of performing signal
direction-of-arrival
Electronics 2020, 9, x FORestimation,
PEER REVIEW [23], and is inherently suitable for an integration in more complex 6 of 18
anchors constellations distributed across an entire area.
2.3. Mobile Tag Architecture
2.3. Mobile Tag Architecture
The mobile tag subject to the positioning shares its architecture with the anchor node with the
exceptionmobile
The tag subject
of the ethernet to the positioning
connection, the SBA and shares its architecture
its associated switch.with
Thethe anchor
power node
supply with the
is provided
exception of the ethernet connection, the SBA and its associated switch. The
by two 1.5 V AA batteries. In this work, the SoC is connected with a planar CP patch antenna, as power supply is provided
by two 1.5
shown V AA batteries.
in Figure In this work,
5a. The radiation patternthe is
SoC is connected
provided with5b.
in Figure a planar CP patchpattern
The radiation antenna, of as
theshown
target
in Figure 5a. The radiation pattern is provided in Figure 5b. The radiation
node is characterized by a main radiation lobe directed along the antenna axis and exhibits pattern of the target nodea
issufficiently
characterized by a main radiation lobe directed along the antenna axis and exhibits
regular behavior around it, [20]. This shape resembles that of a typical Planar Inverted F- a sufficiently
regular
Antenna (PIFA), around
behavior it, [20]. This
as implemented shape mobile
in generic resembles that of [24].
terminals, a typical Planar
In fact, PIFAInverted
antennas F-Antenna
cannot be
(PIFA), as implemented in generic mobile terminals, [24]. In fact, PIFA
considered to have “pure” radiation patterns, as these devices always show an upward oriented antennas cannot be considerednon-
to have “pure”
isotropic pattern radiation patterns, asathese
when considering typicaldevices
handset always showInanaddition,
use case. upward theoriented
PIFA non-isotropic
is not suitable
pattern
for CP when
whichconsidering a typical handset
is instead preferable for this use case. In addition,
application to face thethe PIFAmultipath
severe is not suitable forthat
effects CP which
can be
ispresent
insteadinpreferable for this application to face the severe multipath effects
many realistic scenarios. More elaborated mobile tag antenna designs can be taken into that can be present in
many
account,realistic scenarios. More
[25], nevertheless theyelaborated mobile tag
require a complex antenna
feeding designs
network andcan
inbe taken
many into account,result
circumstances [25],
nevertheless they require a complex feeding network and in many circumstances
not compatible with the applicative scenario. For instance, this constraint is undesirable for a compact result not compatible
with the applicative
and low cost device scenario.
such as the Forproposed
instance,tag thisnode
constraint
and have is undesirable
therefore not forbeen
a compact
used inand thislow
work. cost
device such as the proposed tag node and have therefore not been used in this work.

(a) (b)

Figure5.5.Experimental
Figure Experimentaltarget
targetnode
node(a)
(a)and
andits
itsradiation
radiationpattern
pattern(b).
(b).

3.3. Distributed
Distributed Positioning
Positioning
3.1. The Position Estimation
3.1. The Position Estimation
The numerical estimation of the mobile tag position is based on signal magnitude level
The numerical estimation of the mobile tag position is based on signal magnitude level
measurements, and is generally pursued by minimizing an objective function. Typical objective
measurements, and is generally pursued by minimizing an objective function. Typical objective
functions are expressed by sums of squared residuals, likelihood functions, posterior density functions,
functions are expressed by sums of squared residuals, likelihood functions, posterior density
risk functions, as well as robust loss functions, with the two most common being the Least Squares (LS)
functions, risk functions, as well as robust loss functions, with the two most common being the Least
and the Maximum Likelihood (ML) functions [1].
Squares (LS) and the Maximum Likelihood (ML) functions [1].
In this paper, we focus on an algorithm based on a derivation of the ML method, whose estimation
In this paper, we focus on an algorithm based on a derivation of the ML method, whose
only requires the assumption of a first order knowledge of the range measurement probability
estimation only requires the assumption of a first order knowledge of the range measurement
distribution function. It is based on the minimization of the calculated variance for both measured and
probability distribution function. It is based on the minimization of the calculated variance for both
expected data from an unknown position, [16]. Its operating principle assumes the acquisition of a
measured and expected data from an unknown position, [16]. Its operating principle assumes the
acquisition of a coherent set of signal levels during network communication between the mobile tag
and a generic number, N, of available anchors. The collected data consist of a steering vector,
containing all the RSSI values obtained from each antenna element of each anchor composing the
constellation; the steering vector is defined as:
Electronics 2020, 9, 1026 7 of 18

coherent set of signal levels during network communication between the mobile tag and a generic
number, N, of available anchors. The collected data consist of a steering vector, containing all the RSSI
values obtained from each antenna element of each anchor composing the constellation; the steering
vector is defined as:  ´ 
 S1 
 ´ 
 S2 
Ś =  . , (1)
 
 .. 
 
 ´ 
SN

where Śi denotes the steering vector of the i-th anchor, with:
 
 s1i 
 s2i
 

Śi =  . , (2)
 
 .. 
 
sMi
 

and where ski is the RSSI collected by the k-th antenna of the i-th anchor, with M the number of
SBA antenna elements. The positioning technique adopted in this work is hereinafter descripted,
considering the i-th steering vector composed by the RSSI data obtained from the M antenna elements
composing the single anchor node. The position estimation relies on the assumption of a known mobile
tag height zTAG (z-axis), used to define the steering vector reference map, M(x,y). The latter is obtained
as the projection of the SBA patterns throughout the (x,y) localization domain calculated by a first-order
propagation model, [16], between a generic tag  D position
 (x,y) and the anchor known position. This is
i
characterized by the path-loss model 20 log λ0 , with Di the distance between the mobile tag at (x,y)
and the i-th anchor node, and λ0 the free-space wave length at the operative frequency. Furthermore,
it is assumed that between the anchors and the mobile tag persists a line-of sight. Under these realistic
hypothesis, the tag position estimation (x̂, ŷ) is obtained by minimizing the variance of the difference
between the steering vector and the expected signal map vector across the entire analysis domain:

(x̂, ŷ) = argmin(x,y) C(x, y), (3)

with n o
C(x, y) = vari Ś − M(x, y) . (4)

In Equation (4), the variance of the difference between the steering vector and the expected signal map
for the actual position is given by:
 2
MN MN
1 X 1 X 
C(x, y) = (si − mi ) − s j − m j  (5)
MN  MN 
i=1 j=1

As demonstrated in [16], the robustness of the estimator in Equation (5) is able to counterbalance
the approximation introduced by the map model and the large variability of possible user node heights.

3.2. Comparison between Triangular and Squared Shapes and Parameters

An a priori analysis of the anchor constellation shapes suitable for a proper coverage of the
positioning domain can be effectively obtained by the results of the information theory. Extracting
the inverse of the measurement covariance matrix, F, as the weight matrix, the variance leads to the
Cramer–Rao lower Bound (CRB) for an unbiased estimator, E. We can write that [26],

var[E] ≥ CRB(x, y) = CRBx (x, y) + CRB y (x, y), (6)

Electronics 2020, 9, 1026 8 of 18

Electronics 2020, 9, x FOR PEER REVIEW 8 of 18

Electronics 2020, 9, x FOR PEER REVIEW 8 of 18
where
PNM h ∂Gn i2 PNM  ∂Gn 2
∂y
CRBx (x, y∑∑ σ∑
n=1 ∂x n=1
) = σ2 ; CRB y (x, y) = ∑2
(7) (7)
CRB (( , , )=σ
)=σ det(F;) CRB ( , ) = σ det(F) (7)
CRB ) ; CRB ( , ) = σ
(( )projection ( )
In Equation (7) Gn represents the on the (x,y) domain (of )the n-th antenna element
radiationIn In Equation
pattern
Equation of the (7)SBA
(7) GGnnrepresents
represents
and σ2 thethe the projection
variance
projectionof the on
on the(x,y)
observed
the (x,y)data.
domain
domain Theofof
CRB therepresents
the n-thantenna
n-th antenna element
theelement
accuracy
limitradiation
radiation pattern
of the positioning
pattern of the of the SBA
estimator
SBA and and
withσ 2 the variance of the observed data. The CRB represents the
respect
σ2 the varianceto itsofseveral
the observedparameters,
data. Thewhich CRB in represents
this work the case
accuracy
are accuracy
the shape limit ofofthe
of the
limit the positioningestimator
constellation
positioning estimator
(i.e., triangularwithrespect
with respect
versus totorectangular),
itsseveral
its severalparameters,
parameters,
their size, which
and in
which inthis
the this work
additive
work
caseare
case arethe
theshape
shapeof ofthe
theconstellation
constellation(i.e.,(i.e.,triangular
triangular2versusversus rectangular),their theirsize,
size,and
andthetheadditive
additive
white Gaussian noise characterized by its variance σ . The rectangular),
Figure 6 illustrates the CRB calculated
whiteGaussian
Gaussiannoise noisecharacterized
characterizedby byitsitsvariance
varianceσσ2. .The2 TheFigureFigure66illustrates
illustratesthe theCRBCRBcalculated
calculated in
in awhite
10 m × 10 m square domain, corresponding to a constellation of three anchors arranged ininan
a 10 m × 10 m square domain, corresponding to a constellation of three anchors arranged ininanan
a 10 m ×triangular
equilateral 10 m square shape,domain, corresponding
with various triangletoedge a constellation
size, and σ of = 1.three
Theanchors
domainarranged
corresponds
 = 1. The domain corresponds to
to
equilateral triangular shape, with various triangle edge size, and σ
σ = is
the equilateral
empty space. triangular
From the shape, withwe
picture various
can see triangle
that the edge size, anderror
minimum 1. expected
The domain corresponds to
in correspondence
theempty
the emptyspace.
space.From Fromthe thepicture
picturewe wecancansee seethat
thatthe theminimum
minimumerror errorisisexpected
expectedinincorrespondence
correspondence
of the
of
anchors
the anchors
andand reducing
reducing
thethe edge
edge
size some
size some
kindkind
ofoffusion
fusion
ofofthe
the
minima
minima
occurs,
occurs,
leading to a
leadingtotoaa
of
unique the anchors
circular and
region reducing
of minimum the edge
around size some
thethe center kind
ofof of
the fusion of the minima occurs, leading
unique
unique circular
circular region
region ofofminimum
minimum around
around center
the center of the thedomain.
domain.From
domain. Fromthe
From
the graphs
thegraphs
graphswe we
wesee see that
see thatthe
that the
the
triangular mesh
triangularmesh can
meshcan return
canreturn an
returnan average
anaverage accuracy
averageaccuracy across
accuracyacross acrossthethe entire
theentire domain
entiredomain
domainwhich which is almost
whichisisalmost
almostflat flat
flatup uptoto7to
up 7
triangular
7 m, m,
while
whilewithin
within the theshape
shape the average
the average CRBCRB increase
increase with
with the
the square
square of
of the
the shape
shape size.
size. In
In Figure
Figure 77we
we
m, while within the shape the average CRB increase with the square of the shape size. In Figure 7 we
report a similar
report analysis considering a square shape constellation.
constellation.In this case we can see
seeaaabehavior
behavior
report aasimilar
similar analysis
analysis considering
considering aasquare
square shape
shape constellation. Inthis
In thiscase
casewe wecancansee behavior
qualitatively similar
qualitativelysimilar to the
similar toto theprevious
the previousone
previous onebut with
one but a
but withmore
withaa moreuniform
more uniform distribution
uniform distributionwithin the
distribution within entire
withinthe thedomain.
entire
qualitatively entire
Figure 8 reports
domain. Figure the8 reports
accuracy theinaccuracy
terms of in CRB
terms for
of bothfor
CRB cases.
both From
cases. this we
From can
this we observe
can observethatthat the
domain. Figure 8 reports the accuracy in terms of CRB for both cases. From this we can observe that
rectangular
the mesh
rectangular leads
mesh to better
leads to average
better values
average for
values the
for accuracy
the accuracy
the rectangular mesh leads to better average values for the accuracy for the ideal mobile tag position for the
for theideal
ideal mobile
mobile tag
tag position
position
estimation,
estimation,
estimation,being being
always
being always
always better
better
betterthanthan
than the the
the correspondingtriangular
corresponding
corresponding triangularmesh
triangular mesharea.
mesh area.Given
area. Giventhis,
Given the
this,the article
thearticle
article
will
willwill
focus focus
on on
focus on
thethethe rectangular
rectangular
rectangular shape,
shape, although
shape,although it
althoughititrequires requires
requires an an additional
an additional anchor
anchor with
additional anchor with respect
withrespect to
respecttotothe the
the
triangular
triangular case.
triangular case.case.

(a) (b) (c)

(a) (b) (c)
Figure 6. Cramer–Rao lower Bound (CRB) evaluation for triangular mesh at different edge sizes: (a) 0.85
Figure6.6. Cramer–Rao
Figure Cramer–Rao lower lowerBound
Bound(CRB) evaluation
(CRB) for triangular
evaluation mesh mesh
for triangular at different edge sizes:
at different edge (a)sizes:
0.85
m; (b) 3.45 m; (c) 6 m.
(a)m; (b)m;
0.85 3.45(b)m;3.45
(c) 6m;
m.(c) 6 m.

(a) (b) (c)

(a) (b) (c)
7. CRB
Figure
Figure evaluation
7. CRB forfor
evaluation rectangular mesh
rectangular at at
mesh different edge
different edgesizes:
sizes:(a)
(a)11m;
m;(b)
(b) 44 m;
m; (c) 7 m.
Figure 7. CRB evaluation for rectangular mesh at different edge sizes: (a) 1 m; (b) 4 m; (c) 7 m.
m.
 Electronics 2020, 9, x FOR PEER REVIEW 9 of 18
Electronics 2020, 9, 1026 9 of 18

(a) (b)

Figure8.8.Calculation
Figure Calculationofofthe
theaverage
averageCRB
CRBforfordifferent
differentmesh
meshshapes
shapesand
andcorresponding
correspondingsizes,
sizes,asasa a
function of (a) the mesh edge size, and (b) the mesh area; σ2= 1.
function of (a) the mesh edge size, and (b) the mesh area; σ = 1.
2

4.4.Impact
Impactofofthe
theConstellation
ConstellationMesh MeshArea
Areaon onLocalization
LocalizationAccuracy
Accuracy
This section analyzes the effects of the mesh area on the localization
This section analyzes the effects of the mesh area on the localization accuracy andand
accuracy error coverage
error of
coverage
the
of proposed system
the proposed within
system a realacase
within realindoor environment.
case indoor On theOn
environment. basis
theofbasis
the previous Section results,
of the previous Section
we focus on a rectangular constellation with 4 identical anchors at its vertexes.
results, we focus on a rectangular constellation with 4 identical anchors at its vertexes. We We experimentally
verify the accuracy
experimentally of the
verify positioning
the accuracy ofbythe
three mesh sizeby
positioning cases. For
three all the
mesh analyzed
size cases,
cases. For the position
all the analyzed
estimation is carried out by the estimator in Equation (3). The analysis compares the experimental
cases, the position estimation is carried out by the estimator in Equation (3). The analysis compares data
with simulations.
the experimental data with simulations.
4.1. Experimental Site
4.1. Experimental Site
The scenario under consideration is a typical small office environment with a total area of 7.5 m by
The scenario under consideration is a typical small office environment with a total area of 7.5 m
6 m, for a total area of 45 m2 , with2the usual furniture. Its map is reported in Figure 9a, and it is possible
by 6 m, for a total area of 45 m , with the usual furniture. Its map is reported in Figure 9a, and it is
to see that it includes furniture, metallic objects and office closets spread around the area. The tests were
possible to see that it includes furniture, metallic objects and office closets spread around the area.
performed while allowing the office employees to continue their habitual tasks. The latter condition
The tests were performed while allowing the office employees to continue their habitual tasks. The
results to affect the variance of the observed data, σ2 , defined in the Equation (7). The picture of the test
latter condition results to affect the variance of the observed data, σ2, defined in the Equation (7). The
site is reported in Figure 9c where we observe the single constellation rectangular mesh installed on the
picture of the test site is reported in Figure 9c where we observe the single constellation rectangular
ceiling, with the three out four anchors visible, and the mobile tag installed on a tripod. This installation
mesh installed on the ceiling, with the three out four anchors visible, and the mobile tag installed on
guaranties a line-of-sight between the anchors and the mobile tag, which is a requirement for a proper
a tripod. This installation guaranties a line-of-sight between the anchors and the mobile tag, which is
positioning estimation.
a requirement for a proper positioning estimation.
The main goal of this work consists on the experimental investigation of the relation between the
The main goal of this work consists on the experimental investigation of the relation between
rectangular mesh size and the position estimation accuracy, thus making the anchor nodes location a
the rectangular mesh size and the position estimation accuracy, thus making the anchor nodes
degree of freedom to design the network. In general, a system architect designer can consider other
location a degree of freedom to design the network. In general, a system architect designer can
degrees of freedom like the number of anchors and their arrangements in different mesh shapes,
consider other degrees of freedom like the number of anchors and their arrangements in different
being the latter having already been considered by an idealized point of view in the previous Section 3.2.
mesh shapes, being the latter having already been considered by an idealized point of view in the
To fully evaluate the effects of the anchor distribution on the localization accuracy, the mobile tag
previous Section 3.2.
position has been estimated on a large number of points spread all over the room; these are visible in
To fully evaluate the effects of the anchor distribution on the localization accuracy, the mobile
Figure 9a. In particular, the grid is composed by 299 points regularly spread throughout the room,
tag position has been estimated on a large number of points spread all over the room; these are visible
wherever it was possible compatibly with the obstacles present in the environment, and spaced by
in Figure 9a. In particular, the grid is composed by 299 points regularly spread throughout the room,
0.5 m × 0.5 m both by row and column, with points on adjacent columns vertically shifted by 0.25 m.
wherever it was possible compatibly with the obstacles present in the environment, and spaced by
The three test anchor constellations are visible in Figure 9b, they are characterized by different edge
0.5 m × 0.5 m both by row and column, with points on adjacent columns vertically shifted by 0.25 m.
thus area sizes, namely a “large”, a “medium” and a “tiny” mesh configurations; Table 1 summarizes
The three test anchor constellations are visible in Figure 9b, they are characterized by different edge
the constellation parameters for each case. Overall anchors were hanged at the ceiling, at height
thus area sizes, namely a “large”, a “medium” and a “tiny” mesh configurations; Table 1 summarizes
Zi = 2.8 m.
 Electronics 2020, 9, x FOR PEER REVIEW 10 of 18

the constellation parameters for each case. Overall anchors were hanged at the ceiling, at height Zi =
Electronics 2020, 9, 1026 10 of 18
2.8 m.

(a) (b)

(c)

Figure
Figure 9. 9.
(a)(a) Bidimensionalmap
Bidimensional mapof
ofthe
the experimental
experimental test
test site,
site,(b)
(b)with
withthe
thesuperimpose
superimposeof of
thethe
three
three
constellations adopted in this work, and (c) the corresponding picture.
constellations adopted in this work, and (c) the corresponding picture.

Table 1. Anchor–mesh configurations.

Table 1. Anchor–mesh configurations.
Configuration
Configuration AnchorPositions
Anchor Positions Mesh
Mesh Area
Area
#1#1(1.50,
(1.50,5.00)
5.00)m;
m;#2
#2(1.50,
(1.50, 1.00)
1.00) m 1717mm22
(A)
(A)Large
Large
#3#3(6.00,
(6.00,5.00)
5.00)m;
m;#4
#4(6.00,
(6.00, 1.00)
1.00) m
m 40% of
40% of room room area
area
#1#1(2.20,
(2.20,3.50)
3.50)m;
m;#2
#2(2.20,
(2.20, 1.20)
1.20) m 7.67.6
mm2 2
(B)Medium
(B) Medium
#3 (5.50, 3.50) m; #4 (5.50, 1.20) m
#3 (5.50, 3.50) m; #4 (5.50, 1.20) 17% of room
17% of room area area
#1#1(2.70,
(2.70,2.65)
2.65)m;
m;#2
#2(2.70,
(2.70, 1.70)
1.70) m
m 1.19 mm2
1.19
2
(C)Tiny
(C) Tiny
#3#3(3.90,
(3.90,2.65)
2.65)m;
m;#4
#4(3.90,
(3.90, 1.70)
1.70) m 2.6% of of
2.6% room
room area
area

4.2.4.2. Simulation
Simulation Results
Results
In In this
this work,the
work, theaccuracy
accuracyof ofthe
the position
position estimator
estimator isisfirstly
firstlyevaluated
evaluatedbyby a set of of
a set simulations.
simulations.
This permits to generate a test bench by which evaluate the results, and in
This permits to generate a test bench by which evaluate the results, and in prospective provideprospective provide a
a model
model for the localization estimator in a given environment. This eventually permits to plan an
for the localization estimator in a given environment. This eventually permits to plan an effective
effective anchor network topology in an optimal and reliable way, avoiding a lengthy on-site
anchor network topology in an optimal and reliable way, avoiding a lengthy on-site verification.
verification.
Given the fact that the selected localization algorithm relies on RSSI measurements as defined for
Given the fact that the selected localization algorithm relies on RSSI measurements as defined
IEEE 802.11x and IEEE802.15.4 wireless network standards, the synthetic localization error based on
for IEEE 802.11x and IEEE802.15.4 wireless network standards, the synthetic localization error based
CRB evaluation provide a first order approximation of the obtainable results, not always reliable yet
on CRB evaluation provide a first order approximation of the obtainable results, not always reliable
necessary the initial
yet necessary stagestage
the initial to define the real
to define distributed
the real positioning
distributed positioninganchors
anchorsnetwork.
network.This
This is
is due
due to
thetoinherent
the inherent formulation of the CRB, cf. Section 3.2, that provides the analytical characterization of of
formulation of the CRB, cf. Section 3.2, that provides the analytical characterization
thethe
localization
localizationerror
errorassuming
assumingthe thesteering
steering vectors composedof
vectors composed ofthe
thereceived
receivedsignals
signalsatat
thethe antenna
antenna
section. In [12], it is discussed how the RSSI can be considered as an output variable of a complex
digital correlation algorithm with a high process gain. Given this, the effective system performance
should be analyzed through an algorithm capable of encompassing every data processing layer, [26].
In particular, the RSSI noise model should be described using its own detection error parameters rather
Electronics 2020, 9, 1026 11 of 18

than the effective physical signal-to-noise ratios: such detection error parameters are defined within
protocol standards, and they depict a less noisy measurement scenario than the straight physical
one [27]. The RSSI noise model can be brought back to an Additive White Gaussian Noise (AWGN)
model, as described in [18]. The simulation process applies the localization estimator described in
Equation (3), by considering the following steering vector in the (x,y) domain:

Ś = hs11 ∨ s11 ∨ . . . ∨ sMN iH = M(x, y) + T́ + Ń, (8)

N
P
where the steering vector Ś length is Mi = MN, while T́ contains the radiation pattern of the mobile
i=1
tag node observed by the N anchors:

T´1   GTAG (ϑ1 , ϕ1 )·IM1 ,1

   
 
T´2   GTAG (ϑ2 , ϕ2 )·IM2 ,1
   
 
T́ =   =  . (9)
   
.. ..

 .  
  . 

´
TN GTAG ϑN , ϕN )·IMN ,1
(


The term GTAG (ϑ1 , ϕ1 ) is the mobile tag antenna gain seen by each anchor antenna element.
The anchor and the mobile tag are considered as points in space. The unknown mobile tag spherical
coordinate (ϕi , ϑi ), as observed by each anchor node installed on the ceiling at (xi , yi ), are:
 y−y 
ϕi = atan x−x i


i

  

  , (10)
π zi −z

 ϑi =



 2 − atan q
 

2 2
(x−x ) +( y−y )
i i

while the noise contribution Ń is described by:

Ń = NMN (µ, σ) = NMN (0, εRSSIMAX ). (11)

Here it is assumed an RSSI noise characterized by a Gaussian model with null mean and variance
εRSSIMAX , [16]. Moving on to the expected RSSI data, the measurement noise term embeds both the
effective RSSI detection error and second order effects (e.g., fast fading). It is possible to get a reliable
characterization of the RSSI variance, removing fast-fading by the rejection techniques described in [12].
Considering an example of the typical recorded RSSI data by the anchor SBA elements 1–7, reported
in Figure 10, the worst cases show a maximum RSSI drift of about 5 dB during a single steering
vector acquisition. Given this observation, we can set for the simulations the value σ = εRSSIMAX = 5.
Note that in a real implementation, the RSSI variance can be reduced by applying a mobile mean along
the acquired steering vectors.
The simulation results across the entire area for the three cases of study are reported in Figure 11,
in terms of Cumulative Error Distribution Function (CDF) for the area within the single rectangular
mesh and across the entire room, respectively. In this set of simulations, we generated 5 RSSI randomly
distributed according the noise model described above at each of the 299 points distributed as in
Figure 9a. From the figure we can observe that, due to the reduced dimension, the tiny mesh is
capable to estimate with comparably smaller error the position within the area defined by the mesh
itself, but it is not suitable for the outer part, as it determines high errors at the border of the room;
this is already predicted by the investigation reported in Section 3.2 and specifically Figure 7a. For the
other mesh sizes, the data of Figure 11 show a convergence between the outer and inner mesh results,
thus confirming that the large mesh provides almost the same results within and across the entire area,
as the two results almost the same.
Electronics 2020,9,
Electronics 2020, 9,1026
x FOR PEER REVIEW 12
12 of
of 18
18

Figure 10. Typical Received Signal Strength Indicator (RSSI) acquisition trace: case C, anchor installed
at room up-right corner (5.75, 5.00) m.

The simulation results across the entire area for the three cases of study are reported in Figure
11, in terms of Cumulative Error Distribution Function (CDF) for the area within the single
rectangular mesh and across the entire room, respectively. In this set of simulations, we generated 5
RSSI randomly distributed according the noise model described above at each of the 299 points
distributed as in Figure 9a. From the figure we can observe that, due to the reduced dimension, the
tiny mesh is capable to estimate with comparably smaller error the position within the area defined
by the mesh itself, but it is not suitable for the outer part, as it determines high errors at the border of
the room; this is already predicted by the investigation reported in Section 3.2 and specifically Figure
7a. For the other mesh sizes, the data of Figure 11 show a convergence between the outer and inner
meshFigure
results,
Figure 10.thus
10. confirming
Typical
Typical Received that the
Received Signal
Signal large mesh
Strength
Strength provides
Indicator
Indicator almost thetrace:
(RSSI)acquisition
(RSSI) acquisition samecase
trace: results
case within
C, anchor
C, anchor and across
installed
installed
the entire area,
at room
at as the
room up-right two
up-right corner results
corner (5.75, almost
(5.75, 5.00)
5.00)m.m. the same.

The simulation results across the entire area for the three cases of study are reported in Figure
11, in terms of Cumulative Error Distribution Function (CDF) for the area within the single
rectangular mesh and across the entire room, respectively. In this set of simulations, we generated 5
RSSI randomly distributed according the noise model described above at each of the 299 points
distributed as in Figure 9a. From the figure we can observe that, due to the reduced dimension, the
tiny mesh is capable to estimate with comparably smaller error the position within the area defined
by the mesh itself, but it is not suitable for the outer part, as it determines high errors at the border of
the room; this is already predicted by the investigation reported in Section 3.2 and specifically Figure
7a. For the other mesh sizes, the data of Figure 11 show a convergence between the outer and inner
mesh results, thus confirming that the large mesh provides almost the same results within and across
the entire area, as the two results almost the same.

Figure 11.
Figure 11. Simulated
Simulated localization
localization error CDFs,
CDFs, with
with the
the three
three anchor
anchor networks.
networks. Data
Data acquired
acquired at
at the
the
299 points
299 points in
in the
the room
room of
of Figure
Figure 9a,
9a, assuming
assuming the
the mobile
mobile tag
tag pointing
pointing upward,
upward, 55 averaged
averaged RSSI
RSSI data
data
for point
for point and
and each
each antenna
antenna element,
element, mobile
mobile tag height== 1.1 m.
tag height

For a better comparison of the data, the results are summarized in Table 2. They clearly show
that a trade-off involved in the anchor distribution exists: a wider mesh increases the sub-metrical
coverage on the full area, while smaller configurations increase the precision inside the mesh yet
degrade significantly the overall area performance. With this conclusion we have an objective way to
evaluate the dimension of the mesh according to the accuracy needed and the dimensions of the room.
The next Section deals with the experimental verification of these results.

Figure 11. Simulated localization error CDFs, with the three anchor networks. Data acquired at the
299 points in the room of Figure 9a, assuming the mobile tag pointing upward, 5 averaged RSSI data
for point and each antenna element, mobile tag height = 1.1 m.
 For a better comparison of the data, the results are summarized in Table 2. They clearly show
that a trade-off involved in the anchor distribution exists: a wider mesh increases the sub-metrical
coverage on the full area, while smaller configurations increase the precision inside the mesh yet
degrade significantly the overall area performance. With this conclusion we have an objective way to
evaluate2020,
Electronics the 9,dimension
1026 of the mesh according to the accuracy needed and the dimensions 13 ofofthe
18
room. The next Section deals with the experimental verification of these results.

Table 2. Simulated
Table 2. Simulated positioning
positioning error
error results.
results.

Configuration
A A BB C C
Configuration Large Medium Tiny
Large Medium Tiny
Mean Error 0.71 m 0.71 m
Mean Error 0.99 mm
0.99 3.11 m 3.11 m
Full Room Full Room
Coverage * 78.26%78.26%
Coverage * 66.52%
66.52% 26.30%
26.30%
WithinWithin MeanArea
the the Mesh Error 0.73 m 0.73 m
Mean Error 0.51 mm
0.51 0.13 m 0.13 m
Coverage * 73.21% 93.81% 100%
Mesh Area Coverage * 73.21% 93.81% 100%
* Percentage of the full room area with sub-metrical localization error (≤1 m).
* Percentage of the full room area with sub-metrical localization error (≤1 m).

4.3. Experimental Results

4.3. Experimental Results
The
The constellation
constellation of ofthe
the44anchors
anchorsnetwork
networkwas wasimplemented
implementedinin the real
the realcase
casescenario
scenario depicted
depictedin
Figure 9c. 9c.
in Figure TheThe
three meshmesh
three configurations represented
configurations in Figure
represented 9b define
in Figure three different
9b define case studies.
three different case
In the experiments described herein after, the mobile tag was installed on a tripod
studies. In the experiments described herein after, the mobile tag was installed on a tripod at at a fixed height of
a fixed
1.1 m and moved across the reference points indicated on the map of Figure
height of 1.1 m and moved across the reference points indicated on the map of Figure 9a, while a 9a, while a convenient
number
convenientof 5number
RSSI traces of 5 were
RSSI acquired
traces wereandacquired
averaged andforaveraged
each point.forThese traces These
each point. were exploited
traces wereto
estimate the mobile tag position by the estimator of Equation (3). Finally, some
exploited to estimate the mobile tag position by the estimator of Equation (3). Finally, some mechanical positioning
techniques
mechanical (i.e., laser position
positioning measurements,
techniques (i.e., laserand mobilemeasurements,
position tag antenna alignment)
and mobile provide a better
tag antenna
consistency
alignment) of the measured
provide a betterdata across the experiments.
consistency of the measured These data
experiments
across have a two-fold objective:
the experiments. These
the main consists in drawing conclusions about the trade-off between
experiments have a two-fold objective: the main consists in drawing conclusions about mesh size, the the
localization
trade-off
accuracy and room
between mesh size,area; the second is
the localization the assessment
accuracy and room about thethe
area; reliability
second isof the
the assessment
simulation presented
about the
in the previous section.
reliability of the simulation presented in the previous section.
Figure
Figure 12
12depicts
depictsthe theCDF
CDFofofthethe
position estimation
position estimation by the measured
by the measuredRSSIRSSI
data data
for theforthree cases
the three
of study,
cases respectively.
of study, Additionally,
respectively. in this in
Additionally, case thecase
this results
the are differentiated
results into twointo
are differentiated sets,two
namely
sets,
the position errors inside the meshes and those across the entire room area,
namely the position errors inside the meshes and those across the entire room area, with each casewith each case of study
of
differently colored.
study differently colored.

Figure 12.
Figure 12. Measured
Measuredlocalization
localizationerror
error CDFs,
CDFs, forfor
thethe three
three cases
cases of study.
of study. Data Data acquired
acquired at the
at the 299 299
points
points in the room of Figure 9a, assuming the mobile tag pointing upward, 5 averaged RSSI
in the room of Figure 9a, assuming the mobile tag pointing upward, 5 averaged RSSI data for point and data for
pointantenna
each and each antennamobile
element, element,
tagmobile = 1.1
height tag height
m. = 1.1 m.

This approach is motivated by the fact that for large areas a system architect could be interested
in covering the domain by implementing a constellation of smaller meshes instead of adopting a
larger one. This solution leads to better accuracy at the expense of increased system complexity.
The localization results for the three experiments is summarized in Table 3.
Electronics 2020, 9, 1026 14 of 18

Table 3. Measured localization error results.

A B C
Configuration
Large Medium Tiny
Mean Error 0.70 m 0.85 m 1.75 m
Full Room
Coverage * 79.13% 72.12% 34.35%
Mean Error 0.66 m 0.47 m 0.32 m
Within the Mesh Area
Coverage * 82.09% 96% 100%
* Percentage of the full room area with sub-metrical localization error (≤1 m).

From the data we can observe what follows: Within the mesh boundary the localization accuracy
increases when reducing the mesh area. Nevertheless, outside the same mesh the system is still
able to provide effective localization with a meaningful accuracy. The wider mesh configuration, ‘A’,
shows a good average localization error across the entire area, balancing its distribution all over the
room. On the contrary, adopting the smallest mesh, i.e., ‘C’, the accuracy degrades significantly when
evaluated in average across the entire room area. The ‘B’ mesh represents effectively a transition
between the two cases, thus confirming the coherence between the experimental data.
It is noteworthy that CRB simulation results as shown in Figure 8 show a qualitative identical
trend when compared with experiments. In comparison with the curves in Figure 8 the minimum
achievable localization error within the mesh area shows a monotonous growth as the corresponding
mesh size increases, exactly as depicted in Table 3. On the contrary, considering the entire room area,
the average error decreases by increasing the mesh size until realizing an “optimum” mesh-size for
which it is minimized.
The collected data leads to a straightforward conclusion: we can assess a tradeoff between network
complexity and estimation accuracy; in fact, the smallest cell leads to the better local accuracy, while the
larger leads to better mean error across the entire room. From this consideration, we foresee that
considering smallest mesh sized repeated side-by-side lead to a better accuracy across larger areas,
in place of one larger mesh. For example, following the data reported in Table 3 and considering the
three mesh sizes, cf. Table 1, we see that the accuracy could be improved by a factor 2 (from 0.66 m to
0.32 m) moving from a single “large” mesh to a combination of about 20 “tiny” meshes, which would
be necessary to cover the same area. Obviously, the significant increase of anchor nodes may cause
a dramatic increase of costs, thus this trading-off drives the optimization of the positioning system
between accuracy and complexity. Nevertheless, this assessment requires an experimental validation,
which is at presently beyond the scope of this work.
Finally, from the comparison between the results provided in Tables 2 and 3, we can observe that
for the cases of study ‘A’ and ‘B’ there are significant matches for both the ‘full area’ and ‘within the
mesh area’. The differences between the two accuracies are between 1 cm and 14 cm, and indicate that
the measured results are consistent with the simulations of Section 4.2. Contrastingly, the comparison
for the case of study ‘C’, i.e., the tiny mesh, reveals the worst agreement for both the data related to
the ‘entire room’ and ‘within the mesh area’. This is due to the effective difference between the mesh
size and the room size, which amplifies the model inaccuracy given the fact that the small number of
points within the mesh causes large relative fluctuation in the results.

4.4. Discussion about the Simulated Positioning Results

As emerges from the previous sections, the behavior of the localization accuracy allows to identify
a trade-off for wider observation domains, where an overall acceptable accuracy is easily achievable
without using a high number of anchors by exploiting larger meshes. By using a simple model like the
one discussed in Section 4.2, the system architect can tune positioning accuracy by reducing their area
and increasing the number of anchors, so that the localization system can be considered scalable for
different application scenarios, [28]. This latter consideration can be sustained by comparing Figures 13
and 14, which show the maps of the localization error obtained with the three mesh dimensions;
 identify a trade-off
achievable without using for wider
a high observation
number ofdomains,
anchors by where an overall
exploiting larger acceptable
meshes. By accuracy
using aissimple
easily
achievable without using a high number of anchors by exploiting
model like the one discussed in Section 4.2, the system architect can tune positioning accuracy by larger meshes. By using a simple
model like
reducing the area
their one discussed
and increasing in Section 4.2, the of
the number system
anchors, architect
so thatcanthe tune positioning
localization accuracy
system can be by
reducing their area and increasing the number of anchors, so that the
considered scalable for different application scenarios, [28]. This latter consideration can be sustained localization system can be
considered
by 2020,scalable
comparing
Electronics Figures
9, 1026 for13different
and 14,application
which show scenarios,
the maps [28].
of This latter consideration
the localization can be sustained
error obtained with
15 ofthe
18
by comparing
three mesh dimensions;Figures 13the and 14, which
results come showfrom the maps of and
simulations the localization
measurements error obtained with
respectively, whose the
three meshare
summaries dimensions;
provided in theTables
results2–3.comeThefrom simulations
simulation modeland showsmeasurements
a significantrespectively,
level of agreement whose
the
with results
summaries come
the measured from
are provided simulations
results,in in
Tables and measurements
2–3. The
particular for simulation
the large and respectively,
model
medium whose
showsmeshes, summaries
a significant are provided
levelconcerns
for what of agreementboth
in
the Tables
with the 2measured
accuracy and and 3. the
The simulation
results, model
in particular
distribution of shows
for the alarge
localization significant
error. and
The mediumlevel ofmeshes,
accuracy agreement
of with
canthe
for what
this model be measured
concerns both
attributed
results,
to the adoption of circular polarized antenna that are inherently robust to multipath, [29,30]. Thisthe
the in
accuracy particular
and thefor the large
distribution and
of medium
localization meshes,
error. for
The what concerns
accuracy of thisboth
modelthe accuracy
can be and
attributed is
distribution
due the of
to thetoadoption localization
changeof circular error. Theantenna
polarized
of polarization accuracy
rotation, thatofare
i.e.,this model
hand can
inherently
left be attributed
robust
versus to multipath,
right to the
hand, adoption
[29,30].
once thatThis
theofis
circular
due to polarized
the change
electromagnetic antenna
wave of that are
polarization
bounces inherently robust
rotation,
a conductive i.e.,tothus
object, multipath,
left [29,30].
hand versus
resulting This ishand,
right
in a cross-polarizeddue toonce
the change
wave that of
the
at the
polarization
electromagnetic
receiver rotation,
antenna that i.e.,
wave left hand
isbounces
therefore versus right
a rejected.
conductive hand,thus
object,
This occurrence onceresulting
that the electromagnetic
justifies in aadoption
the of awave
cross-polarized bounces
wave
coarse at the
channel a
conductive
receiverlike
model, object,
antenna
the one thus
that resulting
is therefore
adopted in a cross-polarized
in thisrejected.
model, This occurrence
although wave at the receiver
justifiesofthe
the adoption more antenna
adoption that
case ofofstudies is
a coarse therefore
channel
should be
rejected.
model, This
like the occurrence
one adopted justifies
considered for a complete model validation. in this the adoption
model, althoughof a coarse
the channel
adoption of model,
more caselike
of the one
studies adopted
should be
in this model, although the adoption
considered for a complete model validation. of more case of studies should be considered for a complete
model validation.

(a) (b) (c)

(a) (b) (c)
Figure 13. Simulated localization error distribution over room area for each mesh configuration: (a)
Figure
Figure
large 13. Simulated
13.
mesh; Simulated
(b) mediumlocalization
localization error
error
mesh; (c) tiny distribution
distributionover
mesh. overroom
roomarea forfor
area each mesh
each configuration:
mesh (a)
configuration:
large mesh; (b) medium mesh; (c) tiny mesh.
(a) large mesh; (b) medium mesh; (c) tiny mesh.

(a) (b) (c)

(a) (b) (c)
Figure 14.
Figure 14. Measured
Measuredlocalization
localizationerror
errordistribution
distribution over
over space
space for for
eacheach
meshmesh configuration
configuration (white(white
area
Figure
area
coincides14. Measured
coincides
with localization
with inaccessible
inaccessible room roomerror
space, distribution
space,
see 9b), over
see Figure
Figure (a)9b),space for mesh;
(a)mesh;
large large each
(b) mesh configuration
(b) medium
medium mesh;
mesh; (c) (white
tiny(c) tiny
mesh.
area coincides with inaccessible room space, see Figure 9b), (a) large mesh; (b) medium mesh; (c) tiny
mesh.
5. Conclusions
mesh.
5. Conclusions
This paper has experimentally assessed the effects of the mesh area on localization accuracy
5. Conclusions
for a This
distributed indoor localizationassessed
system the
based on aofconstellation of on
four anchors composed of
paper has experimentally effects the mesh area localization accuracy for
switched
This beam
paper antennas
has and capable
experimentally to acquire
assessed the communication
effects of the link
mesh RSSI,
area on [15,16]. Due
localization
a distributed indoor localization system based on a constellation of four anchors composed of to its
accuracydirect
for
interoperability
a distributed
switched with an
beamindoor
antennas IEEE 802.11x/802.15.4
localization
and capablesystem based
to acquire radio
on access technology,
a constellation
communication link of in prospective
four
RSSI, anchors
[15,16]. this
to itssystem
Duecomposed of
direct
can be easily
switched beamintegrated
interoperability antennas
with inand
an802.11x/802.15.4
an IEEE existing
capable wireless
to acquirenetwork,
accessachieving
communication
radio theRSSI,
link
technology, ingoal of providing
[15,16].
prospective Due a wireless
this to its direct
system can
data access system
interoperability augmented
with with indoor localization,
an IEEE 802.11x/802.15.4 maintaining
radio access theinsame
technology, communication
prospective this systemlayers.
can
This latter characteristic allows the WSN system architect designers to easily implement also smart
routing network protocols, such as the one presented in [7].
The mesh area influences the localization accuracy significantly, demonstrating a clear trade-off
between in-mesh and out-mesh results. The experimental data in the specific scenario presented
in this work demonstrate that larger mesh provides a significant trade-off between accuracy and
system complexity.
Provided that the aforementioned system is meant to allow an easy to integrate and non-invasive
localization platform, the anchor mesh positions should be easily selected through an a priori process
Electronics 2020, 9, 1026 16 of 18

that does not require complex steps of either calibration or testing. To help on this, the proposed
simulation procedure has been validated by measurements results and thus can be used to forecast the
optimal anchor network mesh area.
Table 4 shows a comparison between the presented system and actual state-of-art to the knowledge
of the authors. Note that the system under consideration in this work is the only one that does not
require any kind of training or supplemental data sources, and despite this its localization accuracy
results are comparable with the one achieved by other state-of-art systems, while maintaining a similar
density of anchors. This feature result is very important, and enabling for the optimization of future
indoor localization system deployment.

Table 4. State of art comparison.

N. of Room Size Anchor Density Mean Error

Ref Technology Training
Anchor (m2 ) (1/m2 ) (m)
[31] 3 CSI/RSSI Yes 40 0.075 0.60/0.90
[32] 4 RSSI + IMU Yes 150 0.027 1.80
[19] 15 RSSI + IMU Yes 1609 0.009 3.42
[18] 6 RSSI + IMU Yes 157 0.038 1.00
[33] 6 RSSI + IMU Yes 1250 0.005 1.00
[34] 4 RSSI Yes 45 0.089 1.22
[15] 4 SBA + RSSI Yes 34.5 0.120 1.08
This Work 4 SBA + RSSI No 34.5 0.120 0.70

Author Contributions: Conceptualization, M.P. and A.C.; methodology, M.P. and S.M.; software, M.P.; validation,
M.L.; formal analysis, M.P. and A.C.; investigation, M.P. and A.C.; resources, A.C. and G.C.; data analysis, M.L.;
writing—original draft preparation, M.P.; writing—review and editing, M.L. and G.C.; writing—final version
A.C.; visualization, M.L.; supervision, A.C.; project administration, G.P.; funding acquisition, G.C. All authors
have read and agreed to the published version of the manuscript.
Funding: This research was funded by Regione Toscana POR CreO FESR 2014–2020—Action 1.1.5.a3—Bando
FAR—FAS 2014, under the “SUMA—Struttura Urbana Multifunzionale Attiva” project.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the
study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to
publish the results.

References
1. Zekavat, R.; Bueher, R.M. Handbook of Position Location: Theory, Practice and Advances; John Wiley & Sons:
Hoboken, NJ, USA, 2011; Volume 27.
2. Liu, H.; Darabi, H.; Banerjee, P.; Liu, J. Survey of Wireless Indoor Positioning Techniques and Systems.
IEEE Trans. Syst. Man Cybern. Part C 2007, 37, 1067–1080. [CrossRef]
3. Pau, G.; Chaudet, C.; Zhao, D.; Collotta, M. Next Generation Wireless Technologies for Internet of Things.
Sensors 2018, 18, 221. [CrossRef] [PubMed]
4. Pita, R.; Utrilla, R.; Rodriguez-Zurrunero, R.; Araujo, A. Experimental Evaluation of an RSSI-Based
Localization Algorithm on IoT End-Devices. Sensors 2019, 19, 3931. [CrossRef]
5. Ferrández-Pastor, F.-J.; García-Chamizo, J.M.; Nieto-Hidalgo, M.; Mora-Martínez, J. Precision Agriculture
Design Method Using a Distributed Computing Architecture on Internet of Things Context. Sensors 2018,
18, 1731. [CrossRef] [PubMed]
6. Montoliu-Colás, R.; Sansano, E.; Gascó, A.; Belmonte, O.; Caballer, A. Indoor Positioning for Monitoring
Older Adults at Home: Wi-Fi and BLE Technologies in Real Scenarios. Electronics 2020, 9, 728. [CrossRef]
7. Liu, M.; Cao, J.; Chen, G.; Wang, X. An Energy-Aware Routing Protocol in Wireless Sensor Networks. Sensors
2009, 9, 445–462. [CrossRef]
8. Lin, X.; Bergman, J.; Gunnarsson, F.; Liberg, O.; Razavi, S.M.; Razaghi, H.S.; Rydn, H.; Sui, Y. Positioning for
the Internet of Things: A 3GPP Perspective. IEEE Commun. Mag. 2017, 55, 179–185. [CrossRef]
9. Chou, P.H.; Hsu, Y.L.; Lee, W.L.; Kuo, Y.C.; Chang, C.C.; Cheng, Y.S.; Lee, H.H. Development of a smart home
system based on multi-sensor data fusion technology. In Proceedings of the 2017 International Conference
on Applied System Innovation (ICASI), Sapporo, Japan, 13–17 May 2017.
Electronics 2020, 9, 1026 17 of 18

10. Hou, Y.; Yang, X.; Abbasi, Q.H. Efficient AoA-Based Wireless Indoor Localization for Hospital Outpatients
Using Mobile Devices. Sensors 2018, 18, 3698. [CrossRef]
11. Silva, P.F.E.; Kaseva, V.; Simona, L.E. Wireless Positioning in IoT: A Look at Current and Future Trends.
Sensors 2018, 18, 2470. [CrossRef]
12. Lymberopoulos, D.; Liu, J.; Yang, X.; Choudhury, R.R.; Handziski, V.; Sen, S. A realistic evaluation and
comparison of indoor location technologies: Experiences and lessons learned. In Proceedings of the 14th
International Conference on Information Processing in Sensor Networks, Seattle, WA, USA, 14–16 April 2015.
13. Chen, Z.; Zou, H.; Jiang, H.; Zhu, Q.; Soh, Y.C.; Xie, L. Fusion of WiFi, Smartphone Sensors and Landmarks
Using the Kalman Filter for Indoor Localization. Sensors 2015, 15, 715–732. [CrossRef]
14. Yi, D.-H.; Lee, T.-J.; Cho, D.-I.D. A New Localization System for Indoor Service Robots in Low Luminance
and Slippery Indoor Environment Using Afocal Optical Flow Sensor Based Sensor Fusion. Sensors 2018,
18, 171. [CrossRef] [PubMed]
15. Maddio, S.; Passafiume, M.; Cidronali, A.; Manes, G. A Distributed Positioning System Based on a Predictive
Fingerprinting Method Enabling Sub-Metric Precision in IEEE 802.11 Networks. IEEE Trans. Microw.
Theory Tech. 2015, 63, 4567–4580. [CrossRef]
16. Passafiume, M.; Maddio, S.; Cidronali, A. An Improved Approach for RSSI-Based only Calibration-Free
Real-Time Indoor Localization on IEEE 802.11 and 802.15.4 Wireless Networks. Sensors 2017, 17, 717.
[CrossRef] [PubMed]
17. Cidronali, A.; Collodi, G.; Lucarelli, M.; Maddio, S.; Passafiume, M.; Pelosi, G. Continuous Beam Steering
For Phaseless Direction-of-Arrival Estimations. IEEE Antennas Wirel. Propag. Lett. 2019, 18, 2666–2670.
[CrossRef]
18. Deng, Z.-A.; Hu, Y.; Yu, J.; Na, Z. Extended Kalman Filter for Real Time Indoor Localization by Fusing WiFi
and Smartphone Inertial Sensors. Micromachines 2015, 6, 523–543. [CrossRef]
19. Wu, C.; Yang, Z.; Liu, Y. Smartphones Based Crowdsourcing for Indoor Localization. IEEE Trans. Mob. Comput.
2015, 14, 444–457. [CrossRef]
20. Hartmann, M.; Nowak, T.; Patino-Studencki, L.; Robert, J.; Heuberger, A.; Thielecke, J. A Low-Cost RSSI-Based
Localization System for Wildlife Tracking. IOP Conf. Ser. Mater. Sci. Eng. 2016, 120, 012004. [CrossRef]
21. Maddio, S.; Passafiume, M.; Cidronali, A.; Manes, G. Impact of the Dihedral Angle of Switched Beam
Antennas in Indoor Positioning based on RSSI. In Proceedings of the 11th European Radar Conference
(EuRAD), Rome, Italy, 6–9 October 2014; pp. 317–320.
22. Maddio, S.; Cidronali, A.; Passafiume, M.; Collodi, G.; Maurri, S. Fine-grained azimuthal direction of arrival
estimation using received signal strengths. Electron. Lett. 2017, 53, 687–689. [CrossRef]
23. Cidronali, A.; Collodi, G.; Maddio, S.; Passafiume, M.; Pelosi, G. 2-D DoA Anchor Suitable for Indoor
Positioning Systems Based on Space and Frequency Diversity for Legacy WLAN. IEEE Microw. Wirel.
Components Lett. 2018, 28, 627–629. [CrossRef]
24. Wu, D.; Cheung, S.W.; Yuk, T.I. A Compact and low-profile Loop Antenna with Multiband Operation for
Ultra-thin Smartphones. IEEE Trans. Antennas Propag. 2015, 63, 2745–2750. [CrossRef]
25. Maddio, S. A Circularly Polarized Switched Beam Antenna with Pattern Diversity for WiFi Applications.
IEEE Antennas Wirel. Propag. Lett. 2016, 16, 125–128. [CrossRef]
26. Cidronali, A.; Maddio, S.; Giorgetti, G.; Manes, G. Analysis and Performance of a Smart Antenna for 2.45-GHz
Single-Anchor Indoor Positioning. IEEE Trans. Microw. Theory Tech. 2009, 58, 21–31. [CrossRef]
27. Chen, Y.; Yang, J.; Trappe, W.; Martin, R.P. Impact of anchor placement and anchor selection on localization
accuracy. In Handbook of Position Location: Theory, Practice, and Advances; John Wiley & Sons: Hoboken, NJ,
USA, 2011; pp. 425–455.
28. Razavi, A.; Valkama, M.; Simona, L.E. Robust Statistical Approaches for RSS-Based Floor Detection in Indoor
Localization. Sensors 2016, 16, 793. [CrossRef] [PubMed]
29. Maddio, S.; Passafiume, M.; Cidronali, A.; Manes, G. A scalable distributed positioning system augmenting
WiFi technology. In Proceedings of the 2013 International Conference on Indoor Positioning and Indoor
Navigation, IPIN 2013, Montbéliard, France, 8–31 October 2013.
30. Maddio, S.; Cidronali, A.; Passafiume, M.; Collodi, G.; Lucarelli, M.; Maurri, S. Multipath Robust Azimuthal
Direction of Arrival Estimation in Dual-Band 2.45–5.2 GHz Networks. IEEE Trans. Microw. Theory Tech. 2017,
65, 4438–4449. [CrossRef]
Electronics 2020, 9, 1026 18 of 18

31. Wu, K.; Xiao, J.; Yi, Y.; Chen, D.; Luo, X.; Ni, L.M. CSI-Based Indoor Localization. IEEE Trans. Parallel
Distrib. Syst. 2012, 24, 1300–1309. [CrossRef]
32. Zhang, R.; Xia, W.; Jia, Z.; Shen, L. The Indoor Localization Method based on the Integration of RSSI
and Inertial Sensor. In Proceedings of the Global Conference on Consumer Electronics, Tokyo, Japan,
7–10 October 2014; IEEE: Salt Like City, UT, USA, 2014; pp. 332–336.
33. Chen, W.; Wang, W.; Li, Q.; Chang, Q.; Hou, H. A Crowd-Sourcing Indoor Localization Algorithm via Optical
Camera on a Smartphone Assisted by Wi-Fi Fingerprint RSSI. Sensors 2016, 16, 410. [CrossRef]
34. Luo, X.; O’Brien, W.J.; Julien, C. Comparative evaluation of Received Signal-Strength Index (RSSI) based
indoor localization techniques for construction jobsites. Adv. Eng. Inform. 2011, 25, 355–363. [CrossRef]

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).