Adaptive Indoor Localization by using Environmental Thresholding and Virtual Fingerprint Technique

Hakan Koyuncu. Int. Journal of Engineering Research and Application www.ijera.com
ISSN : 2248-9622, Vol. 6, Issue 12, (Part -1) December 2016, pp.86-91
www.ijera.com 86 | P a g e
Adaptive Indoor Localization by using Environmental
Thresholding and Virtual Fingerprint Technique
Hakan Koyuncu1
, Baki Koyuncu2
ABSTRACT
Environmental Thresholding and virtual fingerprinting techniques are deployed with Wireless Sensor Nodes
(WSN) to create an adaptive localization system. A virtual fingerprint map of RSSI values is generated across
the test area. RSSI amplitude correction phase is introduced with respect to local environmental parameters on
virtual and recorded RSSI values at fingerprint grid points and unknown object points. Localization algorithms
are employed to determine the unknown object locations. Localization accuracies of around 35cm at a grid
space of 1m are obtained during the calculations.
Keywords: WSN, RFID, Adaptive localization, Virtual Fingerprint, threshold, RSSI, k-NN, weighted k-NN
I. INTRODUCTION
Wireless sensor nodes (WSN) are
commonly used for object localization in recent
years. They are deployed in indoors and outdoors
for position detection purposes. Many localization
algorithms have been introduced over the years by
using received signal strength (RSSI) [1, 2] values.
Radio frequency based devices (RFID) are widely
used as WSN nodes due to their low cost and low
power consumption [3]. Some of the best know
localization algorithms are ranged based [4], range
free [5], direction of arrival (DOA) [6], Frequency
Difference of Arrival (FDA) and Time Difference
of Arrival (TDA) algorithms. Additional
techniques such as triangulation, trilateration,
together with time intervals and RSSI values can be
used to find unknown positions.
In literature, an environmentally adaptive
path loss localization method is developed by
Zhang et al. [7]. In this study, ranging errors are
calibrated by broadcasted voltage levels and indoor
multi-fading together with antenna effects are used
adaptively to generate path lose models. On the
other hand, Janire Larranaga et al. has developed
another environmentally adaptive localization
method [8] where RSSI signal levels are obtained
between the communications of WSNs and the
environmental effects are taken into account by the
signal level changes since the last signal request,
[9].
In the current study, a virtual fingerprint
map is incorporated into localization calculations
[10]. Anchor transmitter WSNs radiate RF signals
across the test area. Transmitted signal amplitudes
decrease in exponential form against distances in
free space [11, 12]. Mathematical formulations of
these exponential forms are deployed and virtual
RSSI values are theoretically calculated at every
grid point. Finally, a virtual fingerprint map is
constructed for each anchor transmitter WSN.
Virtual RSSI values at grid points are adaptively
corrected with respect to environmental factors and
the resultant values are deployed in position
calculations.
Hence, a new localization technique is proposed in
this study where adaptive localization and virtual
fingerprint techniques are combined together. After
a brief introduction in section 1, theoretical
background of adaptive localization and RSSI
correction phase is given in section 2. In section 3,
virtual fingerprint generation is explained. Results
are given in section 4 together with conclusions in
section 5.
II. ADAPTIVE LOCALIZATION
WSNs are employed as transmitters and receivers
during RSSI measurements. Transmitters are
generally placed at the corners of the test area and a
receiver is placed on an unknown mobile object as
shown in Fig. 1. Received RF signals by the object
receiver in RSSI form are sent to a server computer
through a wireless media for further processing.
Figure 1: Block diagram of the proposed system
across the test area
RESEARCH ARTICLE OPEN ACCESS

Environmental factors are identified as the
local obstacles and walls around the test areas and
the negative effects introduced by them. These
obstacles cause multiple reflections of RF signals
and the resultant signal additions and cancellations
cause random variations in RSSI measurements.
RSSI variations must be reduced to minimum
values so that accurate localization calculations can
be carried out. This reduction is identified as the
adaptive correction of RSSI values.
Data frame of N number of RSSI values
with an identity code is broadcasted from a
transmitter WSN. Object receiver WSN records
these RSSI values and sends them to a server
computer with respect to specific transmitter. Mean
value and the standard deviation of N number of
received RSSI values in one frame are defined as
RMean and RSTD. To minimize the random behavior
of RSSI values, a signal interval of (RMean – RSTD <
RMean< RMean + RSTD) is selected and RSSI
measurement values only within this interval are
considered for localizations.
If there are ‘q’ number of RSSI values in RMean ±
RSTD interval, ‘m’ is taken as the number of RSSI
values which are less than or equal to the mean of
‘q’ number of RSSI values. The average value of
‘m’ number of RSSI values is given as
 

m
i
iX RSSI
m
R
1
1
where    

q
J
Ji RSSI
q
RSSI
1
1
(1)
(q–m) is the number of RSSI values whose values are more than the mean of q RSSI values. The average value
of (q–m) number of RSSI values is given as
 



mq
i
iY RSSI
mq
R
1
1
where    

q
J
Ji RSSI
q
RSSI
1
1
(2)
Hence, RSSI value generated at a measurement location with respect to, transmitter A is expressed as
  YAXAA RRRSSI   1 (3)
A is defined as the constant environmental factor
depending on STDAR value of RSSI values arriving
from transmitter A. Similar RSSI values can be
utilized at the same measurement location for
transmitters B, C and D as in Fig.1. Finally, an
average threshold standard deviation, STDT is
considered for all the transmitter WSNs given as
 STDDSTDCSTDBSTDASTD RRRRT 
4
1
(4)
STDT identifies the average environmental
conditions due to the fact that RSSI values and the
resultant RSTD values, change with respect to
environmental conditions.
A is defined in terms of STDT and STDAR as





 

STD
STDASTD
A
T
RT
1
2
1
 for STDASTD RT  (5)





 

STD
STDSTDA
A
T
TR
1
2
1
 for STDASTD RT 
(6)
Calculated ARSSI value in equation (3) is
considered as stable if STDSTDA TR  and A is
calculated as A ∈ (0.5, 1). Calculated ARSSI
value is considered as unstable if STDSTDA TR 
and A is calculated as A ∈ (0, 0.5).
Environmentally adaptive RSSIA value is
calculated by substituting A , RX and RY values
in equation (3). Similarly, adaptive BRSSI ,
CRSSI , DRSSI values can be calculated at the
same measurement point.

III. VIRTUAL FINGERPRINT
GENERATION
In classical fingerprint map generation, RF
signal amplitudes arriving from transmitter WSNs
at each grid point are measured as the fingerprint at
that point. Map of these points is termed as the
fingerprint map. Hence, fingerprint map contains a
collection of RF signal amplitudes corresponding
to all the transmitters at each grid point. Signal
amplitude measurements at object points are
compared with the signal amplitude measurements
at grid points of the fingerprint map. Closest grid
points are used in position calculations.
Virtual fingerprint map, on the other hand,
is generated by using theoretical RF signal
amplitude distributions instead of physical
measurements. Free space propagation of RF
signals between a transmitter and a receiver is
modeled for this purpose. Exponential signal
amplitude decrease between a transmitter and a
receiver against distance is formulated by equation
(7) and plotted in Fig 2.
ax
eCY 
 (7)
Figure2: RSSI distribution against distance
between a transmitter and a receiver
RSSI amplitude, Y, is calculated by using
equation (7) at every grid point with radiated signal
strength C of a transmitter and x distance between
the grid point and this transmitter. ‘a’ is the decay
constant generated during measurements. Hence,
virtual RSSI amplitudes at grid points are
calculated by using radiated signal strengths of
transmitters and the distances of grid points from
these transmitters with equation (7). A 3D
histogram map of calculated RSSI amplitudes at
grid points for transmitter A is displayed in Fig. 3.
Figure3: Virtual fingerprint map across the test
area for one transmitter
3D map in Fig.3 is termed as the virtual
fingerprint map and it is generated for one radiated
signal amplitude from a transmitter at an instant of
time. Similarly, N number of virtual fingerprint
maps can be generated from N number of signal
amplitudes radiated from one transmitter. Hence, N
numbers of virtual RSSI values can be obtained at
each grid point from each transmitter.
RF signal transmission from transmitters
displays random behavior in time domain and
constant C in equation (7) randomly changes. This
is also reflected in virtual RSSI amplitude
calculations at every grid point. N number of
virtual RSSI amplitudes are adaptively corrected
and reduced to one virtual signal amplitude for one
transmitter at each grid point. Final virtual
fingerprint map consists of adaptively corrected
virtual RSSI values as many as the number of
transmitters at every grid point.
Similarly, N numbers of RSSI values are
received by the object receiver from each
transmitter. They are also adaptively corrected and
their number is reduced to the number of
transmitters. Localization algorithms are deployed
with these final adaptively corrected virtual
fingerprint map and final adaptively corrected
object RSSI recordings to determine the object
location.
IV. EXPERIMENTATION AND
RESULTS
A rectangular area of 10mx6m with a grid
spacing of 1m is employed for tests and
measurements. It was a part of a sports hall with
minimum obstacles. WSN transmitters are placed
at the corners at a uniform height of 3 meters. A
student with a WSN receiver is considered as the
mobile unknown object. Jennic type WSN
transmitters and receivers are deployed [13]. RF
signals emitted from the transmitters are received
by the object receiver and stored in a database in a
server computer wirelessly interfaced to the
receiver.

Initially, calibration experiments of the transmitted
RF signals are conducted across the test area.
Transmitted RF signal amplitudes are plotted
against the distances between transmitters and
receivers as shown in Fig. 2. This is repeated for all
the transmitters and the object receiver. An average
curve, RSSI amplitude versus distance, is generated
to derive the equation (7).
a) Virtual Fingerprint map
At the end of the calibration stage,
equation (7) is generated to calculate the virtual
RSSI values at grid points. Grid distance to each
transmitter is the x value in equation (7). 4 virtual
RSSI amplitudes with respect to 4 transmitters are
calculated at every grid point and a virtual
fingerprint map is realized.
Secondly, N number of consecutive RF
signal transmission from each transmitter is carried
out and N number of virtual fingerprint map is
generated for a single transmitter. This constitutes
N number of virtual RSSI values at each grid point
for each transmitter. Similarly, N number of RSSI
recordings is also carried out at unknown object
location.
Finally, N number of virtual RSSI
recordings at each grid point and N number of
object RSSI recordings for each transmitter are
subjected to adaptive RSSI correction procedures.
A resultant virtual fingerprint map is generated
with 4 virtual adaptive RSSI values for 4
transmitters at each grid point together with 4
adaptive RSSI recordings for 4 transmitters at
object point.
b) Localization
Adaptively corrected Virtual fingerprint
map and object RSSI recordings for 4 transmitters
are deployed and object location is calculated by
using localization algorithms such as k-NN and
weighted k-NN across the test area. In case of k-
NN algorithm, ‘k’ number of nearest grid points to
the unknown object location are selected by
considering the smallest signal strength differences
between the grid and object points identified as
Euclidean distances. The object coordinates can be
defined as the average values of these nearest grid
coordinates. In case of weighted k-NN algorithm,
the nearest grid points are individually weighted
with respect to their Euclidean distances with the
object. The object coordinates can be defined as
the summation of the weighted coordinates.
Experimental results with physical and virtual
fingerprint mapping are presented in Fig.4. Object
locations are determined by using k-NN and
weighted k-NN algorithms in both mapping
techniques. No adaptive RSSI corrections are
employed in this phase.
Figure 4: localization with virtual and physical fingerprinting with no adaptive correction
Actual object coordinates are compared
with the calculated object coordinates. The results
revealed an average error distance of 66cm with a
grid space of 1m using virtual fingerprint map.
Secondly, adaptively corrected virtual fingerprint
map and adaptively corrected object RSSI
recordings are deployed and object location is
calculated by using k-NN and weighted k-NN
algorithms. The results are presented in Fig. 5.

Figure 5: Localization with virtual and physical fingerprinting with adaptive correction
As it is seen in Figure 5, object localization
accuracies are better with adaptively constructed
virtual fingerprint map and adaptively corrected
object RSSI recordings.
c) Results
In literature, Object localization
accuracies calculated with physical fingerprint
maps are generally around one grid space. This is
also realized in current study. Average localization
accuracy calculated with physical fingerprint map
is found to be 96 cm with a grid space of 1m. On
the other hand, average localization accuracy
becomes 66cm with virtual fingerprint map
employing the same grid space.
Adaptive RSSI corrections are applied on
both physical and virtual fingerprint maps together
with object RSSI recordings. As it is seen in Figure
5, the localization accuracies are quite improved
compared to Figure 4. Weighted k-NN method
with adaptively corrected virtual fingerprint maps
and object RSSI recordings gives the minimum
average localization error of 35cm with a 1m grid
space.
V. CONCLUSIONS
A hybrid fingerprint localization technique
is developed which is environmentally adaptive
and deploys virtual fingerprint mapping.
Previously, fingerprint maps are generated by
making measurements at every grid point and these
measured RSSI values are stored in a database.
Unknown object RSSI recordings are later
compared with the stored fingerprint values and
location of the object point is estimated by
averaging the nearest grid point coordinates.
In this study, RSSI values at grid points
are calculated theoretically based on RSSI
distributions in free space. Fingerprint map which
is generated by these theoretical RSSI values is
termed as virtual fingerprint map. This mapping
technique saves a lot of time and effort in RSSI
measurements and recordings. Initially, calibration
curves between transmitters and receivers and their
mathematical formulation are determined across
the test area.
Once the virtual fingerprint map is
established, RSSI values on the map and recorded
RSSI values at object locations are adaptively
adjusted to reduce the random variations among
them. Two popular localization algorithms, k-NN
and weighted k-NN, are used in these calculations.
Best localization accuracies are obtained with
weighted k-NN algorithm and adaptively corrected
virtual fingerprint map.
Finally, it can be concluded that the
important issue in this study is to prove the validity
of virtual fingerprint mapping technique coupled
with adaptive RSSI correction. The accuracy
results prove that localization of objects can
produce an accuracy of around 1/3 of the grid
space. This is a good improvement among
localization techniques without making too many
RSSI measurements and including the effects of
environmental problems.
REFERENCES
[1]. P.K. Singh, B. Tripathi, N.P. Singh, “Node
localization in wireless sensor networks”,
International Journal of Computer Science
and Information Technologies, Vol. 2 (6),
2011, pp 2568-2572
[2]. J. Kuriakose, S. Joshi, V.I. George,
“Localization in Wireless Sensor Networks:
A Survey”, CSIR Sponsored X Control
Instrumentation System Conference -
CISCON-2013, pp 73-75
[3]. G. Chandrasekaran, M.A. Ergin, J. Yang, S.
Liu, Y. Chen, M. Gruteser, R.P. Martin,
“Empirical Evaluation of the Limits on
Localization Using Signal Strength”, 6th
Annual IEEE Communications Society
Conference on Sensor, Mesh and Ad Hoc
Communications and Networks, 2009, pp 1-
9

[4]. K. Almuzaini, A. Gulliver, "Range-Based
Localization in Wireless Networks Using
Density-Based Outlier Detection," Wireless
Sensor Network, Vol. 2 No. 11, 2010, pp.
807-814
[5]. T. He, C. Huang, B.M. Blum, J.A.
Stankovic, T. Abdelzaher, “Range-Free
Localization Schemes for Large Scale
Sensor Networks”, Proceedings of the 9th
annual international conference on Mobile
computing and networking, 2003, pp 81-95
[6]. V. Kunin, M. Turqueti, J. Saniie and E.
Oruklu, "Direction of Arrival Estimation and
Localization Using Acoustic Sensor Arrays",
Journal of Sensor Technology, Vol. 1 No. 3,
2011, pp. 71-80
[7]. Zhang, R., Guang, J., Chu, F.H., Zhang,
Y.C.: ‘Environmental adaptive indoor radio
path loss model for wireless sensor
Networks localization’,Int. J. Electron.
Commun., 2011, 65, pp. 1023–1031
[8]. Larranaga, J., Muguira, L., Manuel, J.,
Vazquez, J.: ‘An Environment Adaptive
ZigBee based Indoor positioning
Algorithm’. Int. Conf, on indoor Positioning
and Indoor Navigation (IPIN), Zurich,
2010,pp. 15–17
[9]. J. Larranaga, L. Muguira, J.M.L. Garde, J.I.
Vazquez, “An environment adaptive
ZigBee-based indoor positioning algorithm”,
International Conference on Indoor
Positioning and Indoor Navigation (IPIN),
2010, pp 1-8
[10]. H. Koyuncu, S.H.Yang, “Indoor Positioning
with Virtual Fingerprint Mapping by Using
Linear and Exponential Taper Functions”,
IEEE International Conference on Systems,
Man, and Cybernetics 2013, pp 1052 – 1057
[11]. Wendell T.Hill ,Electromagnetic radiation
Wiley , 2009 , ISBN 978-3-527-40773-6.
[12]. T.S.Rappport. Wireless Communications:
principles and practise , Prentice-Hall
Inc.,New Jersey ,1996
[13]. http://www.jennic.com/jennic_support/appli
cation_notes/jnan-
1052_home_sensor_demonstration

Adaptive Indoor Localization by using Environmental Thresholding and Virtual Fingerprint Technique

More Related Content

What's hot (19)

Viewers also liked (16)

Similar to Adaptive Indoor Localization by using Environmental Thresholding and Virtual Fingerprint Technique (20)

Recently uploaded (20)

Adaptive Indoor Localization by using Environmental Thresholding and Virtual Fingerprint Technique