See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/4151006

A Friis-based Calibrated Model for WiFi Terminal Positioning

Conference Paper · June 2005

DOI: 10.1109/WOWMOM.2005.2 · Source: IEEE Xplore

CITATIONS READS

66 238

5 authors, including:

Frederic Lassabe Philippe Canalda

Université de Technologie de Belfort-Montbéliard Institut FEMTO-ST
22 PUBLICATIONS   262 CITATIONS    51 PUBLICATIONS   404 CITATIONS   

SEE PROFILE SEE PROFILE

Pascal Chatonnay Oumaya Baala

University of Franche-Comté Université de Technologie de Belfort-Montbéliard
63 PUBLICATIONS   262 CITATIONS    44 PUBLICATIONS   251 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Application of Machine learning techniques for Indoor Localization in 5G contexts View project

Indoor Positionning View project

All content following this page was uploaded by Oumaya Baala on 16 March 2017.

The user has requested enhancement of the downloaded file.

 A Friis-based calibrated model for WiFi terminals
positioning
Frédéric Lassabe∗ , Oumaya Baala† , Philippe Canalda∗ , Pascal Chatonnay∗ , François Spies∗
∗ LIFC - Laboratoire d’Informatique de l’Université de Franche-Comté
Numérica - Multimedia Developpement Center
Cours Louis Leprince Ringuet, BP 21126
25201 Montbéliard Cedex, France
Email: {frederic.lassabe,philippe.canalda,pascal.chatonnay,francois.spies}@pu-pm.univ-fcomte.fr
† SeT - Laboratoire Système et Transports
UTBM - Université de Technologie de Belfort-Montbéliard
Rue Thierry Mieg
90010 Belfort cedex, France
Email: oumaya.baala@utbm.fr

Abstract— Two types of applications use indoor posi-
tioning: services linked with mobility, such as guided
tour or meeting systems, and the active security of wire-
less network which locates intrusive unauthorized mobile
terminals. Indoor positioning cannot be managed by a
geostationnary system like GPS. In fact, current researches
are conducted to conceive indoor positioning using wireless
networks such as WiFi. In this paper, we study such a
mechanism and compare our accuracy results to other
solutions. This positioning function is combined to a
mobility prediction mechanism and constitute the mobility
service in a Video on Demand system called MoVie (Mobile
Video).
Keywords: positioning, mobility, wireless LAN.
I. I NTRODUCTION
With the spreading of WiFi networks [1], that enables
mobility and high data rate transfer, new problems arise.
Service continuity is a recurring problem related to
mobility. Indeed, mobile clients may want to change their
access point. In this case, the services provided would
be interrupted if nothing was done. A technique called
handover is used to prevent the services from being
interrupted. It is a mechanism allowing a mobile client
to change his access point transparently. The handover
needs to be planned in advance. Mobility prediction is a
good means to anticipate the handover.
Mobility prediction aims at determining where a mo-
bile terminal is more likely to be in the near future. To
do so, the terminal location is required. It is obtained
through a process called positioning. Several ways of
positioning exist including trilateration [2]. For trilat-
eration we need to know the coordinates of at least
three points and their distance to the mobile which
position must be determined. With the spreading of
mobile networks, the interest in location-aware services
has grown significantly. WiFi networks are commonly
implanted indoors, which raises new problems with
positioning. GPS [3] makes mobile terminal positioning
easy outdoors, whereas indoor positioning is seriously
limited.
In this article, we address the indoor positioning of
WiFi terminals. Indeed, positioning is the first and most
important step in mobility management. To obtain the
position of a mobile terminal, we use trilateration. It
is therefore necessary to compute the distance between
each access point and the mobile terminal. The main
topic of our article is the indoor distance determina-
tion within a heterogeneous environment for which we
choose to use an alternative to the Friis equation [4].
We will focus on a comparison of our technique with
the existing ones through precision tests.
Our work comes within the scope of the MoVie [5]
project. MoVie is a streaming platform for multimedia
contents. MoVie is structured into interoperable modules
(see fig. 1) where the following characteristics are gath-
ered:
• NetMoVie integrates the RTP/RTCP protocol. It
receives a few video sequence qualities and selects
the most adapted one depending on the current
situation.
• SysMoVie gathers ORB components and integrates
the hierarchy of video caches. The strategy of
 WebMoVie
Database / Frontend
Mixers Caches / mandataries
CLIENT
CLIENT
SERVER
SERVER
NetMoVie SysMoVie
Clients
CLIENT CLIENT
SERVER
SERVER
GeoMoVie
Positioning and Mobility Prediction
Fig. 1. The MoVie modules structure.
video cache management is specific to the particular
temporal data.
• WebMoVie represents a query interface of the
MoVie platform. It is the entry point of clients
where they are identified. A trader is created for
each query in order to localize one or more required
video sequences in SysMoVie.
• GeoMoVie tracks the mobile clients and antici-
pates their future moves. It contains positioning and
prediction modules. It also provides the handover
management.
In SysMoVie, a handover cache policy [6] ensures the
continuity of the multimedia flow. The defined cache
management aims at transferring parts of the video
sequences to the nearby caches. Thus it makes it possible
to avoid transmission interruptions. SysMoVie retrieves
information from GeoMoVie to determine which caches
the video sequences parts will be transferred to.
GeoMoVie is split into two components. The posi-
tioning component allows to determine a mobile current
position. The mobility prediction component then uses
the position to guess at the future positions of the mobile
terminal. Setting up GeoMoVie requires a learning pe-
riod. The mobility prediction part is based on the hidden
Markov model [7].
The remainder of this article is organized as follows.
In section II, we present the work related to mobile
positioning. In section III, we expound the context of
our work with the related problems and their potential
solutions. Section IV describes the tests of both the
related work and ours. In section V, we discuss the
possible uses of positioning in mobility prediction. In
section VI, we analyze the results of the tests and present
our conclusions.
II. R ELATED WORK
In this section, we present the major positioning
techniques. All of them use signal strength from the WiFi
peripherals to determine the geographic location of a
mobile terminal. Positioning techniques can be classified
into two main categories.
video The first one is based on a signal strength cartography
Servers
which can be done in two ways. One uses physical
measurement and the other uses a mathematical wave
propagation model to compute the signal strength values
of the map.
The second category of positioning techniques aims
at determining a relation between signal strength and
distance. Several approaches to obtain such a model
have been developed. It makes the location computation
possible using trilateration.
Firstly, we present the cartography-based category of
positioning techniques. The work [8] deals with both
measurement and calculation-based generation of the
signal strength map. Secondly, we present two tech-
niques from the wave propagation family. The first one
[9] uses polynomial regression to obtain a model. The
second one [10] is based on distance calculation through
the use of an alternative to the Friis equation which is a
free-space radio wave propagation basis.
A. Cartography-based positioning
Within the RADAR system [8], the mobile terminal
positioning uses a signal strength map of the covered
area. The signal strength map technique is split into
two steps. The first step is the generation of a signal
strength database. It can be created by measuring or by
computation. The second step uses this database to locate
mobile terminals.
1) Data collection: The first step in the signal
strength cartography-based method is data collection. It
requires all the computers (laptop or access points) to
be synchronized. At each point of the grid, the mobile
terminal broadcasts packets including its position and a
timestamp. Then, the geographic coordinates, the signal
strength space coordinates and the mobile orientation are
stored in a database. The signal strength space is a N -
dimensional space where N is the number of access
points used for positioning. The signal strength space
coordinates are N -uples. These N -uples elements are
signal strength measurements from each access point,
expressed in dBm1 .
1
The dBm is a unit used to quantify the power. It is expressed with
the following formula : PdBm = 10 log10 (PmW )
 2) Calculation-based signal strength map: The Floor
Attenuation Factor model from Seidel and Rappaport
[11] is used to compute the signal strength map. This
model is flexible enough to fit every building topology
while taking into account many sources of signal strength
loss. However, in [8] the model is modified by replacing
ground-related losses with obstacle-related losses. They
particularly address the signal strength absorption by the
walls.
The calculation database generation leads to a signal
strength map similar to the measurement-based one.
Thus, it is unnecessary to take measurements at each
point of the grid to obtain the database.
3) Database exploitation: When the signal strength
database is generated, it is used to locate mobile termi-
nals. A mobile requiring its location broadcasts a packet
to the access points within range. Each access point
then measures the signal strength. The signal sent by the
mobile terminal contains a timestamp so that the access
points compute the position of the mobile terminal at the
same time.
Then, the signal strength measurement from each
access point is compared with the reference points stored
in the database. Points selected in the database approx-
imately match the measure because of the variability
of signal propagation. The closest measurement sets are
returned. Selected reference points are ordered by Eu-
clidian distance2 in the signal strength space. The closest
point is more likely to be the solution. Neighbors of this
solution are considered to be potential solutions. The
geographic location of the mobile device corresponds to
the geographic coordinates of the chosen reference point.
Storing reference measures, builds a signal strength
cartography. This is an application from R2 to RN where
N is the number of access points used to locate the
mobile terminal. For each point defined by its geographic
coordinates (x, y), this application admits an image
(p1 , p2 , ..., pN ) where pi is the measured signal strength
from the access point i.
The access points used in this project were PC running
FreeBSD, equipped with WiFi cards configured as access
points. A two-dimensional positioning is considered.
The cartography-based positioning technique has a 2-
to-3-meter precision with a database generated by data
2
The Euclidian distance between two points A (xa0 , ..., xai ) and
B (xb0 , ..., xbi ) is calculated following this formula :
q drawback. A lot of data are required for the regression to
D = (xa0 − xb0 )2 + ... + (xai − xbi )2
collection. The test bed can be considered compact
(22.5 meters by 43.5 meters) and its layout is quite
homogeneous. It is composed of many offices separated
by walls. The precision is fairly good with this tech-
nique. However, there is a critical drawback: the lack
of dynamism. Indeed, the measurement-based database
generation, which leads to the best performances, is
compromised when the topology or the access points
layout changes. In such cases, the database must be
completely re-generated.
B. A polynomial regression-based propagation model
Wang, Jia and Lee [9] present a positioning technique
based on a radio wave propagation model. This model
aims at expressing the mathematical relation between
the distance from transmitter to receiver and the signal
strength.
First of all, the authors prove the possible use of signal
strength for mobile terminal positioning by measuring
signal strength for 24 hours. Measurements are taken
every 0.5 second. The result analysis shows that signal
strength is more stable at night. It is due to people’s
interferences at daytime. When working time fluctuations
are considered, the signal strength shows a standard
deviation of only 2.26 dBm, proving the possibility of
exploiting the signal strength in mobile positioning.
Then the authors take a series of measurements. The
results are placed in a graph where X-coordinate is the
signal strength and Y-coordinate is the distance. The
graph shows a group of points which can be math-
ematically expressed. The mathematical expression is
obtained by polynomial regression of a degree of 1 to
6. It appears that third degree polynomial regression is
the best in terms of precision computation ratio. Then,
the authors proceed with measurements to compare the
model to reality. At a point of unknown coordinates,
the signal strengths from the access points within range
are measured. The third-degree polynomial expression
is applied to the obtained values. The results are the
respective distances from the access points to the mobile
terminal. The distances allow to calculate the position of
the mobile terminal by using trilateration.
The advantage of this technique is the speed of posi-
tioning. All we need to do is apply the formula obtained
by regression. Then, a simple trilateration returns the
mobile terminal coordinates. However, there is a main
be accurate, which involves a high cost in measurement
time. On top of that, it is possible to be confronted
with singularities in the buildings where the positioning
technique is implemented.
C. Security and positioning
The article [10] deals with security issues. Its first
objective is to locate rogue mobile terminals and ac-
cess points which try to infiltrate a network through
its wireless part. Indeed, WiFi networks are subject to
severe security flaws. An access point can be set behind
the firewall of an Ethernet network, thus allowing client
terminals to connect to any machine without being fil-
tered by the firewall. Unauthorized clients may then gain
access to an enterprise network and steal confidential
data from it.
The authors take signal strength measurements at
many locations of many buildings. The results of these
measurements are used to establish a radio wave prop-
agation model. This model is based on the Friis rela-
tion. The Friis relation expresses the signal strength in
function of distance, in a free space environment. The
Friis-based model is adapted to fit the conditions of
implementation.
The Friis equation can be written as follows:
10 log10 (PR ) − 10 log10 (PT ) =
λ
10 log10 (GR )+10 log10 (GT )+20 log10 ( )−20 log10 (d)
4π
where PR and PT are respectively the received signal
strength and the transmission power. λ is the wavelength,
d is the transmitter-receiver distance and GR and GT are
respectively the receiver and transmitter antenna gains.
The authors consider the 20 coefficient, multiplied by
log10 (d). The value 20 is obtained by the conversion to
dBm. It was decided to change this 20 into a greater
value which takes into account the greater loss from
indoor environment. Indeed, the signal strength is in-
versely proportional to square distance. The increase of
the exponent applied to the distance implies that the
signal strength is more reduced for a given distance. It is
exactly what happens in radio wave indoor propagation.
Increasing the exponent in the Friis equation allows the
computation of a more precise signal strength according
to the distance. The reciprocal formula is then
λ
10 log10 (GR ) + 10 log10 (GT ) + 20 log10 ( 4π ) interfere with the radio wave propagation. The obstacles
log10 (d) =
10i
−10 log10 (PR ) + 10 log10 (PT )
+
10i
where i is the new floating point coefficient replacing 2 building and the nature of the obstacles is similar. When
(included in 20 when expressed in dBm as before) in the facing such unfavorable cases, the signal strength car-
Friis equation.
The precision observed is close to 2 meters. The main
advantage of this technique is its setup speed. In fact,
it is only necessary to supply the program with the
coordinates of the access points. Simply running the
program is enough to position a mobile terminal. How-
ever, some singular geographic points were observed
where the precision was worse than 8 meters. The main
drawback of this technique is the unique exponent used
in the Friis equation.
D. Related work analysis
We focus on 5 criteria : setup time, positioning time,
resources used, precision and reactivity. All are summed
up in the table I.
RADAR [8] Interlink SNAP
Networks [10] WPS [9]
Setup time Very long None Long
Positioning time Medium Short short
Resources used Many Few Few
Precision Very good Medium Good
Reactivity None Total None
TABLE I
R ECAP OF THE MAIN CHARACTERISTICS OF THE RELATED WORK .
The best precision is obtained by the signal strength
cartography-based technique. The other two techniques,
based on radio wave propagation, are precise but they
sometimes lead to singular points. However, it costs lot
of resources and computing time to use a signal strength
map. On top of that, it requires a long setup time and it
has no reactivity when topologic changes occur. These
drawbacks partially affect the polynomial regression-
based technique because of the need for data in order
to obtain the polynomial expression of the distance.
Although its precision is less accurate than that of the
previous technique, the technique based on an alternative
to the Friis equation has a main advantage, which is to
be very quick to setup and use. Thus, it is well adapted
to topologic changes.
Singular points are intrinsic with the topologic het-
erogeneity. Buildings are composed of obstacles which
can be of various natures [12] and their layout can be
irregular. By regular we mean that the distance between
the obstacles is approximately constant in the whole
tography shows better results because it fits the building
 Initial Measurements
Indoors Friis
equation
Calibrated
Regression
Reference
Real Time Measurements
Database
Search
Trilateration
Position
Mobility Prediction
Handover Cache Policy
Fig. 2. Positioning techniques in the MoVie project.
whereas the propagation model-based techniques con-
sider the topology uniformly.
The figure 2 shows the dependances between the
positioning techniques, the mobility prediction and the
handover cache policy within the MoVie project. It
presents the steps followed by each positioning technique
from the setup to the positioning and its application.
III. C ONTRIBUTIONS
The Interlink Networks [10] approach is chosen to
implement our positioning system. It has indeed the
advantages of speed and simplicity. It is interesting with
mobile terminals which have little computation power.
We explain the drawbacks of a uniform computation
in order to determine the distance according to the
signal strength. We first describe the common sources
of radio wave distortion and their predominance within
a heterogeneous environment. Second, we highlight the
radio wave distortion in indoor environment with the
help of our experiments and we test the model of [10],
to reveal its limits in a heterogeneous environment.
A. Common phenomena in radio wave propagation
Radio waves are mainly subject to two phenomena
affecting their signal strength or trajectory: the signal
strength loss through atmosphere, and an extension of
this phenomenon to every object gone through by the
signal.
1) Free-space propagation loss: The first source of
signal strength loss is related to the distance covered by
the signal. The signal strength loses power when going
through the atmosphere. In a free-space environment,
like our atmosphere, the loss can be determined by the
Friis equation:
PR λ 2
= GR GT ( )
PT 4πd
where:
• PR and PT are respectively the power available at
the receiving antenna and the power supplied to the
source antenna;
• GR and GT are respectively the receiver antenna
gain and the transmitter antenna gain;
• λ is the carrier wavelength;
• d is the transmitter-receiver distance.
The Friis equation expresses the signal strength loss
in function of the distance d. This loss depends on the
signal frequency f = λc . The carrier wave speed c in the
atmosphere is approximately c = 3.108 m.s−1 .
2) Radio wave absorption: The radio wave absorption
by the environment is due to the conversion of part
of electro-magnetic wave energy into kinetics energy,
involving a warming of the environment concerned. This
absorption is generally greater than the one bound to the
atmosphere.
3) Radio wave distortion sources: The radio waves
are affected by the presence of topologic components
altering the radio waves trajectory and therefore modify
the signal strength. The phenomenon we are more likely
to observe is wave reflection.
The most common sources of wave trajectory are
metal equipment that induces huge signal reflections,
preventing it from reaching areas theoretically within
range. Devices functioning at frequencies close to WiFi
frequencies also distort the signal by covering it with
great noise. This is for example a potential problem when
using a microwave between two WiFi peripherals.
4) Context: Our experiments take place in the context
of the Numérica building (figure 3). The Numérica3
building topology is heterogeneous. When studying the
first-floor map, it is possible to distinguish two homo-
geneous areas. First comes the corridor which runs all
3
The Numérica center is the building where ISTI and LIFC
(Laboratoire d’Informatique de l’Université de Franche-Comté) are
located.
along the floor and has no obstacle. Second there is the
area composed of offices, separated by partition walls
which alter signal strength. A load-bearing wall and a
partition wall separate both defined areas. Between these
two walls, there are electrical components and water
pipes. Together, they considerably weaken the signal
strength when the radio wave goes through them.
32,6 m
(0,0) 10.5 m
Fig. 3. Numérica center first floor (ISTI area) map.
This characteristic of the building is recurrent on all 3
floors. It makes signal strength loss more complex than
a simple distance-bound loss. A little move can indeed
drastically modify received signal strength. For example,
going through a door can involve the mobile being
hidden by a wall, therefore having the signal strength
received fall greatly with just a few steps.
On the map (see fig. 3), the triangles are the access
points. The dots are the calibration measurement points
and the crosses are the testing points. The origin of the
coordinates is the bottom left corner on the map. The
last access point is not represented because it is in a
near building which is not on the map.
In this section, we present various phenomena inter-
fering with radio wave propagation. These phenomena
are free-space propagation loss, absorption by physical
obstacles and radio wave trajectory distortion. Then, we
describe the topology of the Numérica building, which
includes all the types of obstacles and signal strength
losses previously exposed which makes a radio wave
propagation model hard to establish.
B. Conducted experiments
Two series of experiments were conducted. The first
one aimed at showing radio wave attenuation in function
of distance. The second one aimed at testing some of the
models of the literature previously mentioned in section
II.
1) Attenuation check: The first series of experiments
was conducted on the ground floor corridor of the ISTI4
area in Numérica. An access point was placed at the end
of the corridor. A laptop equipped with a WiFi PCMCIA
card was used to measure the signal strength at various
distances between the access point and the laptop. A
sample of the results is given in graph 4.
Signal Strength in function of Distance
-35
Back
Right
Left
-40 Front
-45
Signal Strength (dBm)
-50
-55
-60
-65
-70
0 500 1000 1500 2000 2500
Distance (cm)
Fig. 4. Signal strength in function of distance.
At each point, four measurements were made, one for
4
Institut des Sciences et Technologies de l’Information
 (x;y) Signal strength Real Computed
each cardinal point. It allows us to study the orientation Coordinates AP facing North distance distance
impact on the measured signal strength. (dBm) (m) (m)
Other measurements took place in the corridor of (4;5,4) 1 -61.4 20.6 18.56
the ground floor. Tests were carried out, placing the (4;5,4) 2 -54.6 4 8.54
(4;5,4) 3 -59.6 30.6 11.87
access point at the other end of the corridor. Various (5;9,1) 1 -56.2 16.6 13.18
access points and various PCMCIA cards were used. All (5;9,1) 2 -42.4 2.2 3.83
measurements produced about the same results. (5;9,1) 3 -59.8 31.8 12.02
The examination of the resulting curve underlines (3,8;12,6) 1 -55.6 13.2 12.67
(3,8;12,6) 2 -42.4 2 3.83
globally decreasing signal strength when the distance
(3,8;12,6) 3 -58.6 31 11.11
between the transmitter (access point) and the receiver
(laptop) increases. However, the decrease is not regular. TABLE II
Many singularities show up. Examining the map of A SAMPLE OF MEASUREMENTS AND DISTANCE COMPUTATION .
Numérica enables us to explain these phenomena.
It appears that the singularities are located at the
same distance from the access point as the building
components likely to distort the signal. It confirms the the first floor of a close building.
impact of the metal devices on the radio wave trajectory. The miscalculation of the distance is significant. It
Although the overall attenuation of the signal is can reach 20 meters in this example. Measurements at
checked, many singularities in the wave propagation are the other points produce similar errors. The experiments
observed. They even exist in free space propagation show that the indoor Friis equation [10] is not adapted
between the transmitter and the receiver. The topology to all types of construction work.
has a significant impact on the wave and taking it into
account is of primary concern in the realization of a C. Distance determination
positioning model. The first step in mobile positioning is to determine the
Moreover, the impact of the orientation is checked. It distance between the known points and the point to be
is however variable and difficult to quantify. Measure- located. To achieve this, a valuable approach is based on
ments from various points reveal that the variation of the Friis equation. An alternative to the Friis equation is
measurement is not the same for a given direction. In- used in function of the nature of the area crossed. This
deed, the study of the results shows that the measurement alternative is based on the partitioning of the territory
taken in front of the access point can be higher as it can into homogeneous zones. The calibration of the model
be lower than the one taken when turning back from the is performed for each homogeneous zone. At last, we
access point. This example is repeated for all the tested propose to calculate the distance by an expression linked
orientations, either presented to the right or the left, in with the access point.
the front or at the back. 1) Choice of the indoor Friis equation: The variation
2) Test of the alternative to the Friis equation [10]: of the propagation indice in the Friis equation makes
From now on, we call the alternative to the Friis equation it possible to express the impact of the topology on
the indoor Friis equation. The impact of the distance and the attenuation of the signal. The propagation index,
topology being checked, a series of tests for validation when applied to the distance, explicitly materializes the
of the positioning technique previously presented in [10] difference in the absorption of the signal according to the
was carried out. To perform the test, the alternative to medium it goes through. However, the alternative to the
the Friis equation is implemented in a C program. The Friis equation presented in [10] has drawbacks because
program makes signal strength measurements on the the medium varies from one building to another: the
available access points and calculates their distance to number of walls, spacing between the walls, the building
the laptop by applying the indoor Friis equation. material and other parameters have to be taken into
The measurements are taken according to 4 directions. consideration to determine the new propagation index.
There are 12 points of measurement : 6 in the offices and Indeed, when the propagation index, replacing the 2
6 in the corridor. The results are presented in table II. in the Friis equation, is calculated, it includes to some
There are 3 access points. Number 1 is the Netgear extent an average of the media which can be encountered
access point, the other two are from the Cisco vendor. within the building.
One of the Cisco access points, number 3, is located on Thus, the walls and all the other fixed obstacles in the
building are taken into account during the calibration of
the system.
2) Calibration of the Friis equation: The model pre-
sented in [9] is calibrated before use to put it in adequacy
with the building where it is implemented. According to
the authors of the indoor Friis equation, a coefficient
of 3.5 in the indoor Friis equation is applicable to any
environment of offices. However, a series of tests led
in Numérica prove that this assertion is not accurate
(see table II). The model has to be calibrated so that
its precision allows the use of the results in mobile posi-
tioning. Moreover, this calibration has to be made after
the partitioning of the building into homogeneous zones.
Afterwards, each zone has to be calibrated independently
of the others to obtain a suitable propagation index.
The calibration of the model induces a new cost
in time in the implementation of mobile positioning.
Several criteria of calibration have to be considered. The
first one is the calibration by a homogeneous zone. The
second one depends on the access point to which the
measurements are related.
Indeed, the medium surrounding the access points is
not always homogeneous. For example, an access point
can be in a nearby building and thus the transmitted
waves go through a street. In this case, the major part
of the medium they go through will be air, which means
a free-space medium. Consequently, its signal strength
will be different from the one transmitted by the in-
building access points at the same distance. Then, a
single propagation index for all the access points causes
a miscalculation of the distance according to the power
of the signal.
To calculate the applicable propagation indices in
the Friis equation, it is necessary to conduct measure-
ments. For each homogeneous zone, we have to measure
the power at several regularly distributed points. These
measurements allow the calculation of coefficients. The
calculation of the propagation index is done by reversing
the Friis equation in order to obtain an expression of
the propagation index in function of the parameters of
distance and power. The coefficient cjk is the coefficient
at the point k for the access point j . This coefficient is
calculated according to this relation:
−SSjk − K
cjk =
10. log10 (dj )
where:
• SSjk is the signal strength at the point k related to
the access point j ;
• K is a fixed value for each access point, where
K = −37 − PT , and PT is the access point current
output power ;
• dj is the real distance to the access point j , known
when processing the measurements.
From now on, it is considered that p points are mea-
sured to calibrate a zone. We also consider that there are
N available access points to locate the mobile terminal.
pN propagation indices can be calculated, with p indices
for each access point. The index associated with an
access point j for a given homogeneous zone z is the
average of propagation indices cjk calculated from signal
strength measurements in zone z . These propagation
indices are used in the indoor Friis equation which makes
it possible to estimate the distance between an access
point and a mobile terminal. We name calibrated model
a model being based on the calibration from a concrete
set of measurements.
IV. C ASE STUDY
We study the precision of the calibrated model within
the Numérica building. We have implemented a Java
applet which allows the display of points and circles
materializing the access points and the distances between
these points and the mobile terminal.
A. Analysis of the results
The measurements and calculations carried out allow
to analyse the precision of the studied models ( [9] and
[10]) and that of the calibrated model (part III-C.2).
1) Testing of the calibrated model: To study the
precision profit of the calibrated model, the alternatives
to the Friis equation were confronted with a new set
of measurements. Six new points were measured in
random directions. For the 6 points, calculations of the
distance and position were processed. Six points were
also measured on the ground floor in order to study the
impact of a change of topology within the same building.
The real coordinates and the measured power are
provided to the Java applet which computes the dis-
tances with the methods presented in [10] and [9] and
according to the calibrated model. Then, the position of
the mobile terminal is calculated by trilateration. The
error made during the positioning process is calculated
and displayed too. We use these results to analyze the
models.
2) Precision analysis: The measurement of the pre-
cision is done by computing the distance between the
calculated position and the real position where the
measurement was made. The first set of measurements
was carried out on the ground floor of Numérica. The in calculations and not to distribute them on the whole
measurements obtained are shown in table III. building by using an average.

Real (x;y) Error for the indoor Error for the V. A PPLICATION TO THE MOBILITY PREDICTION
coordinates Friis equation [10] (m) calibrated model (m)
(3,6 ; 12,3) 37,43 33,68
Once the distances between the access points and the
(9,5 ; 13) 44,06 30,05 mobile terminal are determined, it is possible to proceed
(1,6 ; 2) 30,79 43,48 to the mobility prediction. First of all, the geographical
(4,4 ; 19,7) 36 27,4 position of the terminal is calculated using trilateration.
(8,1 ; 23,8) 37,17 21,87
Then the computed data are exploited by the mobility
(3,4 ; 26,9) 27,9 23,44
prediction system. We briefly describe the trilateration
TABLE III technique. Finally, we quickly expound the mobility
P RECISION OF THE POSITIONING MODELS , GROUND FLOOR OF prediction system and its use within the MoVie project.
N UMÉRICA .
A. Determination of the position
The calculation of the terminal position is done by
In most cases, the calibrated model precision is better trilateration. Indeed, the distances between the access
than the indoor Friis equation one. The point, where points and the mobile terminal are known. Moreover,
the precision of the indoor Friis equation is better, is the access points are positioned at known coordinates.
placed in a stairwell behind a load-bearing wall. This One can thus deduce the position of the mobile terminal.
wall is not considered in the same homogeous zone for To do so, it is necessary to solve a system of equations
the use of the propagation index of two of the available whose solution is the mobile terminal position. In prac-
access points. It explains the highest degree of accuracy tice, because of the approximate distance, the solution
of the indoor Friis equation compared to the calibrated obtained is not unique. However, the position can be
model. In other cases, the calibrated model shows better approximated.
precision than the indoor Friis equation. B. Mobility prediction
The lack of precision of our model in the absolute is
Once we know the position of a mobile terminal at
explained by the use of the model calibration on the first
different periods, it is possible to predict its movements.
floor, with access points placed in different positions.
The prediction is split up into two stages. The first
The measurements on the ground floor do not correspond
stage is the installation of the prediction system after
to the model established for the first floor.
learning it from the peregrinations of various users on
The second set of measurements was taken on the first
the network.
floor and produced the results presented in table IV.
Our mobility prediction system is based on the hidden
Markov model [7]. Indeed, the position of the mobile
Real (x;y) Error [10] (m) Calibrated Error [9] (m)
coordinates model error (m) terminals is known with a degree of uncertainty. The
(1,6 ; 2,2) 15,7 2,72 20,44 observations are different from the physical states. We
(7 ; 3) 20,08 7,62 19,62 are confronted with a double stochastic process, to which
(8,6 ; 14,5) 24,08 16,86 23,89 the use of the hidden Markov model is well adapted.
(7,6 ; 25,2) 48,87 26,32 17.29
(3,6 ; 24,8) 30,6 25,06 6,58
The cells of the network are partitioned into under-
(12,4 ; 11) 36,95 16,56 48,86 cells corresponding to the zones bordering on the close
cells. Each of these under-cells is then broken up into
TABLE IV 4 states, each one being a direction (North, South,
P OSITIONING MODELS PRECISION , FIRST FLOOR OF N UMÉRICA . East, and West). Therefore, it is possible to increase
the precision of the model by taking its direction into
account without having to use several former states.
This series of measurements underlines the influence The training is done regularly to overcome the changes
of the building heterogeneity in Numérica on the pre- of the user’s mobility patterns. It is done from time
cision of positioning. All the methods used, even when to time. In the graph representing the states of the
calibrated for Numérica, have such bad precision that it is hidden Markov model, the training makes it possible to
clearly necessary to include precisely the fixed obstacles determine the probabilities of transition from one state to
another. When the probabilities are calculated, it is possi-
ble to exploit the prediction model. Various possibilities
exist to obtain a succession of states corresponding to
a succession of observations but we kept the Viterbi
algorithm.
C. Handover cache policy
As we said in the introduction, this work comes within
the scope of the MoVie project. In this project, the
strategy of video cache management is specific in the
way that it considers mobility. To load caches with video
sequence MoVie tries to predict which video will be seen
in the near future. In fact, we rely on a prediction of the
users position to prefetch the video they are watching in
the cache they will depend on in a near future. We name
this technique "handover prediction". We describe it in
detail in [6].
In this technique, we determine the position of a user.
Using this position and the history of the system, we
calculate the probability of the user’s presence in each
cache zone of the neighborhood. According to these
probabilities, we decide to load the cache serving the
zone or not. When we decide to load a cache with a
video sequence, we determine the amount of data to be
loaded. We have the choice between the base layer only
or the base layer and the enhancement layer (see [6] for
details).
Complete document
Partial document
0.4 0.1
0.9
.4 T =
a=0
0.1 0.2 0.07
a=0
.6 T
= 0.7
0.07 0.06
Hit rate Hit rate
Quality rate weighted by the quality
Bandwidth gain
Fig. 5. User in a corridor.
According to the simulations we made (see fig. 5),
the handover prediction to prefetch caches gives good
results balanced between bandwidth usage and reactivity.
Compared to a method which completely loads caches to
avoid cache miss, we have a low bandwidth usage and
compared to a method which does nothing to prevent
handover, we have good reactivity and we avoid stream-
ing failure. In figure 5, a is the threshold determining
partial prefetching and T is the threshold determining
the complete prefetching of the video sequence.
VI. C ONCLUSIONS AND PROSPECTS
The experiments we carried out led us to several con-
clusions. The observations also opened up new prospects
for us within the framework of mobile-terminal position-
ing.
A. Prospects
The analysis of the calibrated model tests shows that
the topology should be considered more precisely. On
top of that, the calibrated Friis equation is too sensitive
to the distance variation from the transmitter.
1) Taking into account the topology more precisely:
The relative inaccuracy of the topology in the places
where positioning is practiced is observed especially
when the signal strength is either very weak or very
strong. It is necessary to address the empiricism of
the calibration when the peregrinations of a mobile
terminal have to be tracked, especially in a hetero-
geneous environment. This is possible by determining
the signal strength along a way for which the effects
of the obstacles are taken into account. The effects
must be expressed mathematically and integrated into
the computation of the signal strength loss. Limited
(1) measurements could be taken in ideal conditions in order
to evaluate the various phenomena separately.
2) Variable indices in the Friis equation: Even when
calibrating the alternative to the Friis equation, the results
(2) can still include great errors. These errors come from
the determination of the index in the formula giving the
distance according to the signal strength:
−SS − K
log10 (d) =
10.i
where K is a value depending on the hardware proper-
ties.
We consider the possibility of determining the value of
the index used in the mathematical expression according
to the received signal strength. The phenomena radio
wave propagation is subject to are complex. Not all of
them can be taken into account because the generated
computations take too much time and are too frequent:
the positioning has to be almost instantaneous in the
perspectives of mobility prediction.
B. Conclusion
The study of radio wave propagation led us to the
examination of current and future positioning techniques.
Indoor positioning is rarely covered and implemented
whereas numerous positioning techniques exist for out-
door positioning. In an indoor environment, existing
model accuracy is insufficient or the systems are not
available to the general public.
The lack of indoor positioning systems led us to build
a positioning system whose bases are WiFi type of
networks and peripherals. The WiFi technology has been
chosen because of its availability to the general public
and its potential high data throughput.
During this survey, new problems arose. First of all,
the complexity of the topology impact on radio wave
propagation was revealed. It led us to the proposition of
a calibrated model adapted to heavy constraints. Then,
our method followed an experimental approach. It made
it possible to build a positioning system which we tested
and which we compared favourably to the other systems,
whether indoor or mixed.
The precision obtained allows us to consider applica-
tions of mobile digital guides requiring a precision of
about 1 meter. Our work comes within a more complete
project in which the positioning system proposed couples
well with a learning mobility prediction system.
Numerous experiments were carried out to build and
test the positioning system. A case study allowed us to
determine better precision than that of existing models.
However, current precision is not accurate enough for
the moment and needs to be improved.
[6] D. Charlet, P. Chatonnay, and F. Spies. Hand-over video cache
policy for mobile users. In J.B. Stefani, I. Demeure, and
D. Hagimont, editors, Proceedings of 6th IFIP International
Conference on Distributed Applications and Interoperable Sys-
tems (DAIS 03), volume LNCS 2893, pages 176–186, 2003.
[7] Lawrence R. Rabiner. A tutorial on hidden markov models and
selected applications in speech recognition. volume 77, pages
pp. 257–286. IEEE, February 1989.
[8] Paramvir Bahl and Venkata N. Padmanabhan. RADAR: An
in-building RF-based user location and tracking system. In
INFOCOM (2), pages 775–784, 2000.
[9] Y. Wang, X. Jia, and H.K Lee. An indoors wireless positioning
system based on wireless local area network infrastructure. In
6th Int. Symp. on Satellite Navigation Technology Including
Mobile Positioning & Location Services, number paper 54,
Melbourne, July 2003. CD-ROM proc.
[10] Inc Interlink Networks. A practical approach to identifying and
tracking unauthorized 802.11 cards and access points. Technical
report, 2002.
[11] S.Y. Seidel and T.S. Rappaport. 914 mhz path loss prediction
model for indoor wireless communications in multi-floored
buildings. In IEEE Transactions on Antennas & Propagation,
February 1992.
[12] B. Mac Larnon. TAPR’s Spread Spectrum Update, chapter
VHF/UHF/Microwave Radio Propagation : A Primer for Digital
Experimenters. Tucson Amateur Packet Radio Corporation,
1998.

ACKNOWLEDGMENT
We would like to thank Damien C HARLET for his
work on the handover cache policy.

R EFERENCES
[1] B.P. Crow, I. Widjaja, J.G. Kim, and P. Sakai. IEEE 802.11
Wireless Local Area Networks. IEEE Communications Maga-
zine, 35(9):116–126, September 1997.
[2] B.T. Fang. Trilateration and extension to global positioning sys-
tem navigation. Journal of Guidance, Control, and Dynamics,
9:715–717, December 1986.
[3] US Army Corps of Engineer, editor. Engineering and Design
- NAVSTAR Global Positioning System Surveying. Number EM
1110-1-1003. Department of the Army, Washington, DC, July
2003.
[4] L.V. Blake. Radar Range-Performance Analysis. Artech House
Radar Library, December 1986.
[5] Julien Bourgeois, Emmanuel Mory, and François Spies. Video
transmission adaptation on mobile devices. Journal of Systems
Architecture, 49:475–484, 2003.

View publication stats