Caso de Éxito

Caso de Éxito
Organización en la que se ha implantado el proyecto:
Universidad Politécnica de Valencia
TITULO: Diseño óptimo de la red inalámbrica y sistema automático de
reubicación de usuarios para balancear la carga de la red
Antecedentes/Problemática
Desde el inicio de la creación de tecnología inalámbrica para su uso en el entorno local,
tanto el sector empresarial como el académico han vislumbrado las grandes posibilidades
que ésta podía ofrecer. Tras la publicación del estándar IEEE 802.11 en 1997 y la creación
de la alianza Wi-fi en 1999 para promover, asesorar y certificar los productos de redes de
área local inalámbricos de alta velocidad, muchas empresas e instituciones han optado por
las redes inalámbricas de área local para permitir el acceso a su red de datos y a Internet.
En la universidad Politécnica de Valencia, desde el inicio de su planificación, se ha
perseguido el objetivo de proveer el 100% de cobertura en todos sus campus. Actualmente
cuenta con 575 APs para ofrecer servicio a más de 41.000 personas entre alumnos,
profesores, investigadores y personal de servicios. En la actualidad provee ampliamente
servicios de Telefonía, Televisión y video bajo demanda y se están incorporando sistemas
automáticos de reubicación de usuarios con el objetivo de balancear la carga de la red
inalámbrica y permitir mayor aprovechamiento del ancho de banda disponible. Además son
numerosas las aplicaciones que se pueden obtener a partir de los datos obtenidos de la
propia red inalámbrica. Un gran ejemplo de ello es el estudio de la cantidad de
desplazamientos entre los distintos edificios que existen dentro de toda la universidad.
Objetivos
El despliegue de una red inalámbrica grandes
dimensiones con múltiples edificios tropieza con
una serie de dificultades. Algunas se intuyen a
priori y otras se descubren durante el estudio
para su implantación. En cualquiera de los casos,
para poder desarrollar con éxito un proyecto de
tal envergadura es necesario contar con una
estrategia de trabajo correcta.
Estas estrategias que denominaremos “Directivas
de Estudio”, se deben aplicar en cada uno de los
edificios a estudiar y son las siguientes [1] [2].
• Estudio e inspección detallada del edificio,
tanto visualmente mediante un recorrido del
mismo, como con el estudio de los planos.
• Realización de una serie de medidas iniciales
para obtener la media de atenuación de señal por
pared.
• Realización de los cálculos para la obtención de
la distancia de cobertura de un punto de acceso
WLAN en función de las paredes que atraviesa
[3].
• Establecer el número de puntos de acceso
WLAN necesarios por planta y diseñar el mapa de
coberturas teóricamente.
• Situar los puntos de acceso en los lugares
elegidos y comprobar que la cobertura se ajusta
al diseño teórico calculado, verificando su
viabilidad.
• En caso que no se consiguiera la cobertura
estimada recolocar los puntos de acceso o añadir
más.
• Hacer una planificación de canales adecuada
para evitar interferencias, así como re-asignar
canales
adecuadamente
para
evitar
interferencias.
La Universidad Politécnica de Valencia tiene 3
campus principales, Campus de vera, Campus de
Alcoy y Campus de Gandia. Sólo en el Campus de
Vera se cuenta con 50 edificios repartidos en 2
kilómetros cuadrados.
Tras el despliegue de red, se han añadido
sistemas en la red que permiten aprovechar de
manera eficaz dicha red
Valores Añadidos: Sistema de estudio de la
movilidad en el Campus Utilizando redes
inalámbricas [7]
Desarrollo de un sistema de balanceo por
usuario para Telefonía IP en redes de área
local inalámbricas [4] [5]
Aprovechándonos de la tecnología inalámbrica, se
pueden obtener muchos beneficios. Uno de ellos
es el estudio de la movilidad de los usuarios en la
red cuando existe prácticamente una cobertura
total en todo el campus. En la Universidad
Politécnica de Valencia, a través de la red
inalámbrica diseñada en los 2 kilómetros
cuadrados del campus principal (Campus de Vera)
y en los campus de Alcoi y Gandia, hemos podido
observar cuales son los sitios más visitados,
detectar la mejor situación en caso de haber
puntos críticos, si las personas se tienen que
desplazar a lugares muy lejanos de sus
despachos, etc. Este sistema nos ha permitido
observar dónde se deberían localizar las
impresoras de red, servidores, salones de
reuniones, etc. e incluso añadir más puntos de
acceso debido a la afluencia de usuarios a un
lugar determinado. La movilidad de los usuarios
también
permite
saber
cual
es
su
comportamiento en cuanto a los lugares más
visitados o a múltiples desplazamientos entre
edificios lejanos. Además, el sistema permite ver
la movilidad de los profesores y alumnos entre los
diferentes campus. Uno de los mayores beneficios
que se obtienen es que se puede utilizar para
relocalizar servicios y departamentos dentro del
campus con el objetivo de evita demasiados
desplazamientos.
La telefonía IP en redes de área local
Inalámbricas se está haciendo cada vez más
popular debido a que permite un ahorro de coste
considerable. A pesar de que las redes de área
local inalámbricas de gran extensión, proveen
alta disponibilidad y redundancia, carecen de
suficiente ancho de banda cuando hay muchos
dispositivos conectados. En nuestra red estamos
implementando un sistema que reasigna la
conexión del dispositivo de telefonía IP
inalámbrico a otro punto de acceso cuando el
sistema detecta que éste está sobresaturado.
Hemos diseñado el algoritmo necesario para lleva
a cabo este sistema y las medidas obtenidas del
sistema muestran que se realiza de manera
óptima, en muy poco tiempo y transparente para
los usuarios sin haber consumo extra de ancho de
banda en la red. Se puede leer más sobre este
sistema desarrollado en la referencia [4].
Para el test y prueba de rendimiento del sistema
en la red de inalámbrica de la Universidad
Politécnica de Valencia se utilizó una centralita
de código abierto Asterix y teléfonos duales Wifi
(SmartPhones). Podemos decir que cada punto de
acceso soporta, teóricamente, 144 teléfonos IP,
utilizando la codificación de audio G.711 y por
tanto, utilizando el sistema de reubicación, todos
los estudiantes, profesores y personal de servicios
de la Universidad de Valencia pueden disfrutar de
dicho servicio. Dicho estudio se puede observar
en [5].
IPTV en la red inalámbrica de
Universidad Politécnica de Valencia [6]
la
Uno de los servicios multimedia ofertados en la
red de la Universidad politécnica de Valencia es
la televisión sobre IP. Este servicio funciona
correctamente sobre la red cableada, pero suelen
existir algunos problemas en la red inalámbrica,
sobre todo cuando los usuarios transitan a lo
largo del campus. Nosotros hemos propuesto una
solución basada en servidor para minimizar la
pérdida de paquetes y reducir la pérdida de
servicio cuando existe roaming entre puntos de
acceso en la red inalámbrica. Este sistema se
basa en la modificación de los protocolos
multicast para que los dispositivos elijan el mejor
punto de acceso multicast.
.
Referencias
[1] Jaime Lloret Mauri y Jose Javier López
Monfort. Despliegue de Redes WLAN de Gran
Extensión, el Caso de la Universidad Politécnica
de Valencia. XVIII. La Coruña (España). 10-12 de
Septiembre de 2003
[2] Jaime Lloret Mauri, Jose Javier López Monfort
y Germán Ramos. Wireless LAN Deployment in
Large Extension Areas: The Case of a University
Campus. Communication Systems and Networks
2003. Benalmádena, Málaga (España). 8-10 de
Septiembre de 2003
[3] Jaime Lloret, Jose J. López, Carlos Turró y
Santiago Flores. A Fast Design Model for Indoor
Radio Coverage in the 2.4 GHz Wireless LAN. 1st
International
Symposium
on
Wireless
Communication Systems 2004 (ISWCS'04). Port
Louis (Isla Mauricio). 20-22 de Septiembre de
2004.
[4] Miguel Garcia, Diana Bri, Carlos Turró, Jaime
Lloret. A User-Balanced System for IP Telephony
in WLAN. The Second International Conference on
Mobile Ubiquitous Computing, Systems, Services
and Technologies (UBICOMM 2008). Valencia
(España). 29 septiembre - 4 octubre de 2008.
[5] Miguel Edo, Miguel Garcia, Carlos Turro and
Jaime Lloret. IP Telephony development and
performance over IEEE 802.11g WLAN. The Fifth
International Conference on Networking and
Services (ICNS 2009). Valencia (España). 20-25 de
Abril 2009.
[6] Alejandro Canovas, Fernando Boronat, Carlos
Turro and Jaime Lloret. Multicast TV over WLAN
in a University Campus Network. The Fifth
International Conference on Networking and
Services (ICNS 2009). Valencia (España). 20-25 de
Abril 2009.
[7] Miguel Garcia, Sandra Sendra, Carlos Turro,
Jaime Lloret. People Mobility Study in a
University Campus using WLANs. The third
International Conference on Mobile Ubiquitous
Computing, Systems, Services and technologies
(UBICOMM 2009). Sliema (Malta). 11-16 de
Octubre de 2009.
DESPLIEGUE DE REDES WLAN DE
GRAN EXTENSION: EL CASO DE LA
UNIVERSIDAD POLITECNICA DE
VALENCIA
Jaime Lloret Mauri
Departamento de Comunicaciones
Universidad Politécnica de Valencia
e-mail : [email protected]
Abstract- This article deals with the issues related with the
deployment of wireless LAN (WLAN) of great extension.
Specifically, the studies and works developed towards the
set-up of the campus WLAN of the Universidad Politécnica
de Valencia are presented. The paper includes the
solutions taken for solving difficulties and provides a
structured method consisting on different phases which
has been applied as a working strategy in this work and
would be useful in future WLAN deployments. Colour
maps with the radio coverage of different buildings are
presented. Also wall absorption indoor and interference
between channels are discussed. The minimization in the
quantity of WLAN Access Points has been an important
premise in this work, in order to minimize budget and
interferences. Now, the university network is on the
installation phase according to the guidelines o this work
I.
INTRODUCCIÓN
Desde su aparición en el mercado las redes locales
inalámbricas (WLAN) basadas en el estándar 802.11b, [1]
[2], han experimentado un crecimiento de mercado
espectacular, tanto por las prestaciones que ofrecen como
por el bajo coste que tienen hoy en día los equipos de
transmisión. La instalación de una de estas redes en un
pequeño entorno doméstico o de oficina no supone una gran
complicación técnica, más allá de enchufar los equipos e
instalar el software necesario en los ordenadores a enlazar.
Sin embargo, cuando los requisitos exigidos a estas redes
aumentan, por ejemplo cubrir una distancia mayor o bien
proporcionar cobertura más allá de una vivienda o planta de
un edificio, el usuario se encuentra con ciertas limitaciones
técnicas que requieren un estudio detallado de la instalación
y que se escapan del sencillo concepto del ‘plug-and-play’ al
que está acostumbrado dicho consumidor de equipos
informáticos y de multimedia.
Sin embargo la complicación todavía pueda aumentar
más, cuando lo que se quiere es cubrir un área muy extensa
que incluye varios edificios así como espacios abiertos. Uno
de estos casos, es la cobertura completa de un campus
universitario. Sin duda es un caso muy interesante, dado el
interés que puede despertar a los habitantes de una
Universidad la posibilidad de conseguir cobertura de red en
José Javier López Monfort
Departamento de Comunicaciones
Universidad Politécnica de Valencia
e-mail : [email protected]
cualquier lugar del campus, para conectar su ordenador
portátil o incluso su ordenador de mano.
Internacionalmente estas redes inalámbricas estan muy
desarrolladas, existiendo proyectos muy interesantes,
fomentados a veces por ayuntamientos que quieren dotar de
cobertura a sus vecinos, como por comunidades de usuarios
que deciden establecer su propia red particular [3]. Existen
numerosas posibilidades en el uso de esta tecnología [4] que
no requiere licencia para operar.
Por todo ello resulta indispensable que las universidades
y en concreto la Universidad Politécnica de Valencia,
cuenten con ésta tecnología de futuro.
II. ESTRATEGIAS DE TRABAJO
El despliegue de una red de las dimensiones que se
pretende tropieza con una serie de dificultades. Algunas se
intuyen a priori y otras se descubren durante el estudio para
su implantación. En cualquiera de los casos, para poder
desarrollar con éxito un proyecto de tal envergadura es
necesario contar con una estrategia de trabajo correcta.
Estas estrategias que denominaremos “Directivas de
Estudio”, se aplicarán en cada uno de los edificios a estudiar
y son las siguientes.
• Estudio e inspección detallada del edificio, tanto
visualmente mediante un recorrido del mismo, como con
el estudio de los planos.
• Realización de una serie de medidas iniciales para
obtener la media de atenuación de señal por pared (se
detalla en el punto III).
• Realización de los cálculos para la obtención de la
distancia de cobertura de un punto de acceso WLAN en
función de las paredes que atraviesa.
• Establecer el número de puntos de acceso WLAN
necesarios por planta y diseñar el mapa de coberturas
teóricamente.
• Situar los puntos de acceso en los lugares elegidos y
comprobar que la cobertura se ajusta al diseño teórico
calculado, verificando su viabilidad.
• En caso que no se consiguiera la cobertura estimada
recolocar los puntos de acceso o añadir más.
III. PÉRDIDAS DE PROPAGACIÓN EN LAS PAREDES
Un efecto propio de los edificios que no ocurre en
exteriores, son las pérdidas de propagación por causa de las
paredes. Este efecto, junto con el multicamino, resultan muy
difíciles de evaluar y existen muchos trabajos publicados
sobre modelos matemáticos de propagación en interiores.
Sin embargo a efectos prácticos de diseño, podemos recurrir
a sencillos modelos estadísticos de absorción de las paredes,
para intentar predecir cuantas paredes puede llegar a
atravesar la señal WLAN sin perder la conectividad.
Existe otro aspecto ha tener en cuenta y que hemos
comprobado a lo largo de toda la campaña de medidas en el
campus. Consiste en el distinto comportamiento que
presentan las paredes de edificios diferentes. Esto es debido
a los diferentes materiales y/o técnicas constructivas que se
emplean en cada edificio. Es por ello, que en cada uno de los
edificios hay que estimar la atenuación de las paredes.
La técnica que se ha empleado consiste en localizar en el
edificio, a ser posible, una zona de paredes consecutivas
(generalmente un pasillo de despachos contiguos), fig. 2.
Una vez localizado se miden las atenuaciones de las paredes
situando trasmisor y receptor a un metro de la pared y a cada
uno de los lados. Se miden dichas atenuaciones y
posteriormente se calcula una media, que será la que no
servirá para atenuación del resto de paredes del edificio. La
tabla siguiente muestra un ejemplo de los resultados
obtenidos es uno de los edificios.
Fig. 1. Paredes consecutivas utilizadas en las medidas de la
atenuación media de las paredes de un edificio.
Pared
Pérdidas [dB]
0-1
7.98
2-3
6.04
4-5
7.16
6-7
2.34
8-9
0.03
10-11
2.36
12-13
4.62
Media
4.36
Con ayuda de la información obtenida y aplicando la
ecuación de propagación, es posible calcular una tabla que
proporcionará información sobre la distancia que se puede
alcanzar en función del número de paredes que se atraviesan.
1 pared 2 paredes 3 paredes 4 paredes 5 paredes
Distancia
(m)
85
60
49
42
38
Hay que señalar que durante las medidas se comprobó
que en los baños y sanitarios, la señal perdía mucha
potencia, habiendo casos en los que se llegó a perder hasta
incluso 20 dB. De acuerdo estas medidas, se concluye que
esto es debido a la absorción de las cañerías que atraviesan
las paredes en estos lugares. Por tanto, durante la realización
del diseño se deberá tener en cuenta que si estamos en una
zona alejada del punto de acceso que se supone debe dar
cobertura en un lugar, y justo antes tenemos unos sanitarios,
ascensores o escaleras, la experiencia nos indica que en estos
casos la señal recibida va sea de menor potencia que si fuera
una habitación normal.
La propagación de la señal entre plantas contiguas de un
edificio es muy baja, debido al forjado metálico que las
separa que actúa a modo de pantalla. Aunque si nos
encontramos justo encima del PA de la planta inferior o
viceversa se consigue recibir algo, la señal se atenúa
rápidamente en cuanto nos alejamos. Sólo se consigue un
buen paso de señal entre plantas si el edificio dispone de
patios interiores acristalados (deslunados, tragaluces) y el
PA se sitúa allí. Esta opción se ha ensayado con éxito en
edificios que reunían dichas condiciones ahorrando PAs,
pero no es posible aplicarla en la mayoría.
Por otro lado, el residuo de propagación entre plantas
puede dar lugar a interferencias entre canales en algunos
puntos, por lo que tendremos que diseñar un plan de
frecuencias adecuado, como se explica en el punto siguiente.
IV. INTERFERENCIAS ENTRE CANALES
El estándar 802.11b dispone de 13 canales dentro de la
banda ISM que se corresponden con 13 frecuencias de la
portadora para los países que adoptan la directiva de la ETSI
Sin embargo el ancho de espectro utilizado por cada canal se
solapa con los canales adyacentes causando interferencias.
Las interferencias son mayores cuanto más cerca estén los
canales. La siguiente tabla muestra el nivel de dichas
interferencias, clasificado en tres niveles.
Adyacentes
Canal 1 2 3 4 5 6 7 8 9 10 11 12 13
1
2
3
4
5
6
7
8
9
10
11
12
13
Canales que causan interferencias
Canales con riesgo a interferencias
Canales con poca o ninguna interferencia
Atendiendo a la tabla anterior, si tenemos que
seleccionar el máximo número de canales simultáneos sin
ninguna interferencia, usaremos los canales: 1 – 5 – 9 – 13.
Sin embargo si toleramos una ligera interferencia que en la
práctica no degrada el sistema usaremos: 1 – 4 – 7 – 10 – 13.
Por tanto esta segunda opción es mucho más eficiente ya que
proporciona un canal más, totalizando 5.
Para poder reutilizar estos canales, tendremos que
alejarnos lo suficiente para que la interferencia sea nula,
tanto desde un punto de vista horizontal como vertical. La
figura 2 muestra como distribuir 4 canales en un área
horizontal repitiendo canales para minimizar las
interferencias. Cada color representa uno de los 4 canales sin
interferencias entre sí. En la figura 3 se realiza lo propio,
pero teniendo en cuenta las plantas de un edificio.
Tras estos estudios previos y con los parámetros de
propagación, se evalúa si con un sólo punto de acceso
estratégicamente colocado se puede cubrir una planta del
edificio. Si resulta inviable se estudia con dos puntos de
acceso. En ningún caso se han necesitado tres.
Posteriormente al estudio teórico se comprueba in-situ la
cobertura calculada teóricamente con un PA de pruebas.
Cabe la posibilidad de que la señal no llegue tan lejos cómo
se esperaba, o que por lo contrario, ésta sea capaz de llegar a
puntos que en un principio y teóricamente no parecían
factibles. Esto se debe fundamentalmente al efecto del
multicamino, muy difícil de evaluar teóricamente. Si la
disparidad es muy grande, se puede plantear en función de
estas medidas, añadir un PA más en caso de falta de
cobertura o bien eliminar un PA en caso de sobrar.
Al finalizar el estudio de cada edificio se presentan dos
cifras de PA necesarios: una para asegurar una cobertura del
100% en toda el área y otra que viene dada por el intento de
economizar puntos de acceso, que asegurando una cobertura
del 95% reduce el número de PA necesarios en torno al
10%. En la figura 4 a) y b) se muestran las medidas finales
de cobertura para dos edificios del campus cubiertos por un
PA y dos PA respectivamente. Se puede apreciar que se ha
sacrificado la cobertura en algunas zonas extremas con el fin
de no incrementar el número de PA.
Fig. 2. Distribución óptima de 4 canales en un área horizontal.
planta 4
Respecto a los exteriores, también se han realizado
estudios de cobertura. En estos casos los cálculos son más
sencillos puesto que no se presentan obstáculos. En la figura
4 c) se muestra la cobertura del paseo central de la
Universidad. En estos casos se sustituye el monopolo
integrado del los PA por una antena exterior de mayor
ganancia. En este caso se ha solucionado empleando dos
antenas de cobertura sectorial
VI. CONCLUSIONES
planta 1
Fig. 3. Distribución óptima de 4 canales en un edificio.
V. RESULTADOS DE LAS MEDIDAS DE
COBERTURA
Siguiendo las directivas de estudio mencionadas en el
punto II, se han estudiado la totalidad de los edificios del
campus de Valencia de la Universidad Politécnica de
Valencia. En total son más de 50 edificios de múltiples
plantas que se extienden sobre un campus de unos 2
kilómetros cuadrados.
Como se describe en el punto II, la primera parte del
trabajo ha sido el estudio de los planos de los edificios y una
inspección de los mismos. Tras ello se han efectuado unas
pruebas de atenuación de paredes. Ha resultado sorprendente
durante el estudio comprobar la gran variabilidad que
presenta la atenuación de paredes entre edificios. El hecho
de que el campus de la UPV se haya ido construyendo a lo
largo de mucho tiempo y el que los materiales y las técnicas
constructivas hayan evolucionado a lo largo de este periodo
contribuye decisivamente a ello.
Mediante este trabajo se ha realizado un estudio
exhaustivo y completo de la cobertura de un campus
Universitario. Como resultado del mismo se obtienen los
lugares óptimos para el emplazamiento de los PA wireless y
mapas de cobertura de todos los edificios del campus, así
como de exteriores. Finalmente, se ha desarrollado un
método de trabajo que ha resultado satisfactorio y que podrá
contribuir positivamente en otros estudios similares en el
fututo. Actualmente se está en proceso de instalación de la
red en el campus de la UPV en base a este trabajo.
AGRADECIMIENTOS
Los autores quieren agradecer al Centro de Proceso de
Datos de la U.P.V. y en especial a Carlos Turró, por la
colaboración mantenida en este proyecto.
REFERENCIAS
[1] M.S. Gast, "802.11 wireless networks : the definitive
guide", Ed. O'Reilly, Sebastopol, 2002
[2] B. O'Hara, "The IEEE 802.11 handbook : a designer's
companion", IEEE Press, New York, 1999
[3] http://www.wirelessanarchy.com/
[4] http://www.gaips.upv.es/
a) Primera planta del edificio de la Biblioteca
b) Primera planta del edificio de la Escuela de Telecomunicaciones
c) Cobertura de exteriores. Paseo y Ágora.
–30 > S > -50 dBm
–50 >S > -70 dBm
–70 >S > -80 dBm
Fig. 4. Resultados de cobertura en diferentes edificios del campus
S < -80dBm
WIRELESS LAN DEPLOYMENT IN LARGE EXTENSION
AREAS: THE CASE OF A UNIVERSITY CAMPUS
Jaime Lloret, Jose J. Lopez, Germán Ramos
Department of Communications & Department of Electronics
Polytechnic University of Valencia
Ctra Nazaret-Oliva s/n, 46730 GRAO DE GANDIA
Spain
Abstract
This article deals with the issues related to the
deployment of wireless LAN (WLAN) of large
extension. Specifically, the studies and works developed
towards the set-up of the campus WLAN of the
Polytechnic University of Valencia are presented. The
paper includes the solutions used to solve difficulties and
provides a structured method consisting of different
phases which have been applied as a working strategy in
this work and would be useful in future WLAN
deployments. Color maps with radio coverage of
different buildings are presented. In addition, indoor wall
absorption and interference between channels are
discussed. The minimization in the quantity of WLAN
Access Points has been an important premise in this
work, in order to minimize budget and interferences.
Now, the university network is in the installation phase,
according to the guidelines of this work.
Key Words
WLAN, wall loss, radio coverage, 802.11
1. Introduction
Since the market appearance of the wireless local area
networks (WLAN) based on the 802.11b standard, [1]
[2], a spectacular market growth has been experienced,
due to the features they offer and the low costs of the
transmission equipment of today. The installation of one
of these networks in a little house or office environment
is very easy, and is not technically complex.
Nevertheless, when the requirements demanded from this
network increase, for example, to cover a greater
distance or more than one house or floor of a building,
the user has to face several technical limitations that
require a more in depth study of the installation.
However, this complication could increase when to cover
a vast area with several buildings and open areas inside is
needed. One example of this situation is a university
campus where is very interesting to the students. It brings
them the possibility of having a network connection
wherever they are, and use their laptops or handheldcomputers furnished with wireless cards.
Internationally, these wireless networks are widely
developed with interesting projects, sometimes promoted
by the city councils that want to offer coverage to its
neighbors, or by groups of users that want to establish
their own particular network [3]. There are many
possibilities for using this technology and a license is not
necessary to work with it.
2. Working strategies
The deployment of a network of this dimension is faced
with several problems. Some of them are easy to find a
priori for, but others appear during the specific
implantation study. To develop a big project like this
successfully, it is always necessary to use a correct work
strategy from the very beginning.
These strategies we will call “Study Directives”, and will
be applied to each building of the network. These
directives are:
• A visual study and inspection of the building through
walking around them, and using the plans.
• The carrying out of an initial set of measures to
obtain the mean attenuation of the signal across the
walls (this topic is developed at point 3).
• The carrying out of calculations to obtain the
coverage distance of a WLAN access point as a
function of the walls that it needs to cross (point 3).
• Establish the WLAN access point numbers needed by
the floor, and design the theoretical coverage maps.
• Locate the access points of the selected ones and
check that the coverage adjusts to the theoretical
design to validate its viability.
• If the estimated coverage is not obtained, replace the
access points or add some new ones.
3. Propagation losses in the walls
The three main mechanisms of radio propagation are
attributed to reflection, diffraction, and dispersion. These
three effects cause distortions in the radio signal that
suffers attenuation due to losses in its propagation [4].
Common effects on the buildings that do not appear in
free field are losses through the walls, roofs, and floors.
This effect, together with the multipath and diffraction
caused by the corners, is very difficult to evaluate, and
there are many publications about mathematical models
of indoor radio propagation [5][6][7]. The propagation
losses value is given in the next equation, where the
effect of the losses due to multipath effect is added [8]
[9]:
Gtx and Grx = Antenna gains (at the transmitting and
receiving sides).
20log(4π/λ) propagation loss at 1 meter in free field.
Lms = Multipath loss.
According to the equipment used (AP power rated at 16
dBm, and both antennas at 2 dBi), the power received at
1 meter from the AP is:
Prx1m = -40 dBm
(3)
The method employed for estimating wall absorption
consists of locating an area of consecutive walls (usually
a corridor of office rooms) in the building.
L(dB) = Lo + 10n log(d) + kF + IW + Lms
Lo = power losses (dB) at a distance of 1m (40 dB at
2.4 GHz frequency)
n = attenuation variation index with the distance
(n=2)
d = distance between transmitter and receiver
k = number of plants that the signal crosses
F = losses through the floors
I = number of walls that the signal crosses
W = wall losses
Lms = multipath effect losses
Nevertheless, from a practical design point of view, we
will use simple statistical models of the wall’s absorption
in order to predict how many walls the WLAN signal
will be able to cross whilst maintaining connectivity.
The reception power (applied to a wireless network of
these characteristics) is given by the following equation
when the propagation of the signal crosses i:
Pr = Ptxap + Gtx + Grx – 20log d – 20 log (4π/λ) – Σ Lpi
– Lms (1)
Pr = Received power
Ptxap = Transmitted power by the access point
Gtx = Transmitter gain
Grx = Receiver gain
d = Distance between transmitter and receiver
20 log (4π/λ) = 40 dB for 2.4 GHz
ΣLpi = Propagation losses due to the walls
Lms = Propagation losses due to multipath effect
The value of Lms has been estimated by means of field
measures, obtaining a value of between 12 dB and 20 dB.
In order to assure coverage, the worse case scenario
corresponding to 20 dB was taken in this work. Next, the
received power at a distance of 1 meter from the wireless
access point is calculated:
Prx1m = Ptxap+Gtx+Grx-20log(4π/λ) –Lms
Ptxap = Transmitted power by the access point
(2)
Fig. 1. Consecutive walls used in the measurements for
obtaining the mean loss through the walls of a building.
According to figure 1, the transmitter is located at a fixed
position ‘0’, 1 meter apart from the wall and a series of
measures are taken at points 1-3-5-…-13 using a wireless
card connected to a laptop, and signal monitoring
software. Using equation (1), (2) and (3), it is deduced:
Pr=Prx1m -20log(d) – ΣPi. So it is possible to calculate the
loss through the first wall.
L0-1 = -40 – 20 log (d) – Pr1
Pr1 is the received power at point ‘1’ and d=2 in this
case. In order to compute the loss at the second wall:
L2-3 = -40 – 20 log (d) – Pr3 – L0-1
Where Pr2 is the received power at point ‘3’ and d=4.5 in
this case (wall separation is 2.5 meters for all rooms).
After measuring the losses through each wall, a mean
value is computed. This mean value will be employed as
a value of reference for all the walls of this building. The
next table shows the values of attenuation obtained from
a specific building. As can be seen, not all of the walls
produce the same attenuation in spite of the fact that they
are made from the same materials. This variation is
produced by the multipath effect (which is unknown).
Obtaining a mean value the error caused by multipath
effect is reduced.
Wall
0-1
2-3
4-5
6-7
8-9
10-11
12-13
Mean
Loss [dB]
7.98
6.04
7.16
2.34
0.03
2.36
4.62
4.36
Next, we are ready to estimate the number of walls the
wireless signal can cross without loss of connection.
Below is the deduced expression for threshold power:
Pu = Treshold power
Lp = Loss per wall
n = number of walls crossed
The following expression can be used to obtain the
number of walls that the signal can cross within a
specific threshold power:
Prx1m − 20 log d − Pu
Lp
And this other expression provides the maximum
distance as a function of the number of crossed walls:
d = 10
Prx1 m − n · L p − Pu
20
Applying last equation, a table that shows the
maximum distance allowed as a function of number of
walls crossed by the signal is shown below. The power
threshold is fixed at -80 dBm, which is the typical
sensitivity value for the majority of commercial wireless
LAN cards (at a transmission speed of 11 Mbps).
1 wall 2 walls 3 walls 4 walls 5 walls
Distance (m)
85
60
1
7
2
5.25
3
6.5
4
6
5
8.20
Afterwards, a check on resemblances between the models
and the real measures taken in the buildings was carried
out. The validity of these models was confirmed. The
next table shows the results obtained in various
buildings:
Pu =Prx1m – 20 log (d) – n.Lp
n =
Model
Lp (dB)
49
42
38
It would be interesting to notice an irregular behavior in
the signal attenuation when crossing a wall next to a
toilet. Is this case, the loss caused by these walls is
significantly greater than that caused by the common
walls of the building. This loss can be as great as 20 dB.
The pipes embedded in the walls of these pieces probably
cause this behavior. Consequently, special treatment of
these walls can be taken in the computation of coverage
in the building. As a rule of thumb placement of access
points in the proximities of toilets must be avoided, so a
big area of poor coverage will be created on the other
side of the toilet.
It is understandable that different buildings built in
different periods and with different building techniques
will have different wall attenuation characteristics.
Consequently, the study of wall loss would be made in
each building of the campus. However, in order to save
field measures in all the campus buildings, they have
been classified according to the building techniques and
material employed for construction. After some
fieldwork, five different models have been established.
All the buildings are matched with one on these models:
Building
Sports
DCAN
EUITI
I3
DOEFFC
5I
5D
5E
1B Architecture
1E Computer Sc. Faculty
1F DSIC
1G EUI
ETSA Block 2
ETSIA Block 3
3A - 2E
3M Fine Arts
Mean value
Mean
7
5.25
6.41
6.12
6.47
6.33
6.53
6.6
5.83
6.81
7.35
6.32
4.66
5.62
5.19
8.21
6.29
Model
1
2
3
4
3
3
3
3
4
1
1
3
2
4
2
5
The mean value has also been obtained. It can be used in
case of unknowing the model classification of the
building. After obtaining the loss of signal power due to
the wall, next step is to examine the building plane to be
covered with access points. With the obtained data the
wireless coverage can be designed from the far away
points on the map. As shown in figure 4a, the
intersection between the cover zones designed will be the
zone where the access point would be installed.
The propagation signal between adjoining floors in a
building is quite low, due to metal wrought between that
acts as a front. Although if you are exactly above the AP
of the inferior plant or vice versa, it is possible to receive
a slight signal, and the signal attenuates quickly as soon
as you go away. A good sign is only achieved between
plants if the building has crystal interior patios (crystal
skylights) and the AP is located there. This option has
been carried out with success in buildings that gathered
these conditions, saving APs, but it is not possible to
apply it to most of them. As an example of this, the
building of the Higher Polytechnic School of Gandia
(figure 4b), where advantage has been taken of the
crystal skylights to minimize the number of access
points.
On the other hand, the residual propagation between
plants can cause interferences between channels in some
points, so an appropriate plan of frequencies is needed,
and it is explained in the following point.
4. Interferences between channels
The 802.11b standard (in ETSI countries) provides 13
channels inside the ISM band, which belong to 13
frequencies between 2412 MHz and 2472 MHz.
However, the spectrum width used by each channel is
overlapped by the adjacent channels, causing
interferences. These interferences are higher in closer
channels. The following table shows the level of
interference classified in three levels.
Adjacent
1 2 3 4 5 6 7 8 9 10 11 12 13
Ch
1
2
3
4
5
6
7
8
9
10
11
12
13
Interferences between channels
Risk of interferences between channels
Few or no interference between channels
Concerning the previous table, in order to select the
maximum number of simultaneous channels without any
interference, the channels: 1 - 5 - 9 – 13 must be used.
However, if a slight interference that does not degrade
the system in the practice is tolerated the channels 1 - 4 7 - 10 – 13 can be used. This second option provides one
usable channel more, 5 in total. To be able to reuse these
channels, you would have to go far away enough to have
no interference, from the horizontal and the vertical
points of view. Figure 2 shows how to distribute 4
channels over a horizontal area, repeating channels to
minimize interference. Each color represents one of the 4
channels without interferences between them. In figure 3
is demonstrated the same as before but taking into
account the floors of the building.
5. Outdoor coverage study
Outdoors, it should be considered that you could have
different types of antennas (directional, omni-directional,
etc.) to offer the highest coverage. Once the different
radiation antenna diagrams have been studied, the
position, the height, and the angle of inclination would
be
Fig 2. An optimal 4-channel distribution over a horizontal area.
Floor 4
Floor 1
Figure 3. Optimal 4-channel distribution in a building.
chosen to install and cover the desired area. In this case
the antennas were placed where it was necessary to have
good coverage. To find the corresponding mathematical
calculation, the following equation can be used:
Perceived = Ptxap + Gtx + Grx – Lprop
Ptxap = Power transmitted by the AP.
Gtx = Transmitter gain.
Grx = Receiver gain.
Lprop = 20log(4πd/λ) propagation losses
6. Measurement results
Despite there are available WLAN design software [10],
we used our own strategic following the study's directive
mentioned at point 2. All the buildings on the campus of
the Polytechnic University of Valencia have been
studied. There are more than 50 buildings with quite a lot
of plants spread over two square kilometers. The first
part of this work consisted on measure walls attenuation
and the second one was the study of it in the planes of
the buildings. During the study, it has been interesting to
contrast the different wall attenuation values among
buildings. The campus of the Polytechnic University of
Valencia has been built over different time periods and
the materials and the constructive techniques that have
evolved along this period have affected decisively in the
development of this work.
After these previous studies, and with the propagation
parameters obtained, it is checked if only one AP can
cover all the floor of the building. If it is not possible, it
is tried using two AP.
Afterwards, the covering, theoretically calculated, is
tested in situ through a test of AP. It is possible the signal
does not to arrive as far as expected, or otherwise, it is
possible that it arrives at areas that initially and
theoretically did not seem feasible. It is due to the
multipath effect, and it is very difficult to evaluate
theoretically. If there is a high disparity between design
and in situ measures, you can add or remove one AP if it
is needed. In the final report of each building, two results
are presented regarding the number of AP used: the first
one ensures 100% coverage of the whole area, and the
second one economizes the number of access points
assuring a 95% coverage and reducing the number of
necessary AP to around 10%. In figure 4c and 4d
measurements for the final coverage of two building
using one and two AP are shown. It can be appreciated
that in some extreme areas coverage has been sacrificed
with the purpose of not increase the number of AP.
Outdoor coverage studies have also been carried out. In
these cases, calculations are easier due to the fact that
obstacles are not presented. In figure 4e coverage of the
central walk of the University campus is shown. In these
cases, the integrated monopole of the AP is substituted
by more external antennas. It has been solved using two
sectorial antennas in particular, although in figure 4e the
alternative of a single omni-directional antenna is also
shown.
7. Conclusions
In this work, a complete study of the coverage of a
University campus has been developed. As a result of
this study, the optimal locations for the wireless AP are
obtained, and indoor and outdoor coverage maps for all
the buildings of the campus are drawn.
Finally, a working method has been developed providing
quite satisfactory results. It could be able to contribute
positively in other similar studies in the future. At the
moment, the wireless network in the campus of the
Polytechnic University of Valencia is in the deployment
phase following the guidelines of this work.
8. Acknowledges
The authors want to thank the Computer Center of the
Polytechnic University of Valencia and especially to
Carlos Turró, for his collaboration maintained throughout
this project.
References
[1] M.S. Gast, "802.11 wireless networks: the definitive
guide", Ed. O'Reilly, Sebastopol, 2002
[2] B. O'Hara, "The IEEE 802.11 handbook: a designer's
companion", IEEE Press, New York, 1999
[3] http://www.valenciawireless.org
[4] http://www.sss-mag.com/indoor.html
[5] C.C. Chiu and S.W. Lin, "Coverage prediction in
indoor wireless communications," IEICE Trans.
Commun.,vol. E79-B,no.9,pp. 1346-1350,Sep. 1996
[6] W.C. Chang, C.H. Ko, Y.H. Lee, S.T. Sheu, Y.J.
Zheng,"A Novel Prediction System for Wireless
LAN Based on the Genetic Algorithm and Neural
Network", Proc. IEEE 24th Conference on Local
Computer Networks, Oct. 1999, Lowell, Ma, USA
[7] R. A. Valenzuela, "A Ray Tracing Approach to
Predicting Indoor Wireless Transmission", IEEE
Vehicular Technology Conference, Secaucus NJ,
May 18-20, 1993, 214 – 218
[8] http://www.sss-mag.com/pdf/1propagation.pdf
[9] José Maria Hernando Rábanos “Transmisión por
radio”, Editorial C.E. Ramón Areces S.A., 1997
[10] S.J. Fortune, D.M. Gay, B.W. Kernighan, O.
Landron, R. A. Valenzuela, M.H. Wright, WISE
Design of Indoor Wireless Systems, IEEE
Computational Science and Engineering, 2, 1, pp. 5868 (Spring, 1995).
a) Access point place choice.
b) Higher Polytechnic School of Gandia lower and first floors of A building
c) Library. First floor
d) Higher Technical School of Telecommunications Engineering building. First floor
e) Outdoor coverage. Walkway and Agora with 2 directive antennas and with 1 omni-directional antenna.
–30 > S > -50 dBm
–50 >S > -70 dBm
–70 >S > -80 dBm
Fig. 4. Covering results in different campus buildings
S < -80dBm
A Fast Design Model for Indoor Radio Coverage in
the 2.4 GHz Wireless LAN
Jaime Lloret, Jose J. López, Carlos Turró, Santiago Flores
Department of Communications
Universidad Politécnica de Valencia, Camino Vera s/n 46022 Valencia, SPAIN
[email protected]
Abstract—In this paper, an empiric radio coverage model for
indoor wireless LAN is presented. This model has been tried out
in a vast series of large extension buildings obtaining successful
results. The objective of the model is to facilitate the radio design
work of a wireless LAN by means of straightforward
calculations, because the use of statistical methods is very time
consuming, and difficult to put in practice at most situations. Our
model is based on a derivation of the free field propagation
equation taking into account the building structure and its
materials and it has been tested on a large scale design of a
WLAN network of over 400 access points. Finally, the paper
includes the guidelines to setup a home, building, University
campus, or city wireless LAN easily in three simple steps.
Key Words-WLAN, wall loss, radio coverage, 802.11, 2.4 GHz
I.
INTRODUCTION
Since the launch of the wireless local area networks
(WLAN) based on the 802.11b and 802.11g standard, a
spectacular market growth has been experienced, due to the
features offered and to the low costs of the necessary
transmission equipment. Installation of one of these networks
at home or in an office environment is straightforward, and
technically affordable [1] [2] [3].
Nevertheless, when the requirements demanded to this
network increase, for example, to cover a great area, more
than one house or different floors in a building, the user has to
face several technical limitations that require more in depth
study of the installation. Additionally, this complication could
increase a bit more when the full coverage a vast area with
several buildings is needed. In this case, statistical models and
ray tracing based methods are not always affordable.
Furthermore, results of analytical and statistical methods can
be unimplementable due to the real structure of the building
and even to aesthetical considerations.
On the other hand, wireless networks are widely
developed around the world with interesting projects,
sometimes promoted by the city councils that want to offer
coverage to its neighbors, or by groups of users that want to
establish their own particular network. There are many
possibilities for using this unlicensed radio band technology,
and simple models like the presented one, will be useful for
these tasks.
This paper is structured as follows: Section 2 describes the
general approach of our model. Section 3 details the
calculations involved. Section 4 is devoted to the field testing
and finally section 5 concludes the paper.
II.
GENERAL APPROACH
The deployment of vast extension WLAN is faced with
several problems. To develop a big project successfully, it is
always necessary to use a smart work strategy from the very
beginning that could minimize the design effort.
We call that, “Design steps”, and will be applied to each
building of the network. These steps are:
1) Visual study and inspection of the building through
walking around them, and using its floor plan.
2) To carry out an initial set of measures to obtain an
estimate of wall propagation losses.
3) Try a suitable location for each access point and carry
out the calculations in order to estimate the coverage
across the building, as a function of the obstructing
walls in the light of sight and the distance to the access
point.
Step three is then repeated until a suitable location for all
access points has been reached. Fig. 1 illustrates this
approach.
III.
WALL AND FLOOR PROPAGATION LOSSES
The three main mechanisms of radio propagation are
attributed to reflection, diffraction, and dispersion. These three
effects cause distortions in the radio signal that suffers
attenuation due to losses in its propagation [4]. Common
effects on the buildings that do not appear in free field are
losses through the walls, roofs, and floors. This effect,
together with the multipath and diffraction caused by the
corners, is very difficult to evaluate, and there are many
publications about mathematical models of indoor radio
propagation [4] [5] [6].
The propagation losses value is given in equation (1),
where the effect of the losses due to multipath effect is added
[7] [8]:
L(dB) = Lo+10n log(d)+ΣKiFi+ΣIjWj+Lms
Where:
(1)
Figure 2. Consecutive walls used in the measurements, and check points for
obtaining the mean loss through the walls of a building.
Figure 1. Design step 1: Initial set of measures to estimate wall propagation
losses.
Lo = power losses (dB) at a distance of 1m (40.2 dB at
2.44 GHz frequency)
n = attenuation variation index with the distance (n=2)
d = distance between transmitter and receiver
Ki = number of floors of kind i in the propagation path
Fi = attenuation of one floor of kind i.
Ij = number of walls of kind j in the propagation path
Wj = attenuation factor of one wall of kind j
Lms = Propagation losses due to multipath propagation
and light of sight interferences effect.
There are a lot of studies devoted to the characterization of
propagation losses through building materials (walls [10] [11]
and floors [12]) or studies focused to solutions for specific
buildings [13] [14]. It has also been demonstrated the
dependence of the attenuation respect to the angle of
interception with the wall or floor [15][16].
Nevertheless, from a practical design point of view, we
will use simple statistical models of the wall’s absorption in
order to predict how many walls the WLAN signal will be
able to cross whilst maintaining connectivity. As shown on
section four, this simple method will provide fairly good
estimations.
So, as stated in our general approach, we will begin by
measuring the propagation loss of each type of wall in our
environment. To do this, we will locate an area of consecutive
walls (usually a corridor of office rooms) in the building, as
shown in Fig. 1.
According to Fig. 2, the transmitter is located at a fixed
position ‘0’, 1 meter apart from the wall and a series of
measures are taken at points 1-3-5-…-13 using a wireless card
connected to a laptop, and signal monitoring software. Using
equation (1) it is possible to calculate the loss through the first
wall.
L0-1 = Lo – 20 log (d) – Pr1 + Lms1
(2)
multipath propagation losses we compute a mean value. This
mean will be employed as a reference for all the walls of this
type. Next table shows the values of attenuation obtained from
a specific building. As can be seen, not all of the walls
produce the same attenuation in spite of the fact that they are
made from the same materials. This variation is produced by
the multipath propagation and light of sight effects. By
obtaining a mean value we reduce that source of error. Results
can be seen on TABLE I.
TABLE I.
Wall losses, and mean value.
Wall
0-1
2-3
4-5
Mean (Lp)
It is notable to remark that we obtain a value adjusted to
that obtained on references [10] and [11], but using a very
simple process.
Now, we are ready to estimate the number of walls the
wireless signal can cross without loss of connection. The
deduced expression for the threshold power follows:
Pu =Prx1m – 20 log (d) – ΣLipi
(3)
Where Lo = 40.2 dB and Pr2 is the received power at point
‘3’ and d = 4.5 meters in this case (wall separation is 2.5
meters for all rooms). So, to obtain a suitable value without
(4)
Prx1m = Power received at 1meter
d = Distance between the transmitter and the receiver
ΣLipi = Propagation losses due to the walls of kind i
The following expression can be used to obtain the
number of walls that the signal can cross within a specific
threshold power:
n=
Prx1m − 20 log d − Pu
Pr1 is the received power at point ‘1’ and d = 2 meters in
this case. In order to compute the loss at the second wall:
L2-3 = Lo – 20 log (d) – Pr3 – L0-1 + Lms2 - Lms1
Loss [dB]
-1.021
15.352
2.802
5.711
∑L P
i i
(5)
And this other expression provides the maximum distance
as a function of the number of crossed walls:
d = 10
Prx1 m − n ·
∑ L i Pi − Pu
20
(6)
Figure 3. Design step 2A: Attempts for a suitable access point location by
estimation of the coverage across the building.
As an example of application, the TABLE II shows the
maximum distance allowed as a function of the number of
same type walls crossed by the signal. The power threshold is
fixed at -80 dBm, which is the typical sensitivity value for the
majority of commercial wireless LAN cards (at a transmission
speed of 11 Mbps).
TABLE II. Maximum distance allowed as a function of traversed walls.
1 wall 2 walls 3 walls 4 walls
Distance (m) 50.64 14.19
13.60
7.05
It would be interesting to notice an irregular behavior in
the signal attenuation when crossing a wall next to a toilet. Is
this case, the loss caused by these walls is significantly
greater than the caused by the rest of the walls in the building.
This loss can be as great as 20 dB. The pipes embedded in the
walls of these pieces probably cause this behavior.
Consequently, special treatment of these walls can be taken in
the computation of coverage of the building. As a rule of
thumb placement of access points in the proximities of toilets
must be avoided, so a big area of poor coverage will be
created on the other side of the toilet (as it is shown in Fig. 3).
Additionally it must be taken into account the metallic
objects (rails, fences, statues, etc.) in the direct path, in these
cases the measure suffers errors above +/-2 dB.
Now, the building map and the wall losses mean value are
only needed in order to design the coverage area as it is
shown in Fig. 3 and Fig. 4.
IV.
PRACTICAL VS THEORETICAL
MEASUREMENTS
All the buildings on the campus of the Polytechnic
University of Valencia have been studied. There are more than
50 buildings with quite a lot of floors spread over two square
kilometers.
It is understandable that different buildings built in
different periods and with different building techniques will
have different wall attenuation characteristics. Consequently,
the study of wall loss would be made in each building of a
extended area. So, in order to save field measures in all the
campus buildings, they have been classified according to the
building techniques and material employed for construction.
A check on resemblances between the models and the real
measures taken in the buildings was carried out. The validity
Figure 4. Design step 2B: Attempts for a suitable access point location by
estimation of the coverage across the building.
of these models was confirmed. The TABLE III shows the
results obtained for standard brick walls in different year-ofconstruction buildings:
TABLE III. Attenuation mean value for different campus buildings.
Building
Sports
DCAN
EUITI
I3
DOEFFC
5I
5D
5E
1B Architecture
1E Computer Sc. Faculty
1F DSIC
1G EUI
ETSA Block 2
ETSIA Block 3
3A – 2E
3M Fine Arts
Mean value
Mean
7
5.25
6.41
6.12
6.47
6.33
6.53
6.6
5.83
6.81
7.35
6.32
4.66
5.62
5.19
8.21
6.29
A global mean value has also been obtained. It can be used
in case of unknowing the model classification of the building.
After obtaining the loss of signal power due to the wall,
next step is to examine the building map to be covered with
access points. With the obtained data, the wireless coverage
can be designed from the far away points on the map. As
shown in Fig. 5, the intersection between the cover zones
designed will be the zone where the access point would be
installed, Fig. 6.
Now, in order to validate the model, a series of field
measures has been taken in different points in the building and
compared with the predicted ones using our model. As it is
shown in TABLE IV, the error is comprised in the range
between +/- 3dB, which should be enough for the majority of
the situations. The standard deviation of the error is comprised
between 1.6 and 1.9 dB, depending of the building. As
illustration example, the measures obtained for two typical
buildings between the 50 studied are shown.
Figure 5. Selection of the access point placement taking into account wall
losses.
To assure a suitable reception for all wireless cards, we
limit the power range at -80 dBm for 802.11b, and -68 dBm
for 802.11g, so measures further from that point are not taken
into account.
The propagation signal between adjoining floors in a
building is quite low, due to metal wrought between that acts
as a front. Although if you are exactly above the AP of the
inferior plant or vice versa, it is possible to receive a slight
signal, and the signal attenuates quickly as soon as you go
away. A good sign is only achieved between plants if the
building has crystal skylights and the AP is located there. We
have used our model in a 3D fashion to get a good estimate on
these cases.
On the other hand, the residual propagation between plants
can cause interferences between channels in some points, so
an appropriate 3D frequency planning is needed.
To conclude, we present in Fig. 7 an example of the
prediction maps obtained using the proposed model.
Figure 6. Access point placement and coverage area.
TABLE IV. Model prediction, compared with field measure and error, for two
different buildings. Building A
Model predict.
-63.88
-66.61
-76.19
-68.79
-79.62
-79.62
-77.24
-69.04
-73.89
-67.54
-77.84
-76.89
-77.00
-66.77
-68.27
-64.90
-68.44
-76.53
-76.95
Measured
-62
-67
-79
-68
-79
79
-75
-71
-71
-64
-79
-74
-79
-68
-66
-64
-67
-77
-74
Model predict.
-70.28
-76.19
-77.72
-75.16
-67.27
-73.73
-70.28
-74.61
-77.56
-74.57
-74.97
-63.93
-69.55
-77.47
Measured
-73
-76
-78
-74
-65
-74
-69
-77
-76
-77
-78
-65
-67
-75
Error
1.88
-0.39
-2.81
0.79
0.62
0.62
2.24
-1,96
2.89
3.54
1.16
2.89
2.00
1.23
2.27
0.90
1.44
-0.47
2.95
Building B
Figure 7. Example of the prediction maps obtained using the proposed model:
[-30,-50 dBm]
]-50,-70 dBm]
]-70,-80 dBm]
< -80 dBm
Error
-2.72
0.19
-0.28
1.16
2.27
-0.27
1.28
-2.39
1.56
-2.43
-3.03
-1.07
2.55
2.47
V.
CONCLUSIONS
An empiric radio coverage model for indoor wireless LAN
based on a straightforward derivation of the free field
propagation equation taking into account just the walls
traversed has been presented and validated.
This validation has been done extensively in different
buildings built in different periods and with different building
techniques. We have found that our model produces better
approximations when the wall attenuation factor of a building
is correctly selected from the database of constructive
materials collected during the project. Nevertheless, this is not
critical, and quite profitable and close to reality results can be
obtained using a general value.
It has been compared the predicted value respect to the
measured one. The conclusion is that our model produce
errors generally under +/- 3dB, with a standard deviation
around 1.8 dB.
An exhaustive and successfully campaign of field
measures, on more than 50 building in the campus of the
Universidad Politecnica de Valencia (Spain) guarantee its
usability and employment in future projects.
Our model has been used for the design of the wireless
deployment using the standards 802.11b and 802.11g at our
University campus. It is applicable to other Universities,
office buildings, etc.
ACKNOWLEDGEMENTS
We want to acknowledge all the students that participate in
the exhaustive field measurement campaign for their valuable
effort, and to the University Computer Center for their support
in this initiative.
REFERENCES
[1]
Jeffrey Wheat, Designing a Wireless Network, Syngress Publishing,
Rockland, 2001
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
M.S. Gast, 802.11 wireless networks: the definitive guide, Ed. O'Reilly,
Sebastopol, 2002
B. O'Hara, The IEEE 802.11 handbook: a designer's companion, IEEE
Press, New York, 1999
C.C. Chiu and S.W. Lin, “Coverage prediction in indoor wireless
communications”, IEICE Trans. Commun.,vol. E79-B,no.9,pp. 13461350,Sep. 1996
W.C. Chang, C.H. Ko, Y.H. Lee, S.T. Sheu, Y.J. Zheng, "A Novel
Prediction System for Wireless LAN Based on the Genetic Algorithm
and Neural Network", Proc. IEEE 24th Conference on Local Computer
Networks, Oct. 1999, Lowell, Ma, USA
Rappaport, Theodore S., Wireless Communications: Principles and
Practice, Prentice Hall Publications, NJ, 1996.
R. A. Valenzuela, "A Ray Tracing Approach to Predicting Indoor
Wireless Transmission", IEEE Vehicular Technology Conference,
Secaucus NJ, May 18-20, 1993, 214 – 218
T. Frühwirth, J.R. Molwitz and P. Brisset, “Planning Cordless Business
Communication Systems”, IEEE Expert Magazine, Special Track on
Intelligent Telecommunications, February 1996
F. Agelet, A. Formella, J. Rabanos, F. de Vicente and F. Fontan,
“Efficient Ray-Tracing Acceleration Techniques for Radio Propagation
Modeling”, IEEE Trans Vehicular Tech, 49 #6 p. 2069
John C. Stein, Indoor Radio WLAN Performance Part II: Range
Performance in a Dense Office Environment, Harris Semiconductor
(Intersil)
Robert Wilson, Propagation Losses Through Common Building
Materials 2.4 GHz vs 5 GHz. Magis networks Inc.
http://www.magisnetworks.com/pdf/cto_notes/E10589PropagationLosse
s.pdf
D. Suwattana, J. Santiyanon, T. Laopetcharat, “Study on the
Performance of Wireless Local Area Network in a Multistory
Environment with 8-PSK TCM”, International Technical Conference
On Circuits/ Systems, Computers and Communications
J. S. Davis II, Measurements in Cory Hall at UC Berkeley.
http://www.wireless.per.nl:202/multimed/cdrom97/2_4ghz.htm
Dan Dobkin, Indoor Propagation and Wavelength, WJ Communications,
2002 http://www.wj.com/pdf/techpubs/Indoor_prop_and_80211.pdf
R.F. Rudd, “building penetration loss for slant-paths at l-, s- and cband”, International Conference on Antennas and Propagation, March
2003
Adi Shamir; “An Introduction to Radio Waves Propagation: Generic
Terms, Indoor Propagation and Practical Approaches to Path Loss
Calculations, Including Examples”, RF Waves - White Paper.
http://www.wtwo.net/testowo/rfwaves/
A User-Balanced System for IP Telephony in WLANs
Miguel Garcia1, Diana Bri2, Carlos Turró3, Jaime Lloret4
Polytechnic University of Valencia, Spain
1
[email protected],[email protected],[email protected],[email protected]
Abstract
Wireless IP telephony is becoming very popular
between the users of large Wireless Local Area
Networks (WLANs). This increment has been mainly
caused because it allows cost savings. Although, large
WLANs use to provide high availability and
redundancy, WLANs lack on the available throughput
when there are many IP telephony users. In this paper
we will present a system that reallocates IP telephony
devices when the system detects that an AP is
overloaded. The algorithm and the frames used for our
proposal are described. Measurements taken show
how the bandwidth consumed is transferred from an
access point to another when our proposal is running.
Finally, we will show the time needed to re-associate
remainder users.
1. Introduction
In last years, WLANs are reaching much popularity
by their main advantages: mobility, low cost
technology and large scalability [1]. When the
appeared, users only transmitted best effort
information. But, now, WLANs are used for many
types of traffic such as: data traffic, multimedia traffic,
telephony traffic, etc.
When we talk about telephony traffic in WLAN, we
identify this traffic with VoIP (Voice over Internet
Protocol) or IP telephony. IP telephony is defined as
transmission of voice and fax phone calls over a
packet-based IP data network [2]. This service is a
substitute for the fixed-line telephony service which
has become a commodity.
There are two fundamental attributes for users:
price and quality of service. These features are basics
to compete with fixed-line telephony.
Nowadays, IP telephony subscriber numbers don’t
stop increasing, 120% from 2005 to the first half of
2006, reaching to 155,401 users in June 2006 and so
on [3]. But most of IP telephony users are still more
innovative than potential users.
Although, this kind of telephony will be the most
used in the future, there is needed to improve some of
its features (e.g. the quality of service because the
delay is very critical in voice services).
In this work we will analyse the need of
reallocating remainder users that sometimes may exist
in large WLANs. Then, we will describe our proposal
and how it performs. We will show the evaluation of
our proposal in the WLAN of the Polytechnic
University of Valencia.
The remainder of the paper is structured as follows.
Section 2 describes some related works. The section 3
formulates the problem that we have when there are a
lot of users connected to an unique access point and it
is required a bandwidth to initiate an IP call. Our
proposal is explained in section 4. The performance
evaluation and the measurements of the system are
shown in section 5. Finally, in section 6, we conclude
the paper giving the benefits of our proposal.
2. Related works
Today, there are many works where the authors talk
about wireless IP telephony as a system that improves
the communications between users [4]. The IP
telephony and VoIP concepts are closed. In many of
the papers that we are going to see, the authors write
about VoIP, but the voice transmission mechanism
presented is used by the IP telephony.
When we talk about IP telephony and WLAN, first
we need to know whether it is possible the use of IP
telephony on IEEE 802.11 b/g wireless networks.
Theoretical studies shown in [5] [6] give that in most
cases these types of wireless networks meet the
requirements of IP telephony. On the other hand, there
are some practical studies. Dutta et al. in [7] checked
the performance of wireless Internet telephony in
depth. In this work we observe the behaviour of the IP
telephony when there is another type of traffic in the
network.
In [8], Hederson et al. presented a study about data
traffic in a wireless network of 550 access points and
7000 users. In this paper we can see the increase of
VoIP traffic on WLAN in last years. The authors have
also studied the mobility of users and noted that the
user does not depend on a fixed position. In addition,
they did a study about the amount of corporative
network ingoing and outgoing VoIP traffic, average
length of calls, total VoIP traffic, etc.
These communication systems operate properly via
Internet when there is a control of users connecting to
the network. One way to perform this task is the
admission control. Reference [9] presents an admission
control system of IP calls to obtain a QoS adequate. In
[10] there is a similar work but in this case there is a
system of VoIP over ADSL.
These systems must keep always the user
connected. When the system discards a user in the
admission control when the user is trying to connect to
an access point (AP), it must be reassigned to another
AP. In [11] the authors present a study of the access
selection problem in a multi-access wireless network.
They propose an access selection solution, in which
the arriving users as well as a few ingoing users are
reassigned according to the new systems’ conditions.
This solution selects candidate users to a vertical
handover and anticipates the user context transfer.
We have not found any work where there is a
system of balance of users to improve the wireless IP
telephony. Because of it, in this work we submit a
user-balanced system that improves the performance of
the network when we have a sensitive communications
that need to have a QoS adequate end to end.
3. Problem formulation
One of the main drawbacks of the WLANs in the
2.4 GHz band is their bandwidth when it is compared
with wired networks. Since their appearance, the
available bandwidth has been increased continuously
as new coding techniques have been applied. It has
gone from 1 Mbps up to 108 Mbps (currently offered
by some specific vendors when an extra codification is
applied).
The bandwidth of the deployed WLANs is shared
by different type of users with specific bandwidth
requirements [8].
The Polytechnic University of Valencia is 40 years
old. It is currently formed by 3 campuses and the main
campus has around 50 buildings spread out in 2 km2.
There are around 4,000 researchers and educational
personnel, around 1,500 staff and around 35,000
students in the three campuses.
The WLAN of the Polytechnic University of
Valencia [12] is formed by 575 access points (AP), 33
APs are in the Campus of Gandia, 42 APs are in the
Campus of Alcoy and 500 APs are in the main Campus
(Campus de Vera).
In the WLAN of our university the type of traffic
going through the access points is mainly:
• Http.
• Smtp, Pop3 and Imap4.
• Chat protocols.
• File transfers protocols.
• Multimedia streaming.
• VoIP and IP telephony protocols.
• Other traffic.
Figure 1 shows the evolution of the number of users
during 500 hours in a regular indoor access point of
the Polytechnic University of Valencia. It gives the
number of users use to vary from 0 to 149. The
average value during this time was 9 users.
The most critical protocols are the VoIP and IP
Telephony protocols, because there has not to be
packet loss and delays between packets. The
bandwidth requirements for the G.711 coding are 64
Kbps bitstream for a signal sampled at 8 kHz [13].
Moreover, at this bitstream we must add the control
traffic that H.323/SIP introduces in the network.
Because the available bandwidth of an access point
is shared between the devices connected to it, there
may not be enough available bandwidth to offer the
required Quality of Service for the IP telephony
devices. In order to avoid this problem, it is needed a
system to avoid having too much IP telephones on the
same access point.
We propose to establish a minimum throughput
threshold and when the number of devices connected
to the access point makes to achieve this threshold an
IP telephony devices has to be thrown from the access
point. This IP telephony device will be reassociated
automatically to another access point balancing the
load of the whole system.
4. Proposal
Our proposal is based on the network architecture
shown in Figure 2. This network is formed by IP
telephony devices connected to the access points.
These APs are connected to the wired network through
switches. Finally, there is a server which is responsible
for managing the number of users that are associated
with each access point.
In the proposed user-balanced system, the IP
telephony devices used WPA instead of WEP (as
security protocols) because it prevents the user having
to do the re-association and validation process every
time it leaves an access point. In WPA these processes
are automatic.
160
140
Switch
120
Number of Users
100
Acces
Point A
80
Switch
Acces
Point B
Database
Server
60
Acces
Point C
40
20
0
0
100
200
300
400
500
Hours
IP phone
IP phone
Figure 1. Number of users in an AP during 500 hours.
Figure 2. Network architecture of our user-balanced system.
Router(config)#dot11 association mac-list 770
Router(config)#access-list 770 deny $mac-list
0000.0000.0000
Router(config)#access-list 770 permit 0000.0000.0000
ffff.ffff.ffff
Router(config)#exit
Router#clear dot11 client $mac-list
case of an affirmative response, the IP telephony
device A will remain in that AP. If not, the algorithm
waits 2 seconds and clears the access mac-list from the
APi letting the IP telephony device A reassociate and
the IP telephony device will be marked in the server
database as unmovable for next rounds.
Our algorithm has three main processes. These are
the following:
• Association process: When a client comes
online, it broadcast a probe request. All APs
that receive this request will respond with
information about the AP such as RF hops to
the backbone, load, and so on. If more than one
AP replies, then the client will decide which
AP to associate with, based on the information
returned from the AP. In order to maintain the
association, APs broadcast ‘beacons’ at
periodic intervals. A beacon contains details
similar to that in the probe response. The client
listens to all APs in their coverage area and
builds an information table about the APs. The
association process is illustrated in figure 4.
• IP telephony device maintenance decision:
Using SNMP, the AP informs the server about
its associated clients, so the server registers the
new clients in its database. The server
maintains a database of the clients associated to
each access point related with their signal
strength and their signal quality. When the
server registers an IP telephony device and
calculates that the throughput of that AP
reaches a threshold, it looks for the IP
telephony device in the middle coverage range
and throws it from this access point using the
code shown in figure 5.
• Re-Association process: When an IP telephony
device is threw from an AP, it will try to
authenticate to a new AP automatically. The
reassociation process is shown in figure 6. If
the server doesn’t find its MAC address in any
neighbouring access point for 2 seconds, it
erases the access mac-list from the AP.
Figure 3. Code to throw an IP telephony devices.
The server is constantly receiving SNMP traps from
the APs and the switches, and thus it knows how many
stations and IP telephony devices, and their signal
level, are in each AP. On the other hand, the server can
calculate the available throughput in every AP. Then,
when the available throughput of an AP has reached a
minimum threshold in this AP, the server selects an IP
telephony device to move to another AP. It is based on
the signal level reported by the APs for each client (it
is given by the command “show dot11radio
associations all-clients” in an Cisco Aironet Access
Point). A very high signal level means that the phone
is very close to the AP and it is probably not in the
roaming range, so we select an IP phone with a signal
level less than a threshold value. A very far client
could mean that this access point is the most close and
the last one to be associated. So, our algorithm selects
those phones that are in the middle range.
In order to move an IP telephony device, named A,
from an AP, named i, the server selects the MAC
address of the IP telephony device A in the APi, then it
sends an message to the access point to throw that IP
telephony device. It doesn’t allow the IP telephony
device to associate again to that access point. It is done
by creating an access mac-list which doesn’t allow that
MAC to associate to APi. Then it is disassociated the
IP telephony device A from APi. The phone will
reassociate to another AP in its coverage range
automatically. The code used in a Cisco Aironet access
point to perform these tasks is shown in figure 3.
Maybe the IP telephony device A can’t associate to
another AP. In order to take into account this, the sever
enquiries neighbouring APs to know if the IP
telephony device has associated to anyone of them. In
Calculate
Throughput
Yes
¿Action
needed?
Select Phone to
move
Phone A
Find AP Phone A AP i
is attached to
Block A
association at AP i
Disassociate A
from AP i
Receive Data
¿Is A attached
to any AP?
Yes
No
Unblock A
association at AP i
Mark A as
unmovable
Wait a little time
No
Figure 5. User-balanced algorithm
Database
Server
Database
Server
Acces
Point A
Acces
Point B
Steps to Association:
Client sends probe.
Acces
Point A
Acces
Point B
APs send Probe Response.
Steps to Re-association:
Database server send a
disassocite IP telephony device
from AP (A)
Client evaluates AP response selects
best AP.
Adapter listens for beacons
from APs.
Client sends authentication
request to select AP (A).
Adapter evaluates AP beacons,
selects best AP.
AP (A) confirms authentication.
Adapter sends association
request to selected AP (B).
Clients sends association
request to select AP (A).
AP (B) confirms association
and registers adapter.
AP (A) confirms association and
registers client in the database server.
Initial Connection to an Access Point
Figure 4. Association process.
5. Deployment and measurements
5.1. Test bench
In order to show the performance of our proposal,
we made several types of measurements. The devices
used in our test bench were Cisco Aironet access
points series 1100 and Cisco Catalyst 2950T-24
switches with 100BaseT links. We separated part of
the real network in order to take the most accurate
measurements in a closed environment without the
interference of external devices and to avoid variations
due to external factors. We used Asterisk [14], an
Open source PBX, in order to register the IP telephony
devices. The number of IP telephony devices and the
test procedure is described in the next subsections.
This system is designed to be executed even when a
conversation is running. This situation is the most
critical, for this reason the following measures are
made when there are several conversations running.
5.2. Re-association time measurements.
5.2.1. In different Access Points.
This section gives the time taken by an IP telephony
device to reassociate to another access point. We have
AP (B) informs AP (A) of
association with AP (B).
Roamign from Access Point A to Access Point B
Figure 6. Re-association process.
measured just a unique IP device to know the effect of
our proposal. In figure 7 it is shown that IP telephony
device starts to transmit in the fourth second. In the
second 15 the device is threw of the AP. The peak
indicates the disassociation process. The peak has a
value of 25 KB/s approximately. Then we observe that
there is an interval of 3 seconds where there is not data
transmission, this is due to the re-association process.
In the figure 8 obtained the same measurement for
the process described in figure 7. In this case we got
the number of packets per second. The change of AP
implied a peak of around 300 packets per second.
5.2.2. In the same Access Point.
This section shows the time needed by an IP
telephony device to reassociate to the same AP. The
procedure is the same of the last section, but this time
the access point lets the device to reassociate to it.
In the figure 9 we can observe that in the 15th
second there is a reassociation. When the reassociation
is conducted in the same AP, there is only needed 2
seconds to reassociate. Comparing these measurements
with the ones given by figure 7 we see that when the
reassociation process is conducted between different
APs, there are more delay and more control traffic.
Figure 10 shows the same behaviour, obtaining a peak
of 41 packets per second.
25000
350
300
20000
Packets/s
Bytes/s
250
15000
10000
5000
200
150
100
50
0
0
0
10
20
30
40
50
0
60
10
20
Se conds
Figure 7. Bytes/s in the network.
40
50
60
50
60
Figure 8. Packets/s in the network.
20000
45
18000
40
16000
35
14000
30
Packets/s
Bytes/s
30
Seconds
12000
10000
8000
6000
25
20
15
4000
10
2000
5
0
0
0
10
20
30
40
50
60
0
10
Seconds
Figure 9. Bytes/s in the network.
5.3. Bandwidth measurements.
5.3.1. When the IP telephony devices are in the
same AP.
Then, we measured the traffic in a network where
there were 7 IP telephony devices, connected to an
AP1, talking with 7 IP telephony devices that were
connected with AP2. In this case we connected an IP
telephony device every 5 seconds and, when all of
them were working, we waited 1 minute to see how the
network performed.
In figure 11 we observe that every 5 seconds the
bandwidth used increases due to the connection of the
devices. When the system converges (in the 31th
second) we obtain an average load of 53326.4 Bytes/s.
The average number of packages obtained in figure 12
when the network converges is around 94.5 packets/s.
5.3.2. When there are AP re-associations.
This subsection shows how involves the traffic
when there are 7 IP telephony devices calling while
our proposed system is running. They start the call
every 5 seconds sequentially. The 7 IP telephony
devices are joined to an access point and they try to
call to 7 IP telephony devices in a second access point.
The system throws some of them as they appear in the
network in order to balance dynamically the load. The
system makes the IP telephony devices join the second
access point in order to join the first one. In figure 13
we can see that the bytes/s in the backbone is quite
lower than the one obtained in the previous section.
20
30
Seconds
40
Figure 10. Packets/s in the network.
When the network has converged, there are around
26251.7 Bytes/s.
Figure 14 shows the number of packets per second
while our proposed system is running. When the
network has converged, there is an average value of
49.3 packets per second although there are sporadic
peaks due to IP telephony protocol issues.
6. Conclusions
In this paper we have proposed a new user-balanced
system for wireless IP telephony. It is currently used in
the Polytechnic University of Valencia WLAN. This
system is based on collecting information on a server
from the various IP telephony devices. With this
information, the system is responsible for associating
the IP phones to the APs who possess best features.
Then we have made measurements to check the
system operation. The measurements show that the
system has a good behaviour. When the IP telephony
device performs a reassociation process among APs,
the time since the disassociation until the proper
functioning of the application is 3 seconds. When the
device is disassociated and becomes associated with
the same AP only spend 2 seconds. On the other hand,
our measurements show that there is less traffic in the
backbone when the system is running.
As the system is based on the devices of the
network rather than on the IP Telephony software or
even on the IP Telephony devices, the same results
will be obtained in our system when an access point
fails down.
70000
120
60000
100
Packets/s
50000
Bytes/s
40000
30000
80
60
40
20000
20
10000
0
0
0
10
20
30
40
50
Seconds
60
70
80
0
90
10
30
40
50
Seconds
60
70
80
90
80
90
Figure 12. Packets/s in the network.
Figure 11. Bytes/s in the network.
80000
900
70000
800
700
Packets/s
60000
50000
Bytes/s
20
40000
30000
600
500
400
300
20000
200
10000
100
0
0
0
10
20
30
40
50
Seconds
60
70
80
90
Figure 13. Bytes/s in the network.
One of our future works is create a rule to calculate
the number of IP telephony devices when this system
is running.
Currently we are improving the system, to be able
to link the device according to a Wi-Fi positioning
system such as the one developed by the same authors
in reference [15]. On the other hand we have
implemented the same type of handoff for regular
devices when they cause too much load on one AP.
7. References
[1] Kit-Sang Tang, Kim-Fung Man and S. Kwong, Wireless
Communication Network Design in IC Factory, IEEE
Transactions on industrial electronics, vol. 48, nº. 2, pp. 452459, Hong Kong, April 2001.
[2] O. Hersent, D. Gurle, and J-P. Petit. IP Telephony.
Addison Wesley, 2000.
[3] I. D. Constantiou and K. Kautz, Economic factors and
diffusion of IP telephony: Empirical evidence from an
advanced market, Pergamon Press, Inc., Telecommunications
Policy, Vol. 32, Issue 3-4, pp.197–211, NY, April 2008.
[4] M. Hassan, A. Nayandoro, M. Atiquzzaman. Internet
telephony: services, technical challenges, and products. IEEE
Communications Magazine, vol.38, no.4, pp.96-103, April
2000.
[5] D.P. Hole, F.A. Tobagi, "Capacity of an IEEE 802.11b
wireless LAN supporting VoIP" IEEE International
Conference on Communications 2004, vol.1, pp. 196-201,
20-24, Paris (France), June 2004.
[6] L. Cai, Y. Xiao, X. Shen and J. W. Mark, “Voice Over
IP-Theory and Practice”. International Journal of
Communication Systems, vol. 19, Issue 4, pp. 491-508. 13
April 2006.
0
10
20
30
40
50
Seconds
60
70
Figure 14. Packets/s in the network.
[7] A. Dutta, P. Agrawal, S. Das, et al., “Realizing mobile
wireless Internet telephony and streaming multimedia
testbed”, Computer Communications, vol. 27, Issue 8, May
2004, Pages 725-738.
[8] T. Henderson, D. Kotz, and I. Abyzov, The changing
usage of a mature campus-wide wireless network. 10th
Annual international Conference on Mobile Computing and
Networking, Philadelphia, USA, Sep 26 – Oct 1, 2004.
[9] István Szabó. On call admission control for IP telephony
in best effort networks. Computer Communications, vol. 26,
Issue 4, pp. 304-313, March 2003.
[10] A. Ram, L. A. DaSilva and S. Varadarajan, Admission
control by implicit signaling in support of voice over IP over
ADSL. Computer Networks, vol. 44, Issue 6, pp. 757-772,
April 2004.
[11] V. A. de Sousa, R. A. de O. Neto, F. de S. Chaves, L. S.
Cardoso, F. R. P. Cavalcanti. Access selection with
connection reallocation for multi-access networks.
International Telecommunications Symposium, pp.615-619,
3-6 Sept. 2006
[12] Jaime Lloret Mauri, Jose Javier López Monfort y
Germán Ramos, Wireless LAN Deployment in Large
Extension Areas: The Case of a University Campus,
Communication Systems and Networks 2003, Benalmádena,
Málaga (Spain), September 2003.
[13] ITU-T Rec. G.711. General Aspects of Digital
Transmission Systems Terminal Equipments - Pulse Code
Modulation (PCM) of Voice Frequencies. 1972.
[14] Jim Van Meggelen, Leif Madsen & Jared smith.
Asterisk: The future of the telephony. O’Reilly. USA. 2005.
[15] Miguel Garcia, Carlos Martinez, Jesus Tomas and Jaime
Lloret, Wireless Sensors self-location in an Indoor WLAN
environment, International Conference on Sensor
Technologies and Applications (SENSORCOMM 2007),
Valencia (Spain), October 14-20, 2007.
IP Telephony development and performance over IEEE 802.11g WLAN
Miguel Edo1, Miguel Garcia2, Carlos Turro3 and Jaime Lloret4
Universidad Politécnica de Valencia, Camino Vera s/n, 46022, Valencia (Spain)
1
[email protected];[email protected];[email protected];[email protected]
Abstract
With the adoption of the Wireless LAN technology
as one of the main ways to access the enterprise
network, IP services have found another place where
they can be implemented. In this paper we will show
the test and performance used to develop the IP
telephony network over the 802.11g wireless LAN of
the Polytechnic University of Valencia. In order to
make these measurements, we have used the Open
Source PBX & Telephony Platform: Asterisk and
SmartPhones. We will show the results obtained about
the delay, the jitter and the number of lost packets
when the SmartPhones are in the same wireless cell
and when they are roaming. Finally, we will calculate
the amount of IP phones that can be working in a
single access point and the bandwidth wasted in
different cases. This work can be used to design VoIP
and IP telephony wireless networks and to design
wireless IP phones relocation algorithms.
1. Introduction
IP telephony, or VoIP (Voice over IP), enables
voice communications over networks based on the
Internet Protocol (IP). Therefore, it allows significant
advantages. On one hand, IP phones communications
within the intranet are free. This is most interesting for
companies or institutions which have several branches
or for mobile employers who are moving inside the
intranet with their mobile devices. On the other hand,
the huge investments that have to be done by the
companies or institutions to purchase a Private Branch
Exchange (PBX) can be reduced by using PBXs based
on free software. They provide the same functionality
as a traditional PBX.
From several years ago, wireless networks have
been achieving great popularity because the
deployment of these networks are low cost while
provide us quite mobility and scalability [1].
These networks have evolved quickly to meet the
needs of these users: more security and bandwidth. It is
therefore evolved from IEEE802.11b, which is used to
supply a theoretical bandwidth of 11Mbps, to
IEEE802.11a and IEEE802.11g [2] which provides a
theoretical bandwidth of 54Mbps. This technology is
always in progress.
Although IP telephony was firstly deployed for the
wired network, it can also be deployed on the wireless
network. One of the main advantages of the IP
telephony over a wireless network is that it allows
mobility of the people while they are talking.
Currently, most PDA's and SmartPhones
incorporate Wireless LANs connectivity that is a
considerable advantage because such devices can be
used as either cell or VoIP phone. It is a very important
feature because as long as we have WLAN coverage
we will be able to make VoIP calls.
In this paper we will show how an IEEE 802.11g
WLAN performs using IP telephony with a PBX:
Asterisk [3].
The rest of the paper is as follows. Section 2 gives
previous studies and implementations of IP Telephony.
Section 3 shows the network deployment and the main
features of the IP PBX. In Section 4, we will show the
measurements of the delay, jitter, packet loss and
bandwidth for different scenarios. Finally, section 5
will present the conclusions.
2. Related Work
In the literature, we can find several publications
dealing with IP telephony over wireless networks. In
some sections there is a discussion about the ability of
the IEEE 802.11b/g networks to handle VoIP traffic.
In such papers there are theoretical studies on the
feasibility of IP telephony over WLAN [4] [5] and
also, in some of them, the quality of calls using
different audio codecs G.711 and G.719 [4] is checked.
In other papers, the behavior of IP telephony over
WLAN, when these wireless networks have another
kind of traffic, is analyzed [6].
Furthermore, some studies have been conducted on
IP Telephony Asterisk-based PBX [7], where an
experimental assessment of the IEEE 802.11b standard
to support VoIP on a wired network has been carried
out.
Nevertheless, none of the studies aforementioned
has dealt with IEEE 802.11g networks. Neither any of
them have studied how roaming affect the phones or
the required bandwidth for the phones to have enough
quality of service. Eventually, none of these
considerations have been applied to the SmartPhones
case.
UPV
Network
Asterisk
AP
Cisco Aironet 1130AG
3. Network Deployment and IP PBX
The WLAN of the Polytechnic University of
Valencia [8] is formed by 575 access points (APs)
spread in 3 campuses. 33 of them are in the Campus of
Gandia, 42 APs are in the Campus of Alcoy and 500
APs are in the main Campus (Campus de Vera).
The access points are Cisco Aironet 1130AG Series
APs which use IEEE 802.11a/b/g standard and provide
speeds up to 108Mbps. They are installed to allow the
users a continuous coverage as they roam throughout a
facility. They incorporate the 802.11i IEEE-Compliant
standard (WPA2-Certified and WPA-Certified) which
allows interoperability with other manufacturers. The
coverage of each access point varies between 30 m at
54 Mbps and 137 m at 1 Mbps for indoor
environments.
The IP Telephony PBX is a server that runs a
software desktop application. Asterisk is a freesoftware application (under GPL) performing a
function as a regular telephone PBX. Such as many
PBX, it is possible to connect a specific amount of
phones to make calls between each other and even to
connect to a VoIP provider or to the PSTN (Public
Switched Telephone Network).
The basic package of Asterisk includes many
features that were previously available only in
expensive proprietary systems such as creation of
extensions, sending voice messages to e-mails,
conference calls, voice interactive menus and
automatic call distribution.
Asterisk supports various VoIP protocols including
SIP (Session Initiation Protocol) [9]. SIP is a protocol
for controlling and signaling systems used primarily in
IP Telephony which was developed by the IETF (RFC
3261). The protocol allows to start, to modify and to
finalize multimedia sessions with one or more
participants and its greatest advantage lies in its both
simplicity and consistency.
The audio codecs supported be the Asterisk are the
following ones: G.711 ulaw, alaw G.711, G.723.1,
G.726, G.729, GSM, iLBC, LPC10, Speex.
IP PBX
IP Phone
Nokia E65
Figure 1. Network architecture.
4. Real Measurements
In this section we will show the measurements
carried out in our experiment to evaluate the network
performance.
4.1. Test Bench
In order to test the network performance and
analyze which features offers, we will use two
SmartPhones Nokia E65 and an Asterisk VoIP
telephone PBX. The SmartPhones will be connected to
the Asterisk PBX through the wireless network of the
Polytechnic University of Valencia.
The IP telephones incorporate the IEEE 802.11g
standard and WPA encryption. This is necessary
because the connection to the wireless network of the
Polytechnic University of Valencia is established by
using IEEE 802.11g, WPA encryption with Protected
EAP (PEAP) and EAP-MSCHAP v2 authentication,
thus ensuring a secure communication. These phones
support the SIP protocol which we use to connect with
the Asterisk PBX.
The packages that have been captured for further
study are RTP packets on UDP/IP using the network
analyzer Wireshark [10].
As for the audio codec, we will use the G.711. This
codec will give us the best voice quality as it does not
use any compression. It is the same codec used by the
network ISDN (Integrated Service Digital Network)
and the sound quality is like a conventional telephone.
It also has the lowest latency since there is no need for
compression, which leads to less processing load.
50
45
Call3
Call5
Call9
Call7
50
Max Delay (ms)
40
35
35
30
25
20
15
30
25
20
15
10
10
5
5
0
0
2
4
6
8
10
12 14 16
Time (s)
18
20
22 24
26
28
0
30
Call1
Call2
Call3
Call4
Call5
Call6
Call7
Call8
Call9
Fig 3. Delay average and maximum delay per call.
Fig 2. Measures of the call delay
6
Call1
Call3
Call5
Call7
6
Call9
5
Jitter average (ms)
Max Jitter (ms)
5
4
Jitter (ms)
Jitter (ms) Delay average (ms)
45
Delay (ms)
Delay (ms)
40
Call1
3
2
4
3
2
1
1
0
0
2
4
6
8
10
12
14 16
Time (s)
18
20
22
24
26
28
30
Fig 4. Measures of the call jitter
4.2 30 second call
In order to analyze the performance and quality of
calls we have made 9 calls of 30 second each one,
testing the delay, jitter, bandwidth and packet loss.
The packets are captured from the IP Phone to the
Asterisk PBX.
0
Call1
Call2
Call3
Call4
Call5
Call6
Call7
Call8
Call9
Fig 5. Jitter average and maximum jitter per call.
6ms and is maintained almost constant all the time
around 1 ms.
In Figure 5 we have an average of the jitter and
maximum jitter per call. In the graph we see that the
jitter in none of the 9 calls exceeds 6 ms giving an
average of 1.11418 ms jitter and 3.9244 ms of
maximum jitter average.
4.2.1 Delay tests
4.2.3 Lost packets testing
The Figure 2 shows the data obtained in the delay
measurements. It only shows 5 out of 9 calls because
we want the graph to be understandable. As we can
see, the 5 calls are around 30 second long and none of
them exceed 50ms delay.
In Figure 3 we have the highest average delay and
delay per call. Visually we can see that the average
delay time is around 20ms. Performing the overall
average of 9 calls we obtain a delay of 19.6767 ms and
the average maximum delay is 39.0122 ms
4.2.2 Jitter testing
In Figure 4 we can see the results of tests done
about Jitter. In neither of the cases, the jitter exceeds
In Figure 6 we can see the results of the tests
carried out concerning lost packets. In calls 1 and 2 we
can see how the number of lost packets grows rapidly
around 14 seconds after the call starts because the IP
phone buffer is filled up. Nevertheless, under no
circumstances is this packet loss appreciable during the
communication.
In the following figures 7 and 8 we can see that the
packet loss in calls is very high. The packet loss
average is 162.22 packages: a 12.39 % packet lost.
Although it may seem a very high rate, this does not
affect the conversation. The 9 calls have been of
excellent quality. The transmitted packets average is
1329.77 for the 9 30-second calls.
16
14
Call1
Call3
Call5
Call7
1600
Call9
Lost Packets
12
Lost Packets
Lost packets
Packets/Call average
1400
10
8
1200
1000
800
6
600
4
400
2
200
0
0
2
4
6
8
10
12
14 16
Time (s)
18
20
22
24
26
28
0
30
Call1
25,00%
Call2
Call3
Call4
Call5
Call6
Call7
Call9
60
Lost Packets average (%)
Call1
Call2
50
Delay (ms)
20,00%
Lost Packets
Call8
Fig 7. Transmitted packets and lost packets per call.
Fig. 6. Extent the packets lost in the calls.
15,00%
10,00%
40
30
20
10
5,00%
0
0
0,00%
Call1
Call2
Call3
Call4
Call5
Call6
Call7
Call8
Call9
Fig 8. Percentage of lost packets per call.
4.4. Roaming testing
It has done the same procedure as in section 4.2 but
in this case we have made two calls. One of the phones
is static in an Access Point and the other is moving
around the wireless network at the UPV in the Campus
of Gandia. The data displayed on the following points
are those obtained from the phone on the move.
4.4.1 Testing Delay
The figure 9 shows the data obtained in the analysis
of the delay in testing Roaming. As we can see the
same thing that happens in figure 1, the delay is kept
around 20 ms but in this case because the information
has to go through many more network equipment
ranges between 25 ms and 15 ms with an average of
19.7803 ms very similar to data obtained in section
4.2.1.
In this test, the maximum average delay is 49.92 ms
whereas it was 39.0122 ms in the 30-second calls
between the two phones using the same Access Point.
4.4.2 Jitter Testing
In figure 10 we can see the data obtained in the
analysis of jitter in the roaming test. As shown in the
figure, jitter is around 0.5 and 1.5 ms very similar to
tests carried out in section 4.2.2 with the difference in
30
60
90
120
150 180 210
Time (ms)
240
270
300
330
360
Fig 9. Extent of the delay in Roaming
this case that the maximum jitter average of 3.9244 ms
increases to 6.03ms. On the other hand, the average
jitter has not increased over the average obtained in
section 4.2.2: 1.0626 ms this time, which is a positive
development.
4.4.3 Lost packets test
In figure 11 we can see the data obtained in the
analysis of roaming calls regarding lost packets. As in
previous points, we have something similar to what
happened in the 30-second calls. In this case we have a
lost packet average per call of 8.00 % without causing
any problem in this communication. On the other hand,
the maximum packet loss is caused at times when the
buffer is full in the IP phones. As we can see, these
maximum losses are equidistant.
4.5. Test of effective bandwidth
In figure 12 we show the test of effective bandwidth
in the network of the Polytechnic University of
Valencia. These tests show us that we have a capacity
of around 20000 Kbps with an average of 18514.31
Kbps in the IEEE 802.11g wireless network. Then, in
the following paragraphs we will see the bandwidth
occupied by IP phones and then will calculate the
amount of IP phones that theoretically might work in
this wireless network.
16
7
Call1
Call2
Call2
12
Lost Packets
5
Jitter (ms)
Call1
14
6
10
4
3
2
8
6
4
1
2
0
0
0
30
60
90
120
150
180 210
Time (s)
240
270
300
330
0
360
30
60
90
120
150 180 210
Time (ms)
240
270
300
330
360
Fig. 11. Lost packets in roaming
Fig 10. Measures on Roaming Jitter
120
20000
100
IP BW (Kbps)
BW (Kbps)
15000
10000
80
60
40
5000
20
Call1
Call3
Call5
Call9
Call7
0
0
0
100
200
300
Time (s)
400
500
600
700
Fig 12: Effective bandwidth in the IEEE 802.11g network.
4.5.1 Bandwidth test using the G.711 audio codec
At this point we are going to see the occupied
bandwidth both by the set of the 9 of 30-second calls
and by the roaming calls, using in both tests the audio
codec G.711
In Figure 13, we can see how the calls 1 and 5 have
a drop of bandwidth, this is due to packet loss because
the buffer of the IP phone in both cases is full. This
was explained earlier in paragraph 4.2.3.
On the other side, in the moments where there is no
loss of packets we can see that the bandwidth is around
80 and 110 Kbps with an average of 89Kbps.
In figure 14 we can see the average bandwidth per
call that is always between 100 and 120Kbps:
concretely 110.0444 Kbps. This is very important
because then we will calculate the theoretical number
of phones that can operate on the wireless network of
the UPV.
In Figure 15, we may see something similar that
happens with the 30-second call bandwidth. On this
occasion, the bandwidth has a mean of 90.1559 Kbps
very similar to 89Kbps in the 30 second’s average
calls.
On the other hand, what does vary significantly is
the maximum bandwidth average which raises from
0
2
4
6
8
10
12
14 16
Time (s)
18
20
22
24
26
28
30
Fig. 13. Measures of the 30-second call bandwidth
110.0444 Kbps up to 128Kbps. This is very important
in order to calculate the maximum number of phones.
According to the paragraph above the effective
bandwidth in the IEEE 802.11g network of the
Polytechnic University of Valencia is 18514.31 Kbps.
In our case, the maximum bandwidth that generates
our phones when they were in a position to roaming
(worst situation) was 128Kbps. Following these steps
we can say that the theoretical number of phones that
our wireless network support per access point is
approximately 144.
5. Conclusions
In conclusion, we can say that the IEEE 802.11g
wireless network of the Polytechnic University of
Valencia could, theoretically, support up to 144 IP
phones per access point using the audio G.711 codec.
This number would be obtained in an ideal situation
where we had always this effective bandwidth with a
small amount of external interference always in a
network devoted solely to IP telephony without any
other type of traffic. This can be improved thanks to
the system proposed by the same authors of this paper
[11].
140
IP BW average (kbps)
140
Max IP BW (kbps)
100
100
IP BW (Kbps)
120
IP BW (Kbps)
120
80
60
80
60
40
40
20
20
Call1
Call2
0
0
Call1
Call2
Call3
Call4
Call5
Call6
Call7
Call8
Call9
0
30
Fig. 14. Measure of the maximum bandwidth average and
maximum bandwidth per call.
30 Call
Roaming Call
Delay
19.67 ms
19.78 ms
Jitter
1.41 ms
1.06 ms
IP BW
89 Kbps
90.15 Kbps
Lost Packet %
12.39%
8.00 %
Table 1. Mean value of the data obtained in the tests
aforementioned.
On the one hand, we have realized that the average
of the data obtained in both cases in discussion (the set
of 9 30-second calls and the roaming calls) are very
similar. This can be seen in Table 1.
By contrast, in table 2 we can see as the maximum
increases markedly in the roaming calls. This is due to
the mobility of the user who makes a reassociation
between different access points constantly necessary.
We also consider it could be interesting to use
SmartPhones with IP phones and WiFi connectivity.
These phones have the ability to connect to a wireless
network, and then use the same device to make VoIP
and cellular calls.
Our future work will be focused on carrying out
performance tests connecting the Asterisk PBX with a
standard PBX in order to make calls outside the
Polytechnic University of Valencia. We will make
those calls using the PSTN (Public Switched
Telephone Network) and with another Asterisk’s
supported audio codecs, e.g. G.723.1, G.726 and
G.729. It will open new research lines about mixed
standard mobile-wireless IP Telephony architectures.
6. References
[1] Kit-Sang Tang, Kim Man-Fung and S. Kwong, Wireless
Communication Network in IC Design Factory, IEEE.
Transactions on industrial electronics, vol. 48, No. 2, pp.
452-459, Hong Kong, April 2001.
60
90
120
150 180
Time (s)
210
240
270
300
330
360
Fig. 15. Roaming bandwidth measures
30s Call
Roaming Call
Delay
39.01 ms
49.92 ms
Jitter
3.92 ms
6.03 ms
IP BW
110.04 Kbps
128 Kbps
Table 2. Average of the maximum values.
[2] IEEE 802.11. It is available in The Working Group for
WLAN Standards www.ieee802.org/11/
[3] Asterisk. It is available enwww.asterisk.org
[4] D.P. Hole, F.A. Tobagi, "Capacity of an IEEE 802.11b
wireless LAN supporting VoIP" IEEE International
Conference on Communications 2004, vol.1, pp. 196-201,
20-24, Paris (France), June 2004.
[5] L. Cai, Y. Xiao, X. Shen and J. W. Mark, "Voice Over
IP-Theory and Practice." International Journal of
Communication Systems, vol. 19, Issue 4, pp. 491-508. 13
April 2006.
[6] A. Dutta, P. Agrawal, S. Das, et al. "Realizing mobile
wireless Internet telephony and streaming multimedia
testbed," Computer Communications, vol. 27, Issue 8, May
2004, Pages 725-738.
[7] G. Agreda, J. Gaviria. "EvaluaciónExperimental the
Capacity of IEEE 802.11b support for VoIP." Converging
technologies applied to mobile computing. Memories
I2ComM 2006. Pp 28-39.
[8] Jaime Lloret Mauri, Jose Javier López Monfort and
German Ramos, Wireless LAN Deployment Extension in
Large Areas: The Case of a University Campus,
Communication Systems and Networks 2003, Benalmadena,
Malaga (Spain), September 2003.
[9] RFC 3261, SIP: Session Initiation Protocol.
[10] Wireshark 1.0.4. It is available in www.wireshark.org
[11] Miguel Garcia, Diana Bri, Carlos Turró, Jaime Lloret. A
User-Balanced System for IP Telephony in WLANs. The
Second International Conference on Mobile Ubiquitous
Computing, Systems, Services and Technologies, 2008.
UBICOMM'08. Publication Date: Sept. 29 2008-Oct. 4 2008
On page (s): 251-256.
Multicast TV over WLAN in a University Campus Network
Alejandro Canovas1, Fernando Boronat2, Carlos Turro3, Jaime Lloret4
Polytechnic University of Valencia
EPSG - IGIC Institute1,2,4
ASIC-UPV3
Ctra. Nazaret-Oliva, S/N
Camino de Vera, S/N
Grao de Gandia
Valencia
[email protected], {fboronat2,jlloret4}@dcom.upv.es, [email protected]
Abstract. One of the multimedia services offered by the campus
network of the Polytechnic University of Valencia is TV over IP.
This service works well in the devices connected directly to the
wired network but we have detected some problems when the
receivers access to the campus network through wireless IEEE
802.11, especially when devices roam across the Campus. In this
paper we propose and evaluate a server-based solution to minimize
the packet loss and reduce the lack of service when the mobile
devices roam from an Access Point to another Access Point in the
wireless network. This solution uses a location system to modify
the behaviour of standard multicasting protocols in order to get a
near-seamless multicast WiFi roaming.
Keywords- IPTV, Multicast
I. INTRODUCTION
The Information and Communications Systems Area of
the Polytechnic University of Valencia (UPV) is offering
the TV over IP multicast service (IPTV) to all the members
of the University since 2004 ([1]). With the integration of
such a service together with data and voice over IP (VoIP),
the UPV offers a complete Triple-play (voice, data & video)
service to its members. The distribution of TV over the IP
campus network reduces the costs in TV infrastructure en
each new building of the campus.
The service offered by the UPV to its members includes
more than simple TV channels. It includes traditional
Terrestrial or Satellite digital TV (Figure 1) and other
Services such as Educational Channels, Event broadcasting,
information channels.
The structure of the campus network and the TV
streaming infrastructure including wireless access will be
presented later.
Nevertheless, whereas we have detected that the service
works well (provides good QoS) in wired receiver devices
(i.e., connected directly to the wired network), on the
contrary, some problems arise when the receivers are mobile
devices (laptops, PDAs…) which are accessing to the
campus network through wireless IEEE 802.11 Access
Points (AP). The service doesn’t provide good QoS to those
devices, especially when the devices are moving across the
Campus.
When roaming situations appear, lots of packets are lost
and the service is stopped during a too long interval (half a
minute, approximately), time needed to rearrange the
multicast tree until the new AP.
We can think that most of the possible scenarios for
clients accessing IPTV services suggest that they will have
low mobility during such sessions, but there may be brief
periods of mobility as the user moves from one WLAN
access area to another and so the problem of vertical
handover needs to be addressed. In this paper we propose a
preliminary solution that minimizes the number of lost
packets and the duration of lack of service intervals when
roaming situations are inevitable.
Figure 1. UPV Television
The rest of the paper is organized as follows. In Section
2, we discuss related work. In Section 3, we present the WI-
FI University Campus Network and the infrastructure for
TV Streaming. In Section 4, a possible solution of the
problem detected when devices are roaming from one AP to
another AP is outlined. Section 5 presents the results of the
evaluation of the proposal. Section 6 presents our
conclusions, discussing some future directions. Finally, the
paper ends with the references.
II. RELATED WORKS
The way of supporting user mobility in IEEE 802.11
networks can have a strong effect on certain types of
services. Roaming is the main cause for heavy packet loss
error in wireless networks. As far as we know, the IEEE
802.11b standard for WLANs allows for handover between
overlapping WLAN cells at the link layer. Since this only
permits connection to one WLAN at a time, it falls into the
category of hard handovers. During handover clients can not
send or receive data and packets queued at the old WLAN
will be lost, making it unsuitable for the handover of
multimedia traffic. The roaming management produces
temporary periods of high packet loss which affect
bandwidth estimations and these underestimations reduce
service performance during longer periods of time than
those caused by the usual roaming process. For real-time
video streaming, roaming directly affects the quality of
video reception and impacts the user satisfaction.
To date, most MAC protocols for wireless networks do
not provide a reliable multicast service. For reliably
multicasting packet over WLAN, it will be necessary to
modify the MAC layer protocol to add recovery mechanism.
Adding local recovery at the MAC layer can greatly
improve the performance for multicast in wireless networks.
Next-generation WLAN standard probably supports the
more reliable multicast/broadcast scheme [2].
In the last years, we have found an intense research
activity on the effects of WLAN over streaming services.
There are multiple possible solutions to decrease the
roaming effects in streaming services, ranging from
modifications on current streaming service devices (clients
and servers) to the design of new intermediate devices to
avoid modifying the service devices. Here we present some
outstanding solutions to improve the quality of such
services, especially in roaming situations.
The first option we can think of is the use of Mobile IP
[3], which allows a Mobile Node (MN) to receive IP packets
through a packet forwarding procedure, but handovers in
Mobile IP are slow and packets can be lost during the
handover procedure, making it unsuitable for the handover
of video traffic.
In [4], an advanced agent-based architecture to provide
guaranteed quality to the Mobile users in a WLAN is
presented for a VoD (Video on Demand) Service. In it, the
Access Points manage the user’s mobility (handoff) and
implement the management policies of the QoS
(reservation, allocation and distribution of the bandwidth).
The service architecture is based on intermediate elements
(virtual servers) and client software modifications.
In [5], a vertical soft handover scheme is presented,
using jitter as the indicator for initiating the handover
process.
A method combining the benefits of multiple
descriptions coding (MDC) and multipath routing is
explained in [6] to improve the quality of streamed video in
WLAN roaming situations. It incorporates channel status
detection mechanism to decide which channel will be
selected or multiple channels will be used to take advantage
of path diversity to deliver the streaming video content. The
loss-rate and round-trip time are used to determine the
channel status by using active probing.
In [7] we can find a proxy-based middleware that
foresees client handoff and manages intermediate buffers
between client and server to reduce the effects of the
handover latency. The proxy manages an intermediate
buffer that stores data during roaming to reduce packet loss.
In [8], a solution to minimize all the negative effects of a
roaming situation in a WLAN, based on a buffering scheme
and the pro-active management of signalling control
messages between clients and server, is proposed. It is
based on off-the-shelf WiFi hardware and unmodified
commercial streaming clients and servers. A Wireless Proxy
(intermediate element) aware of the type of access network
is used to manage client to server signalling. The results
show that the use of such a transparent intermediate element
filtering or forwarding client signalling messages
significantly improves streaming service performance over
WLANs. The results of the tests also show that maintaining
an independent stable channel between server and proxy
helps to reduce roaming effects over the interchanged data.
Our solution, explained in Section 4, decreases the
amount of lost packets during the roaming processes, by
including a new intermediate device and using a WiFi
location system. It doesn’t imply the modification of the
server or clients sides and then is suitable to be implemented
for any kind of networks.
III. THE WI-FI UNIVERSITY CAMPUS NETWORK. TV
STREAMING INFRASTRUCTURE
As stated previously, Polytechnic University of Valencia
has, for more than three years, a system for distributing
IPTV for 38 simultaneous channels with MPEG-2 encoding
at 4-Mbit/s over the network. Reception of TV channels is
either from the user’s PC or, for conventional televisions
with set-top-boxes (STB) connected to the IP network.
This service is used in computers on the wired network
without cuts or reception problems, using multicast services
over the UPV campus network infrastructure, which allows
access for 40,000 users at the University with L3 and L2
Gigabit and Fast Ethernet switches with for IGMP snooping
support. This function intercepts multicast traffic
intelligently to avoid the multicast flow where it is not
necessary by knowing about multicast requirements reading
IGMP report and IGMP join messages between the
multicast routers and network PCs. A sample of the network
infrastructure of the UPV used for IPTV services can be
seen in Figure 2.
Moreover, the University has an 802.11g wireless
network, with over 500 APs deployed, which provides
100% coverage of the campus. Obviously, it is not possible
to deploy directly a IPTV service to the WiFi network
directly because of bandwidth involved. So, we have
developed a testbed for providing IPTV services for mobile
devices using a farm of PCs using open-source VLC
encoder [9]. That PCs scale and compress on the fly each of
the different TV channels with MPEG-4 codec at a speed of
256 kbps and a resolution of 320x240 pixels and recast
those streams to the network with a new multicast group
(one for each TV channel). This recoding enables receiving
with an appropriate bandwidth via the University’s WiFi
network as shown in figure 3.
1
G
bp
s
Figure 2. Network communications for UPVTV
1G
1
s
bp
s
bp
G
Figure 3. Multicast TV roaming setup
Again, it is necessary to comment on the pre-requisite of
IGMP snooping support by the network devices involved to
prevent getting into a WiFi cell more multicast groups than
the required ones. But this infrastructure cannot support
roaming in an appropriate way, because if the user roams
from AP to any other AP, multicast packet loss is
unacceptable, because the new L2 switch doesn’t allow
multicast packets to flow, and this effect is amplified by the
usual requirement of coding systems in having a video key
frame to be able to continue playing. This loss of
connectivity is the problem we try to solve with our
proposal.
Now we have to clarify the concepts of soft handover
and hard handover. While we can find these terms
elsewhere in the literature in some senses, some specific
meaning is required here.
On the one hand, and in the context of our work, we
consider soft handover as a handover in which the same data
is delivered to the mobile device simultaneously via two
access networks. This can be resource intensive but it
reduces the probability of data loss during the handover. On
the other hand, we consider a hard handover as the one in
which data is streamed via one network at any time (at some
specific time, a decision is made to receive the data through
one specific network). Hard handover is more parsimonious
with network resources but it can result in data loss at the
user device, depending on how quickly the handover is
made
and
how
much
data
the
device
is
transmitting/receiving.
As mentioned above, the way IEEE 802.11 allows for
handover between overlapping WLAN cells at the link layer
causes packet loss, making it unsuitable for the handover of
multimedia traffic. Moreover, after a hard handover the
mobile device will not receive multicast due to the path is
blocked by the IGMP snooping functionality.
Finally there are many proposals for location on 802.11
networks. We use location from the network side, as
proposed by the authors in [10], by means of power received
by the APs and probably a field survey. We will use that
location information to prepare multicast traffic in roaming
situations.
IV. OUTLINE OF MULTIMEDIA HANDOVER SCHEME
As stated previously, our proposal solves packet loss on
hard IEEE 802.11 roaming by combining location and
network control in order to achieve near-seamless multicast
roaming without requiring any 802.11 client behavior
modification.
Scheme is depicted on figure 3. When a client registers
on a multicast group via an IGMP Join message, the first
upstream multicast routers sends a SNMP message to a
control server (through standard event notification process),
in order to locate the position of that client (message
number 1 in figure 4). This location system is based on
Power measurements that are also sent to the control server
trough SNMP messages too (message number 2 in figure 4).
When a multicast client is about to leave a WiFi cell, the
location systems send a message to the L2 switches in the
new route from the multicast sender to the planned AP
(message 3). This message must arrive to all L2 switches
that will be in the path of the new multicast route to the
predicted Access Point.
In our real implementation with CISCO switches, that
message sends a “ip igmp snooping vlan static” command to
configure an L2 interface as a member of that multicast
group. As far as we have no opportunity to deal directly
with source code of CISCO operating system, we have
simulated a Telnet session with prerecorded commands to
send effectively the previously mentioned message.
Anyway, the way we think it should be done is through a
standard SNMP message.
After this message we have two simultaneous multicast
paths in which we have flowing the TV session. Then there
are two possibilities: the first one is that the mobile roams as
we have expected because of the information provided by
the location system, or maybe the user has changed its mind
(was my office locked?) and decides not to roam. In each
solution, IGMP timers for multicast sessions will take care
of the unused path and will delete it gracefully. So our
proposal works correctly even for that kind of events.
We have implemented this service on an infrastructure
of APs and switches from Cisco Systems in a building of the
university and we have made several tests of roaming in real
wireless environments using this technology. The scenario
is shown in figure 5. The laptop was running a VLC client
([9]) and a random TV channel was selected to be watched.
The laptop characteristics are: Macbook Pro laptop under
Leopard OS, Intel Core Duo 2,2 GHz, 4GB of RAM. It was
equipped with interior Broadcom WLAN 802.11n adapter.
The laptop started the playout in the coverage area of an
AP (AP 1, in the figure) and moved along the corridor
(following the arrow path) to the covering area of another
AP (AP 2, in the figure).
Our test transmission is a MPEG-4 video at 256 Kbps
with a 320x240 pixels resolution, which is a bandwidth we
have estimated correct for TV reception in mobile devices
like PDAs and Smartphones.
All the tests were made in two ways: one without
including the improvement proposed and other with it. Due
to paper extension limitations, we only show the results of
one of the tests.
Next, we present three figures with the results of lost
packets (figure 6), jitter (figure 7) and bandwidth
consumption (figure 8) measured during the session, using
the Wireshark Network Protocol Analyzer Software
(version 0.99.7). In them, we can observe that when we
don’t implement the proposed improvement, the Mobile
device reception of packets was interrupted from the second
81 to the second 115 (34 seconds later), instant in which it
continued the playout of the TV sequence. This means an
unacceptable lack of service of about 34 seconds (pause in
the playout process and a skip effect with the loss of part of
the TV sequence).
AP2
Figure 4. Message flow for our proposal
5. EVALUATION
Figure 5. Measurement scenario
AP1
Lost Packets (packets)
3500
3000
2500
2000
1500
1000
500
0
0
20
40
60
Time (s)
Normal Roaming
80
100
120
Improved Roaming
Jitter (ms)
Figure 6. Lost Packets
14
12
10
8
6
4
2
0
0
20
40
60
Time (s)
Normal Roaming
80
100
120
Improved Roaming
Figure 7. Jitter
3000
BW(Kbps)
2500
2000
1500
1000
500
0
0
20
40
60
Time (s)
Normal Roaming
80
Improved Roaming
Figure 8. Bandwith Consumption
100
120
The total amount of lost packets in this case, due to the
roaming operation, was 3230 packets. On the other hand,
when we implemented our proposal, in the mobile device
the reception of packets was interrupted from the second 81
to the second 84 (3 seconds later), instant in which it
continued the playout of the TV sequence. This means a
quite short time of service interruption more tolerable by the
users. The total amount of lost packets in this case, due to
the roaming operation, was 422 packets (it means a
reduction of the 87% regarding the number of lost packets
in the previous test, without implementing the
improvement).
All the tests obtained similar results. So, we can conclude
that with our improvement we reduce considerably the
amount of lost packets and the duration of the interval
without service due to the roaming operation.
VI. CONCLUSIONS AND FUTURE WORK
The distribution of IPTV is the natural evolution of the
services offered by the Universities to their members. In the
UPV there is a complete IPTV service, on the one hand,
with satisfactory functionality QoS (and compatibility with
other services) offered to the wired devices but, on the other
hand, with unsatisfactory QoS offered to Mobile Devices.
With the preliminary solution proposed in this paper we
have reduced considerably the number of lost packets and
the duration of the situations of lack of service due to the
roaming situations, and without modifying the server and
clients sides. It has been tested without background traffic
load, so the effects over different levels of load have not
been analyzed. Moreover, the study of multiple users
performing roaming has not been done yet.
Although our solution has proved valid it is not enough,
we will continue investigating on it in order to reduce the
lack of service period during the roaming and to reduce the
number of lost packets. We will study the feasibility of a
similar solution for 3G to WiFi roaming and WiFi to wimax
roaming, because we have some extent of these technologies
deployed across the campus and that kind of support will
provide a new level of service for our users.
On the other hand, Microsoft researchers in conjunction
with some academics proposed in [11] a mechanism by
which a single WLAN interface can be used to
simultaneously access multiple WLANs. If future devices
will have such functionality we can start the handover
process through the second WLAN before losing connection
to the previous one. With buffering techniques no packets or
few packets will be lost.
Moreover, other services such as access control and the
characteristics of the audience of the service are not defined
yet. At this stage, those subjects are being studied.
REFERENCES
[1] C. Turró, J. Pasamar, M. Jiménez y J. Busquets, “Using Intensively the
Network: 152 Mbps of Multicast to Distribute TDT, Satellite and
Educative Channels in the UPV”, Spanish National R&D Network
Bulletin, PP. 77-80, December 2006-January 2007.
[2] J. Ma, X. Feng, Y. Liu and B. Tang, "Video multicast over WLAN,"
IEEE International Symposium on Communications and Information
Technology, ISCIT 2005, vol.2, pp. 1400-1403, 12-14 Oct. 2005.
[3] C. Perkins, “IP mobility support”, RFC 2002, IETF, (1996).
[4] D. Bruneo, M. Villari, A. Zaia, A. Puliafito, “VOD services for mobile
wireless devices”, in Proc. of the Eighth IEEE international
Symposium on Computers and Communications, ISCC03, June 30 July 03, 2003.
[5] Cunningham, G., Perry, P., Murphy, L., "Soft, vertical handover of
streamed video", Fifth IEE International Conference on 3G Mobile
Communication Technologies, 3G 2004, pp. 432-436, 2004.
[6] C-M. Chen, Y-C. Chen and C-W. Lin, “Seamless roaming in wireless
networks for video streaming”, ISCAS (4) 2005, pp. 3255-3258.
[7] P. Bellavista, A. Corradi and L. Foschini, “Application-level
Middleware to Proactively Manage Handoff in Wireless Internet
Multimedia”, 8th International Conference on Management of
Multimedia Networks and Services, MMNS 2005, Barcelona, Spain,
October 24-26, 2005.
[8] M. Vilas, X. G. Paneda, D. Melendi, R. Garcia and V.G. Garcia,
“Signalling Management to Reduce Roaming Effects over Streaming
Services”. In Proc. of the 32nd EUROMICRO Conference on Software
Engineering and Advanced Applications, EUROMICRO, pp. 398-405,
August 29 - September 01, 2006.
[9] VideoLAN - VLC Media Player, http://www.videolan.org
[10] M. García, C. Martínez, J. Tomas and J. Lloret, “Wireless Sensors
self-location in an Indoor WLAN environment”, International
Conference
on
Sensor
Technologies
and
Applications
(SENSORCOMM 2007), Valencia (Spain), 14-20 October 2007.
[11] R. Chandra, Pr. Bahl, and Pl. Bahl, “Multinet: connecting to multiple
IEEE 802.11 networks using a single wireless card”, Proc. IEEE
Infocom, (2004)
People Mobility Behaviour Study in a University Campus using WLANs
Sandra Sendra, Miguel Garcia, Carlos Turro, Jaime Lloret
Universidad Politécnica de Valencia
Camino Vera s/n, 46022, Valencia, Spain
[email protected], [email protected], [email protected], [email protected]
Abstract—Wireless Local Area Networks are becoming
necessary in university campuses and enterprise areas. Taking
advantage of this technology, many benefits can be obtained.
One of them is the use of the user’s mobility information when
there is ubiquity in the wireless network. This information can
tell us which places are most visited, if people have to go to
places that are far from their office, detect the best situation
for critic points, etc. The study presented in this paper show us
the case study of a university campus of two square kilometers
and how we have taken advantage of the information gathered
from the wireless network. This approach can be used by the
enterprises to optimize the sites to place their resources
(network printers, servers, meeting rooms, etc.). Furthermore,
a network administrator will be able to relocate bandwidth in
order to increase the comfort of the end users.
Keywords-People Mobility; people behavior; people traking;
WLANs
I.
INTRODUCTION
Nowadays, wireless local area networks are widely
implemented. The public organisms such as universities,
governments, etc. are well known examples. These wireless
networks are usually based on the IEEE 802.11 b/g standard
[1]. The standard presents many advantages. We can
emphasize some of them:
• The use of a free band in the 2.4 GHz.
• Speeds up to 54 Mbps.
• The user’s comfort is bigger than wired networks.
• Wireless networks allow the access of multiple
computers with a smaller infrastructure cost.
• The compatibility among different devices is very
high, because of the organization Wi-Fi [2].
The IEEE 802.11 b/g networks present the intrinsic
problems of any wireless technology. Some of them are:
• The wireless connections bandwidth is smaller than
in wired connections.
• Are more prone to be attacked because they can be
accessed from anywhere, although there are several
methods to encrypt the communication.
• The roaming can stop any communication between
the devices of the network.
• Wi-Fi is not compatible with other wireless
technologies like Bluetooth [3], UMTS [4], etc.
The wireless LAN network is mainly used to transmit
data, but there are many other applications. One of the most
well known applications is the indoor positioning system.
The localization is made using the access point’s received
signal strength and it is possible the use of different
mathematical methods [5]. There are many other
applications such to provide connectivity in meetings,
wireless VoIP, wireless IPTV and so on. Most of them are a
service guided to final client.
In this paper, we use the data obtained from the WLAN
in order to study the mobility of the users. The roaming
information can be used to know the behaviour of the people
in a place, to relocate the bandwidth and how they change
from a building to another. This information will let us know
the movement of the users, what buildings are most visited,
etc.
The rest of the paper is organized as follows. In section 2,
we discuss the related work. Section 3 shows our university
wireless network. People tracking measurements can be
observed in section 4. Section 5 shows the people behavior
in our campus wireless network. Finally, Section 6 presents
our conclusion and future work.
II.
RELATED WORK
We have found several works related with mobility
people and tracking. Some of them are the following. In [6],
Z. Chen et al. presented a system that works like an indoor
GPS. It uses RFID and provides directional instructions for
users while tracking things. It is called DynaTrack and
consists of three key parts which are the RFID tags and
readers, database servers that hold information about things’
location and the DynaTrack client side interface. This system
uses a dynamic or static system tags, depending on if it is an
object or a person. Also, they tell us that if an object, initially
labeled as static, begins to move, the system is able to
change its address dynamically.
Another example about tracking system is the one
presented by J.G. Markoulidakis et al. in [7]. In this paper,
we can see a new system based in Third Generation Mobile
Telecommunication Systems (TGMTS) and the three basic
types of mobility models that are appropriate for the full
range of the TGMTS design issues. They propose a
methodological modeling approach called Integrated
Mobility Modeling Tool (IMMT). IMMT tries to improve
some aspects of other systems like the validation of the
theoretical input assumptions and analytical models or the
effect of the mobility model accuracy.
The authors in [8] show the possibilities of utilizing
RFID, Wi-Fi and BlueTooth wireless technologies in
personnel/equipment tracking and mapping mine works of
Pollyanna (underground mine in Oklahoma). Other wireless
technologies, as the conventional satellite GPS technology,
are not feasible there. They evaluate the advantages in the
real-time location services (RTLS) technology to determine
their applicability and limitations to underground mining at
the Pollyanna.
In [9], B. Issac et al. presented a predictive mobility
management system which could make mobility on an IEEE
802.11 network more proactive with minimum loss and
delay, when compared to existing schemes. Their proposal is
focused on WLAN installations within a restricted campus
and to predict the mobility path of a mobile node and use that
information to lessen the handoff delay.
A wireless indoor tracking system, based purely in
software because no additional hardware is required, is
described in [10]. It can be used to track and locate both
moving and static WLAN-enabled devices inside a building.
The system uses complex mathematic algorithms and
determines the locations of the mobile devices according to
the received signal strength from visible access points. The
author categorizes the WLAN-based location determination
algorithms, into two groups: deterministic and probabilistic
algorithms. Finally, he concludes the paper making some
reflections about the number of APs and their correct
localization in order to obtain reliable results, among other
things.
There are other works that show a study and even try to
imitate the human behavior movements through simulations.
One of them is the paper presented by T. Liu et al. in [11].
They present a model in order to mimic human movement
behavior. It is built as a two-level hierarchy in which the top
level is the global mobility model or GMM (a deterministic
model that is used to create intercell movements) and the
bottom level is the local mobility model or LMM (a
stochastic model with dynamically changing state variables
to model intracell movement).
Another example about simulation human behavior is
given in the paper presented by C. Bettstetter [12]. It shows a
model that can be used in simulations of mobile and wireless
networks. He uses a combination of principles for direction
and speed control to provide the movement of the users. It
shows the calculation process to simulate changes of speed,
stop-and-go behaviors or address control, among others.
In summary, the works presented previously carried out
studies only related with users' tracking, except for [11] and
[12] that develop simulation models of people's mobility. In
all of these papers, the buildings are considered as objects
instead of groups of APs. This paper presents a different
work; because we studied the users’ mobility through
grouping APs in several buildings, no matter which person
is. With these data we could relocate some services and the
displacement of the users would be more efficient.
III.
WIRELESS NETWORK DESCRIPTION
The Polytechnic University of Valencia is distributed on
three Campuses. One of them is located in Valencia and
contains about 80% of the students and staff of the
University. It has a dimension of about three kilometers long
and one kilometer wide. There are two smaller campuses in
the nearby cities Gandia and Alcoy. There are around 4,000
researchers and educational personnel, around 1,500 staff
and around 36,000 students among the three campuses. The
distribution of students in each faculty is shown in Table 1.
On these Campuses, a wireless IEEE 802.11 b/g network
is deployed. It comprises more than 575 access points to get
full coverage, including not only the buildings and offices,
but the surrounding gardens and open space between these
buildings. So, any person in the UPV can roam seamlessly
between any locations. The distribution of these access
points is: 33 APs are in the Campus of Gandia, 42 APs are in
the Campus of Alcoy and 500 APs are in the main Campus
(Campus de Vera). The APs are installed to allow the users a
continuous coverage as they roam throughout a facility. The
coverage of each access point varies between 30 m at 54
Mbps and 137 m at 1 Mbps for indoor environments.
The access points are from Cisco Systems Inc. (models
1130, 1140 and 1300) and they are configured with three
simultaneous SSIDs, one with VPN authentication, another
one with 802.1X authentication and the last one interacted
into the EDUROAM (European roaming project) for visitors.
Any member of the University, and from others via
EDUROAM, has free access to that wireless network.
IV.
PEOPLE TRACKING MEASUREMENTS
It is quite complicated to predict if the students will visit
more times some buildings than others. This study could be
used to relocate some schools and services in order to obtain
a more effective and efficient distribution or even to help
planning the construction of a new university. In this section,
the measurement process will be explained in order to
analyze the number of users' change between buildings.
A. Baseline measurement
In order to gather information from the wireless network,
the SNMP agent was activated in all wireless APs using only
the required messages. Every time a MAC address is
associated to an AP, it sends a SNMP trap message to a
central server. This information is stored in a database to be
processed and analyzed.
First, APs are grouped according to the building where
they are placed. This activity was not difficult because in our
university each AP has an unique identifier, formed by the
name from the building and the MAC address. All the APs in
a building can be grouped easily using the same badge.
The database contains several tables in order to analyze
the information. There is a table that stores the day and the
month of the information jointly with the AP DNS name and
the MAC that has been associated. Another table relates
every access point with the building where it is placed. These
tables allow us to make several queries such as:
• MACs registered
• Buildings with wireless access points
• MACs in every building
• MACs that roam between buildings
• MACs that roam between buildings every day
• MACs that roam between buildings every month
• MACs associated to every AP during a day
• MACs associated to every AP during a month
• APs where each MAC has been associated in a day
• APs where each MAC has been associated in a
month
TABLE I.
PEOPLE REGISTERED IN EACH BUILDING
Building
E. Politecnica Superior de Alcoy
E.T.S de Ingenieria de Informatica
E.T.S. de Arquitectura
E.T.S. de Gestión de la Edificación
E.T.S. de Ingenieria del Diseño
E.T.S. del Medio Rural y Enología
E.T.S.I. de Agronomos
E.T.S.I. de Caminos, Canales y Puertos
E.T.S.I. de Telecomunicación
E.T.S.I. de Geodesica, Cartografía y Topología
E.T.S.I. de Industriales
E. Politecnica Superior de Gandia
F. de Administración y Dirección de Empresas
F. de Bellas Artes
Total
People registered
2298
3240
3858
2920
4794
1074
1841
3145
1409
1027
3479
2320
2271
2334
36010
Once the queries can be performed, we can process the
information.
B. Data processing
The people tracking measurement process is based on the
number of people roaming among buildings. It is also
measured where the people stops during a period of time.
These measurements let us know the quantity of movements
among all the buildings in the campus. In order to estimate
the time that an user takes to go from the A building to
another B building, we keep in mind that it could be the C
building inside this itinerary. Roaming will exist among the
A building, the C building and the B building, but the
displacement will be considered from the building A to the B
building. This study cannot be considered as a system of
privacy intrusion, because this system does neither save a
correspondence list of each person, nor their MAC
addresses. Only the amount of movements is interesting.
Once all APs of each building have been grouped, the
roaming among the APs of the same building will not be
taken into account because these movements are inside the
same building and the user does not move among buildings.
All these data has been stored in a database during a
month to carry out this study. Firstly, the data have been
purified because there was some information that was not
useful to this study. These data have been taken daily and,
therefore, we can show the information gathered during a
regular day or show the information about monthly activity.
In order to process the data we have used a spreadsheet
Excel 2007 with the NodeXL tool [13]. NodeXL is an
extendible toolkit for network overview, discovery and
exploration. The core of NodeXL is a special Excel 2007
workbook template that structures data for network analysis
and visualization. Six main worksheets currently form the
template. There are worksheets for “Edges”, “Vertices”, and
“Images” in addition to worksheets for “Clusters,” mappings
of nodes to clusters (“Cluster Vertices”), and a global
overview of the network’s metrics (“Overall Metrics”). As
we will see in the following section, this software tool allows
visualize the roaming among the buildings, the quantity of
roaming made, filter the quantity of roaming, etc. NodeXL is
a powerful tool that can help us analyze the behavior of the
network. NodeXL aims to make analysis and visualization of
network data easier by combining the common analysis and
visualization functions with the familiar spreadsheet
paradigm for data handling. The tool enables essential
network analysis tasks and thus supports a wide audience of
users in a broad range of network analysis scenarios.
C. Roaming during a day in the Vera Campus
Table II shows the numbers of changes among the
buildings in one day. The rows represent the number of users
that roam from that building to another. The situations where
there is no mobility among a pair of buildings, that is, there
is no MAC roaming between that two buildings in a day, are
represented with a dash (-). The biggest value obtained in the
user’s mobility during a day is carried out from the building
“E.T.S. de Gestión en la Edificación” and the “E.T.S.
Ingeniería Informñatica” (4912 roamings). This is because of
the buildings situation. The easiest way to access the “E.T.S.
Ingeniería Informática” building is through the “E.T.S. de
Gestión de la edificación” building. Furthermore, this last
building is located in front of a tram stop, so it is an entry
zone to this part of the university. In Table II, it can be seen
that “E.T.S. of Telecommunication” building has many
roamings to other buildings. The reason is similar to the
previous one, in front of this building there is also a tram
stop and this building is located in the central area of the
main campus. Among the “E.T.S. de Telecomunicación” and
“E.T.S. de Caminos, Canales y Puertos” there are 2290
roamings in a day, this is because it is needed to cross the
“E.T.S. de Telecomunicación” building to arrive to “E.T.S.
de Caminos, Canales y Puertos” building. Another building
that has a lot of roamings is the “E.T.S.I. Geodésica,
Cartográfica y Topografía” building. In this case these
roamings were caused because it is placed near the snack
bar. This snack bar has wireless coverage thanks to the APs
of the “E.T.S.I. Geodésica, Cartográfica y Topografía”
building. We will see several movements related with this
building in our studies. Lastly, among the “E.T.S.I. Caminos,
Canales y Puertos” building and “E.T.S. Arquitectura”
building there are 2202 roamings by day. These roamings are
due to: a) the proximity between both buildings and b) the
relationship of contents that are taught in both buildings.
May be students and/or professors walk from one building to
the other in order to carry out theoretical or practice classes.
D. Roaming during a month in the Vera Campus
Table III shows the roamming value carried out during a
month among the buildings of the Vera Campus of the
Polytechnic University of Valencia. In this Table the data
movements from one building to another and viceversa have
been added. That is, we have not considered the direction of
the roaming.
The maximum number of roamings in one month is
carried out among the “E.T.S. de Gestión en la Edificación”
building and “E.T.S. Ingeniería Informática” building. We
explained why before. The number of roamings between
“E.T.S. de Telecomunicación” and “E.T.S. de Caminos,
Canales y Puertos” buildings was 26746. In this case
roamings were due to the proximity of the buildings and
because it is necessary to cross the building “E.T.S. de
Telecomunicación” to arrive to “E.T.S. de Caminos, Canales
y Puertos” when the people come from the tram.
―
15 4912 986
―
―
―
111
195
―
249
―
― 137 200 ―
―
―
―
279
27
7
12
17
―
9
―
2
4
E.T.S. Medio Rural y
―
Enologia
E.T.S.I. Agronomos
―
E.T.S.I. Caminos, Canales y
2
Puertos
―
E.T.S.I. Telecomunicación
2
663 1379 1307 458 20 218 2270
4
94
12
32
491 742 537 475 ― 965 434 ―
16
7
20
7 ― 10 24 ―
―
―
―
―
―
―
―
―
―
―
―
―
―
―
86
―
―
―
52
―
―
―
―
―
E.T.S.I. Geodesica,
Cartografica y Topografía
E.T.S.I. Industriales
E. P. S. Gandia
Facultad Administración y
Dir. de Empresas
Facultad Bellas Artes
―
24
4
―
―
―
― 2202 ―
―
― 248
50
92
304
1
11
―
47
40
―
―
―
―
―
55
―
―
―
―
47
49
E.T.S.I. de Geodesica,
Cartografía y Top.
― 507 28
100
95
50
2
1047
12
5
76
4
72
―
E.T.S.I. Industriales
E. P. S. Gandia
F. de Administración y
Dir. de Empresas
64
―
―
―
―
163
315
1350
6704
845
8591 15180 17704
F. de Bellas Artes
341 ― 494 568 ―
96
10429 1276
F. de Adm.y Dirección
de Empresas
―
E. P. Superior de
Gandia
42
E.T.S.I. de Industriales
40
―
67
7344
E.T.S.I. de Geodesica,
Cartografía y Top.
113
―
136
7906
E.T.S.I. de
Telecomunicación
―
―
E.T.S.I. de Caminos,
Canales y Puertos
―
―
― 701 ―
E. P. S. Alcoy
67 378
389 161 2
E.T.S de Ingenieria de
15643 62768 5806 166
Informatica
E.T.S. Arquitectura
13226 4163 267
E.T.S. de Gestión de la
5535 494
Edificación
E.T.S. de Ingenieria del
98
Diseño
E.T.S. del Medio Rural
y Enología
E.T.S.I. Agronomos
E.T.S.I. de Caminos,
Canales y Puertos
E.T.S.I. de
Telecomunicación
E.T.S.I. de
Agronomos
―
―
E.T.S. de Ing. del
Diseño
E.T.S. del Medio
Rural y Enología
428 ― 385 564 ―
E.T.S. de Gestión de
la Edificación
―
7
ROAMING BETWEEN BUILDINGS IN A MONTH.
E.T.S. de Arquitectura
―
―
E.T.S de Inge. de
Informatica
Facultad De Bellas Artes
―
―
E. P. S. Gandia
―
E.T.S.I. Agronomos
―
E.T.S. Medio Rural Y Enologia
―
E.T.S. De Ingenieria Del Diseño
Facultad De Administración y Dirección
de Empresas
6
E.T.S.I. Industriales
E.T.S. Ingenieria del Diseño
―
―
1296 ―
―
E.T.S.I. Telecomunicación
―
―
E.T.S.I. Geodesica, Cartografica y Top.
3
E.T.S. Arquitectura
E.T.S. Gestion en la
Edificación
―
TABLE III.
E.T.S.I. Caminos, Canales y Puertos
E.T.S Ingeniería Informática
E.T.S. Arquitectura
E. P. S. Alcoy
E.T.S. Gestion en la Edificación
E. P. S. Alcoy
ROAMING BETWEEN BUILDINGS IN ONE DAY.
E.T.S Ingeniería Informática
TABLE II.
70
2
1231 3448
652
9314
502
670
7906
7531
15996 1572
8389
698
1837 2431
1914
3304
7644
3214
7804
334
3470
636
423
365
19
184
20
78
38
3104
3139
487
11437 572
965
708
26746
464
5811
557
577
958
7785 1054 1286 1181
767
979
1162
99
796
39
168
―
E. Roaming between Campuses
Figure 1 shows the values of the roamings carried out
among the different campuses of the Polytechnic University
of Valencia during a month. These campueses are Escuela
Politécnica Superior de Gandia (located in Gandia city),
Escuela Politécnica Superior de Alcoy (located in Alcoy
city), E.T.S. Medio Rural y Enología (located in one of the
main avenues of Valencia City).
480
145 11086 161
230
10
There were also many changes between “E.T.S.
Arquitectura”
building
and
the
“E.T.S.
de
Telecomunicación” building (17704 movements in a month).
The “E.T.S. Ingeniería del Diseño” and “E.T.S.I.
Industriales” buildings have also a lot of roamings. These
buildings have many users registered (see Table I). There are
many roamings between these buildings due to the likeness
of the studies. It seems that there are many subjects imparted
by the same department, so there are professors moving
between these buildings indistinctly.
“F. de Bellas Artes” building had less roamings. We
think that it is because “fine arts” students do not have as
much computers as the students of the other buildings.
Lastly, “E.T.S. Medio Rural y Enología”, the “E.P.S.
Alcoy”, and “E.P.S. Gandia” have very few movements
between them (there are 1350 roamings among “E.P.S.
Alcoy” and “E.P.S. Gandia”). This is because these buildings
belong to different campus located in different cities. There
are users that one day can be in a campus and, after some
hours, they are in another campus. In this case there is a
hard-roaming because the user loses the connection during a
large time because the user is travelling. If we take into
account the buildings of the Vera Campus we observe that
the most number of roamings are between all the campuses
and the Vera Campus.
1164
UPV
Campus de VERA
1844
2768
5497
E.T.S. del Medio Rural y Enología
Escuela Politécnica Superior de Alcoy
2
(Av. Vicente Blasco Ibáñez )
20
Escuela Politécnica Superior de Gandia
1350
Figure 1. Roaming between Campuses
The number of movements in a month between the
Gandia's campus and the Vera's campus are 5497. It has been
the highest value.
There are quite a lot of movements between Gandia’s
campus and Vera’s campus because they are relatively near
(around 56 km.), there is a good public transport
communication and many professors of Gandia's campus
also work in Vera's campus. The number of roamings among
“E.T.S. Medio Rural y Enología” and Vera's campus is 2768
and the roamings between Alcoy's campus and Vera's
campus is quite lower (1844 roamings). We can see in Figure
1 that Vera's campus is the campus that receives more visits.
This result is a prospective fact because Vera's campus is the
main campus of our university and most of the formalities,
solicitudes and administrative issues are made there.
Lastly, we can see that 1350 movements per month are
carrier out among "E.P.S. de Gandia" and "E.P.S. de Alcoy".
The main reason of these movements is the existence of
many professors that give classes in both campuses.
V.
TABLE IV.
PEOPLE BEHAVIOR
In this section, we will evaluate the users' movements by
day in the Vera's campus. We will also analyze the number
of changes carried out per user in a day.
In Figure 2, the 5 highest roaming values between
buildings are shown. The situation of more mobility is given
among “FI” and “GE”. It has a value of 4912. In this figure,
all of displacements shown have a higher value than 1834
movements/day. In this case, the movements are given
among “IND”-“BIB”, “DSIC”-“EI”, “ARQ”-“BIB” and
"CASALU”-“BIB”. With these data we can obtain some
information. E.g. the students of “E.T.S. Arquitectura” and
“E.T.S.I. Industriales” visit the university library more times
than the other students of the university. On the other hand,
there are many movements among the university library
building and the “Student's house” building (this building is
used by the students to study, to connect to Internet and to
develop any activity). This movement is due to the vicinity
of buildings (see Figure 2) and many users that are in one of
the buildings usually visit the other building. In Figure 3, the
10 highest roaming values are shown. In this case, we have
the 5 previous movements (see Figures 2) and 5 more. These
5 new displacements are carried out among “GE”-“DSIC”,
“GE”-“ARQ”, “FI”-“EI”, “ARQ”-“ASIC” and “ARQ”“CCP”. The minimum number of roamings of all
displacements seen in Figure 3 is 1483 per day. One of the
buildings that had more movements is “ARQ” (E.T.S.
Arquitectura). The main reason seems to be because some
services are offered in this building. For example, this
building has some snack bars, and there are some banks in
the bottom plant. We can also find a hairdresser, bookstores,
etc. and it can be found a great number of movements among
this place and other buildings.
Figure 4 represents all the Vera's campus movements
during one day. Almost all of buildings have users' mobility.
We can state that the wireless network of our university is
very robust. This network can support the mobility of all
users giving the appropriate service.
Lastly, we analyzed the number of changes per person.
This information is shown in table IV.
Expression (1) is used to extract the number of changes
per person.
Changes
Person
=
Total _ changes _ between _ buildings
registered _ people _ Building _ 1 + registered _ people _ Building _ 2
(1)
The buildings that have the biggest number of
movements per person are “E.T.S. de Gestión en la
Edificación” and “F. Informática” buildings. They obtained a
value of 0.797 movements per person. The movements
among “IND”-“BIB” and “ARQ”-“BIB” also possess a high
number of changes per person, 0.644 and 0.608 respectively.
VI.
CONCLUSION AND FUTURE WORK
In this article, we have presented a user mobility study
based on the roaming of the MACs in the wireless network
of the “Universidad Politécnica of Valencia”. We have
studied those buildings with higher mobility in the campus.
We have also studied the mobility between campuses.
Buildings with
Roaming
FI-GE
ARQ-BIB
IND-BIB
CCP-ARQ
FI-EI
GE-DSIC
FI-DSIC
EI-GE
TEL-ARQ
CASALU-ARQ
CHANGES/PERSON BETWEEN SOME BUILDINGS .
Registered
People
Building 1
3240
3858
3479
3145
3240
2920
3240
60
1409
―
Registered
People
Building 2
2920
―
―
3852
60
40
40
2920
3858
3858
Changes
Changes/person
4912
2344
2239
1772
1744
1705
1484
1380
1347
1308
0,797
0,608
0,644
0,253
0,528
0,576
0,452
0,381
0,256
0,339
Using the obtained data we can analyze the behavior of
students and professors in the campus. This study let us build
relocation bandwidth scenarios to increase the comfort of the
end user, i.e., if we note that a set of users go far to an area,
we could determine in detail where do they go. May be this
information could be used to reallocate services and
departments in the campus by changing their place or putting
a branch near the appropriate place.
This analysis will be the basis for a dynamic management
and control system of the wireless network. According to the
user mobility, the system will be able to give more
bandwidth in those areas where there are more users, and the
roaming system will be more efficient.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
IEEE
802.11-2007
Standard.
Available
at
http://standards.ieee.org/getieee802/download/802.11-2007.pdf
Wi-Fi Alliance. Available at http://www.wi-fi.org/ [accessed, July 14, 2009]
Bluetooth SIG. Specification of the Bluetooth system. Specification Version 1.1
(Feb. 2002). Bluetooth SIG.
E. Dahlman, B. Gudmundson, M. Nilsson and J.Sköld, “UMTS/IMT-2000
based on Wideband CDMA”, IEEE Communications Magazine, vol. 36, pp 7080, September 1998.
M. Garcia, J. Tomás, F. Boronat and J. Lloret, “The Development of Two
Systems for Indoor Wireless Sensors Self-location”. Ad Hoc & Sensor
Wireless Networks, vol. 8, num. 3-4, pp. 235-258. 2009.
Z Chen and W Chew, “DynaTrack: Dynamic Directional Instructions for
Tracking Assets and People in Office Environment”, Available in
http://hci.stanford.edu/srk/cs377a-mobile/project/final/chen-chew.doc
J.G.Markoulidakis, G.L.Lyberopoulos, D.F.Tsirkas, E.D.Sykas, “Mobility
Modeling in Third Generation Mobile Telecommunication Systems”, IEEE
Personal Communications, vol. 4, Issue. 4, pp 41-56. Aug. 1997.
G. Radinovic and K. Kim, “ Feasibilty study of RFID/ Wi-Fi / BlueTooth
wireless tracking system for underground mine mapping – Oklahoma”, In
procedings of “Incorporating Geospatial Technologies into SMCRA Business
Processes”, Atlanta, GA, March 25 – 27, 2008.
B. Issac, K. Hamid and C.E.Tan, “Wireless Mobility Management with
Prediction, Delay Reduction and Resource Management in 802.11 Networks”,
IAENG International Journal of Computer Science, Vol.35, Issue 3. August
2008.
R. Zhou, “ Wireless Indoor Tracking System (WITS)”, In proceedings of 12th
in a series of conferences within the framework of the European University
Information Systems Organisation (EUNIS), Tartu, Estonia. 28-30 June, 2006
T. Liu, P. Bahl and I. Chlamtac, “Mobility Modeling, Location Tracking, and
Trajectory Prediction in Wireless ATM Networks”,IEEE Journal Selected
Areas in Communications, Vol. 16, No. 6, pp. 922-936. August 1998.
C. Bettstetter, “Mobility modeling in wireless networks: categorization, smooth
movement, and border effects”, In proceedings of ACM SIGMOBILE Mobile
Computing and Communications Review (MC2R), vol.5 no.3, p.55-66. Rome,
Italy. 16-21 July 2001.
M. Smith, B. Shneiderman, N. Milic-Frayling, E.M. Rodrigues, V. Barash, C.
Dunne, T. Capone, A. Perer, and E. Gleave, "Analyzing (Social Media)
Networks with NodeXL", In Proceedings of the Fourth International
Conference on Communities and Technologies, C&T '09. The Pennsylvania
State University, USA. June 25-27, 2009.
Figure 2. 5 highest roaming values in Vera’s Campus
Figure 3. 10 highest roaming values in Vera’s Campus
Figure 4. All roaming values in the Vera’s Campus