Caso de Éxito. TITULO: Diseño óptimo de la red inalámbrica y sistema automático de reubicación de usuarios para balancear la carga de la red

Transcription

1 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 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 1% de cobertura en todos sus campus. Actualmente cuenta con 575 APs para ofrecer servicio a más de 41. 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 5 edificios repartidos en 2 kilómetros cuadrados.

2 Tras el despliegue de red, se han añadido sistemas en la red que permiten aprovechar de manera eficaz dicha red Desarrollo de un sistema de balanceo por usuario para Telefonía IP en redes de área local inalámbricas [4] [5] 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 la Universidad Politécnica de Valencia [6] 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.. Valores Añadidos: Sistema de estudio de la movilidad en el Campus Utilizando redes inalámbricas [7] 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. 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) de Septiembre de 23 [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 23. Benalmádena, Málaga (España). 8-1 de Septiembre de 23 [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 24 (ISWCS'4). Port Louis (Isla Mauricio) de Septiembre de 24. [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

3 and Technologies (UBICOMM 28). Valencia (España). 29 septiembre - 4 octubre de 28. [5] Miguel Edo, Miguel Garcia, Carlos Turro and Jaime Lloret. IP Telephony development and performance over IEEE 82.11g WLAN. The Fifth International Conference on Networking and Services (ICNS 29). Valencia (España) de Abril 29. [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 29). Valencia (España) de Abril 29. [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 29). Sliema (Malta) de Octubre de 29.

4 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 jlloret@dcom.upv.es 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 82.11b, [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 jjlopez@dcom.upv.es 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.

5 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] 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. 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 2 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 82.11b 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 Distancia (m) 1 pared 2 paredes 3 paredes 4 paredes 5 paredes Canales que causan interferencias Canales con riesgo a interferencias Canales con poca o ninguna interferencia

6 Atendiendo a la tabla anterior, si tenemos que seleccionar el máximo número de canales simultáneos sin ninguna interferencia, usaremos los canales: Sin embargo si toleramos una ligera interferencia que en la práctica no degrada el sistema usaremos: 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. Fig. 2. Distribución óptima de 4 canales en un área horizontal. planta 4 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 5 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. 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 1% 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 1%. 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. 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 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, "82.11 wireless networks : the definitive guide", Ed. O'Reilly, Sebastopol, 22 [2] B. O'Hara, "The IEEE handbook : a designer's companion", IEEE Press, New York, 1999 [3] [4]

7 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. 3 > S > -5 dbm 5 >S > -7 dbm 7 >S > -8 dbm S < -8dBm Fig. 4. Resultados de cobertura en diferentes edificios del campus

8 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, 4673 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, Introduction Since the market appearance of the wireless local area networks (WLAN) based on the 82.11b 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

9 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]: L(dB) = Lo + 1n log(d) + kf + IW + Lms Lo = power losses (db) at a distance of 1m (4 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 = Ptx ap + Gtx + Grx 2log d 2 log (4π/λ) Σ Lp i Lms (1) Pr = Received power Ptx ap = Transmitted power by the access point Gtx = Transmitter gain Grx = Receiver gain d = Distance between transmitter and receiver 2 log (4π/λ) = 4 db for 2.4 GHz ΣLp i = 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 2 db. In order to assure coverage, the worse case scenario corresponding to 2 db was taken in this work. Next, the received power at a distance of 1 meter from the wireless access point is calculated: Prx 1m = Ptx ap +G tx +G rx -2log(4π/λ) Lms (2) Ptx ap = Transmitted power by the access point G tx and G rx = Antenna gains (at the transmitting and receiving sides). 2log(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: Prx 1m = -4 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. 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, 1 meter apart from the wall and a series of measures are taken at points using a wireless card connected to a laptop, and signal monitoring software. Using equation (1), (2) and (3), it is deduced: P r =Prx 1m -2log(d) ΣP i. So it is possible to calculate the loss through the first wall. L -1 = -4 2 log (d) Pr 1 Pr 1 is the received power at point 1 and d=2 in this case. In order to compute the loss at the second wall: L 2-3 = -4 2 log (d) Pr 3 L -1 Where Pr 2 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 Loss [db] Mean 4.36

10 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 =Prx 1m 2 log (d) n.lp 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: n P = rx1 m 2 log d P L And this other expression provides the maximum distance as a function of the number of crossed walls: d = p Prx n L 1m p Pu 2 1 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 -8 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) 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 2 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: u Model Lp (db) 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: Building Mean Model Sports 7 1 DCAN EUITI I DOEFFC I D E B Architecture E Computer Sc. Faculty F DSIC G EUI ETSA Block ETSIA Block A - 2E M Fine Arts Mean value 6.29 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.

11 4. Interferences between channels The 82.11b 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 Ch 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: must be used. However, if a slight interference that does not degrade the system in the practice is tolerated the channels 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 = Ptx ap + Gtx + Grx Lprop Ptx ap = Power transmitted by the AP. Gtx = Transmitter gain. Grx = Receiver gain. Lprop = 2log(4πd/λ) propagation losses 6. Measurement results Despite there are available WLAN design software [1], 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 5 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

12 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 1% 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 1%. 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, "82.11 wireless networks: the definitive guide", Ed. O'Reilly, Sebastopol, 22 [2] B. O'Hara, "The IEEE handbook: a designer's companion", IEEE Press, New York, 1999 [3] [4] [5] C.C. Chiu and S.W. Lin, "Coverage prediction in indoor wireless communications," IEICE Trans. Commun.,vol. E79-B,no.9,pp ,Sep [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 18-2, 1993, [8] [9] José Maria Hernando Rábanos Transmisión por radio, Editorial C.E. Ramón Areces S.A., 1997 [1] 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 (Spring, 1995). a) Access point place choice.

13 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. 3 > S > -5 dbm 5 >S > -7 dbm 7 >S > -8 dbm S < -8dBm Fig. 4. Covering results in different campus buildings

14 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 4622 Valencia, SPAIN 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 4 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, 82.11, 2.4 GHz I. INTRODUCTION Since the launch of the wireless local area networks (WLAN) based on the 82.11b and 82.11g 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]: Where: L(dB) = Lo+1n log(d)+σk i F i +ΣI j W j +Lms (1)

15 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 (4.2 db at 2.44 GHz frequency) n = attenuation variation index with the distance (n=2) d = distance between transmitter and receiver K i = number of floors of kind i in the propagation path F i = attenuation of one floor of kind i. I j = number of walls of kind j in the propagation path W j = 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 [1] [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, 1 meter apart from the wall and a series of measures are taken at points 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. L -1 = Lo 2 log (d) Pr 1 + Lms 1 (2) Pr 1 is the received power at point 1 and d = 2 meters in this case. In order to compute the loss at the second wall: L 2-3 = Lo 2 log (d) Pr 3 L -1 + Lms 2 - Lms 1 (3) 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 Loss [db] Mean (L p ) It is notable to remark that we obtain a value adjusted to that obtained on references [1] 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 =Prx 1m 2 log (d) ΣL i p i (4) Prx 1m = Power received at 1meter d = Distance between the transmitter and the receiver ΣL i p i = 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: Prx 2log d P 1m u n = L P And this other expression provides the maximum distance as a function of the number of crossed walls: i i (5) Where Lo = 4.2 db and Pr 2 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 d = Prx n m 2 1 L P P 1 i i u (6)

16 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 -8 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) 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 2 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 5 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 Mean Sports 7 DCAN 5.25 EUITI 6.41 I DOEFFC I D E 6.6 1B Architecture E Computer Sc. Faculty F DSIC G EUI 6.32 ETSA Block ETSIA Block A 2E M Fine Arts 8.21 Mean value 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 5 studied are shown.

17 Figure 5. Selection of the access point placement taking into account wall losses. Figure 6. Access point placement and coverage area. To assure a suitable reception for all wireless cards, we limit the power range at -8 dbm for 82.11b, and -68 dbm for 82.11g, 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 7. Example of the prediction maps obtained using the proposed model: [-3,-5 dbm] ]-5,-7 dbm] ]-7,-8 dbm] < -8 dbm TABLE IV. Model prediction, compared with field measure and error, for two different buildings. Building A Model predict. Measured Error , Building B Model predict. Measured Error

18 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 5 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 82.11b and 82.11g 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, 21 [2] M.S. Gast, wireless networks: the definitive guide, Ed. O'Reilly, Sebastopol, 22 [3] B. O'Hara, The IEEE handbook: a designer's companion, IEEE Press, New York, 1999 [4] C.C. Chiu and S.W. Lin, Coverage prediction in indoor wireless communications, IEICE Trans. Commun.,vol. E79-B,no.9,pp ,Sep [5] 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 [6] Rappaport, Theodore S., Wireless Communications: Principles and Practice, Prentice Hall Publications, NJ, [7] R. A. Valenzuela, "A Ray Tracing Approach to Predicting Indoor Wireless Transmission", IEEE Vehicular Technology Conference, Secaucus NJ, May 18-2, 1993, [8] T. Frühwirth, J.R. Molwitz and P. Brisset, Planning Cordless Business Communication Systems, IEEE Expert Magazine, Special Track on Intelligent Telecommunications, February 1996 [9] 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. 269 [1] John C. Stein, Indoor Radio WLAN Performance Part II: Range Performance in a Dense Office Environment, Harris Semiconductor (Intersil) [11] Robert Wilson, Propagation Losses Through Common Building Materials 2.4 GHz vs 5 GHz. Magis networks Inc. s.pdf [12] 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 [13] J. S. Davis II, Measurements in Cory Hall at UC Berkeley. [14] Dan Dobkin, Indoor Propagation and Wavelength, WJ Communications, 22 [15] R.F. Rudd, building penetration loss for slant-paths at l-, s- and c- band, International Conference on Antennas and Propagation, March 23 [16] 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.

19 A User-Balanced System for IP Telephony in WLANs Miguel Garcia 1, Diana Bri 2, Carlos Turró 3, Jaime Lloret 4 Polytechnic University of Valencia, Spain 1 migarpi@posgrado.upv.es, 2 diabrmo@posgrado.upv.es, 3 turro@cc.upv.es, 4 jlloret@dcom.upv.es 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, 12% from 25 to the first half of 26, reaching to 155,41 users in June 26 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 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 55 access points and

20 7 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 [1] 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 18 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 4 years old. It is currently formed by 3 campuses and the main campus has around 5 buildings spread out in 2 km 2. There are around 4, researchers and educational personnel, around 1,5 staff and around 35, 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 5 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 5 hours in a regular indoor access point of the Polytechnic University of Valencia. It gives the number of users use to vary from 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.

21 Switch Number of Users Acces Point A Acces Point B Switch Acces Point C Database Server Hours Figure 1. Number of users in an AP during 5 hours. Router(config)#dot11 association mac-list 77 Router(config)#access-list 77 deny $mac-list.. Router(config)#access-list 77 permit.. ffff.ffff.ffff Router(config)#exit Router#clear dot11 client $mac-list 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 AP i, 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 AP i. Then it is disassociated the IP telephony device A from AP i. 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 IP phone IP phone Figure 2. Network architecture of our user-balanced system. 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 AP i 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.

22 Calculate Throughput Action needed? Yes Select Phone to move Phone A Find AP Phone A is attached to AP i Block A association at AP i Disassociate A from AP i Receive Data No Yes Is A attached to any AP? Wait a little time Unblock A association at AP i Mark A as unmovable No Figure 5. User-balanced algorithm Acces Point A Database Server Initial Connection to an Access Point Acces Point B Figure 4. Association process. 5. Deployment and measurements 5.1. Test bench Steps to Association: Client sends probe. APs send Probe Response. Client evaluates AP response selects best AP. Client sends authentication request to select AP (A). AP (A) confirms authentication. Clients sends association request to select AP (A). AP (A) confirms association and registers client in the database server. 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 11 and Cisco Catalyst 295T-24 switches with 1BaseT 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 Re-association time measurements In different Access Points. This section gives the time taken by an IP telephony device to reassociate to another access point. We have Database Server Acces Point A Acces Point B Steps to Re-association: Database server send a disassocite IP telephony device from AP (A) Adapter listens for beacons from APs. Adapter evaluates AP beacons, selects best AP. Adapter sends association request to selected AP (B). AP (B) confirms association and registers adapter. 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 3 packets per second 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 15 th 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 1 shows the same behaviour, obtaining a peak of 41 packets per second.

23 Bytes/s Bytes/s Seconds Figure 7. Bytes/s in the network Seconds Figure 9. Bytes/s in the network Bandwidth measurements 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 AP 1, talking with 7 IP telephony devices that were connected with AP 2. 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 31 th second) we obtain an average load of Bytes/s. The average number of packages obtained in figure 12 when the network converges is around 94.5 packets/s 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. Packets/s Packets/s Seconds Figure 8. Packets/s in the network Seconds Figure 1. Packets/s in the network. When the network has converged, there are around 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.

24 Bytes/s Bytes/s Seconds Figure 11. Bytes/s in the network Seconds Figure 13. Bytes/s in the network. Packets/s Packets/s Seconds Figure 12. Packets/s in the network Seconds Figure 14. Packets/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 , Hong Kong, April 21. [2] O. Hersent, D. Gurle, and J-P. Petit. IP Telephony. Addison Wesley, 2. [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 , NY, April 28. [4] M. Hassan, A. Nayandoro, M. Atiquzzaman. Internet telephony: services, technical challenges, and products. IEEE Communications Magazine, vol.38, no.4, pp.96-13, April 2. [5] D.P. Hole, F.A. Tobagi, "Capacity of an IEEE 82.11b wireless LAN supporting VoIP" IEEE International Conference on Communications 24, vol.1, pp , 2-24, Paris (France), June 24. [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 April 26. [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 24, Pages [8] T. Henderson, D. Kotz, and I. Abyzov, The changing usage of a mature campus-wide wireless network. 1th Annual international Conference on Mobile Computing and Networking, Philadelphia, USA, Sep 26 Oct 1, 24. [9] István Szabó. On call admission control for IP telephony in best effort networks. Computer Communications, vol. 26, Issue 4, pp , March 23. [1] 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 , April 24. [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 , 3-6 Sept. 26 [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 23, Benalmádena, Málaga (Spain), September 23. [13] ITU-T Rec. G.711. General Aspects of Digital Transmission Systems Terminal Equipments - Pulse Code Modulation (PCM) of Voice Frequencies [14] Jim Van Meggelen, Leif Madsen & Jared smith. Asterisk: The future of the telephony. O Reilly. USA. 25. [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 27), Valencia (Spain), October 14-2, 27.

25 IP Telephony development and performance over IEEE 82.11g WLAN Miguel Edo 1, Miguel Garcia 2, Carlos Turro 3 and Jaime Lloret 4 Universidad Politécnica de Valencia, Camino Vera s/n, 4622, Valencia (Spain) 1 miedmon@epsg.upv.es; 2 migarpi@posgrado.upv.es; 3 turro@cc.upv.es; 4 jlloret@dcom.upv.es 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 82.11g 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 IEEE82.11b, which is used to supply a theoretical bandwidth of 11Mbps, to IEEE82.11a and IEEE82.11g [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 82.11g 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 82.11b/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

26 experimental assessment of the IEEE 82.11b standard to support VoIP on a wired network has been carried out. Nevertheless, none of the studies aforementioned has dealt with IEEE 82.11g 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 AP Cisco Aironet 113AG IP PBX Asterisk 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 5 APs are in the main Campus (Campus de Vera). The access points are Cisco Aironet 113AG Series APs which use IEEE 82.11a/b/g standard and provide speeds up to 18Mbps. They are installed to allow the users a continuous coverage as they roam throughout a facility. They incorporate the 82.11i IEEE-Compliant standard (WPA2-Certified and WPA-Certified) which allows interoperability with other manufacturers. The coverage of each access point varies between 3 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 s, 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, LPC1, Speex. 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 Test Bench IP Phone Nokia E65 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 82.11g 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 82.11g, 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 [1]. 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.

27 Delay (ms) Call1 Call5 Call9 Call3 Call Time (s) Fig 2. Measures of the call delay Delay (ms) Delay average (ms) Max Delay (ms) Call1 Call2 Call3 Call4 Call5 Call6 Call7 Call8 Call9 Fig 3. Delay average and maximum delay per call. 6 Call1 Call5 Call3 Call7 6 Jitter average (ms) Max Jitter (ms) 5 Call9 5 Jitter (ms) Jitter (ms) Time (s) Fig 4. Measures of the call jitter Call1 Call2 Call3 Call4 Call5 Call6 Call7 Call8 Call9 Fig 5. Jitter average and maximum jitter per call second call In order to analyze the performance and quality of calls we have made 9 calls of 3 second each one, testing the delay, jitter, bandwidth and packet loss. The packets are captured from the IP Phone to the Asterisk PBX Delay tests 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 3 second long and none of them exceed 5ms 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 2ms. Performing the overall average of 9 calls we obtain a delay of ms and the average maximum delay is ms Jitter testing In Figure 4 we can see the results of tests done about Jitter. In neither of the cases, the jitter exceeds 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 ms jitter and ms of maximum jitter average Lost packets testing 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 packages: a % 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 for the 9 3-second calls.

28 Lost Packets Call1 Call5 Call9 Call3 Call Time (s) Fig. 6. Extent the packets lost in the calls. Lost Packets Lost packets Packets/Call average Call1 Call2 Call3 Call4 Call5 Call6 Call7 Call8 Call9 Fig 7. Transmitted packets and lost packets per call. 25,% 2,% Lost Packets average (%) 6 5 Call1 Call2 Lost Packets 15,% 1,% Delay (ms) ,% 1,% Call1 Call2 Call3 Call4 Call5 Call6 Call7 Call8 Call9 Fig 8. Percentage of lost packets per call 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 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 2 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 ms very similar to data obtained in section In this test, the maximum average delay is ms whereas it was ms in the 3-second calls between the two phones using the same Access Point Jitter Testing In figure 1 we can see the data obtained in the analysis of jitter in the roaming test. As shown in the figure, jitter is around.5 and 1.5 ms very similar to tests carried out in section with the difference in Time (ms) Fig 9. Extent of the delay in Roaming this case that the maximum jitter average of ms increases to 6.3ms. On the other hand, the average jitter has not increased over the average obtained in section 4.2.2: ms this time, which is a positive development 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 3-second calls. In this case we have a lost packet average per call of 8. % 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 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 2 Kbps with an average of Kbps in the IEEE 82.11g 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.

29 7 6 Call1 Call Call1 Call Jitter (ms) 4 3 Lost Packets Time (s) 2 Fig 1. Measures on Roaming Jitter Time (ms) Fig. 11. Lost packets in roaming 1 BW (Kbps) Time (s) Fig 12: Effective bandwidth in the IEEE 82.11g network. IP BW (Kbps) Call1 Call5 Call9 Call3 Call Time (s) Fig. 13. Measures of the 3-second call bandwidth 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 3-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 On the other side, in the moments where there is no loss of packets we can see that the bandwidth is around 8 and 11 Kbps with an average of 89Kbps. In figure 14 we can see the average bandwidth per call that is always between 1 and 12Kbps: concretely 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 3-second call bandwidth. On this occasion, the bandwidth has a mean of Kbps very similar to 89Kbps in the 3 second s average calls. On the other hand, what does vary significantly is the maximum bandwidth average which raises from 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 82.11g network of the Polytechnic University of Valencia is 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 Conclusions In conclusion, we can say that the IEEE 82.11g 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].

30 14 IP BW average (kbps) Max IP BW (kbps) IP BW (Kbps) IP BW (Kbps) Call1 Call2 Call1 Call2 Call3 Call4 Call5 Call6 Call7 Call8 Call9 Fig. 14. Measure of the maximum bandwidth average and maximum bandwidth per call. 3 Call Roaming Call Delay ms ms Jitter 1.41 ms 1.6 ms IP BW 89 Kbps 9.15 Kbps Lost Packet % 12.39% 8. % 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 3-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 , Hong Kong, April Time (s) Fig. 15. Roaming bandwidth measures 3s Call Roaming Call Delay 39.1 ms ms Jitter 3.92 ms 6.3 ms IP BW 11.4 Kbps 128 Kbps Table 2. Average of the maximum values. [2] IEEE It is available in The Working Group for WLAN Standards [3] Asterisk. It is available enwww.asterisk.org [4] D.P. Hole, F.A. Tobagi, "Capacity of an IEEE 82.11b wireless LAN supporting VoIP" IEEE International Conference on Communications 24, vol.1, pp , 2-24, Paris (France), June 24. [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 April 26. [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 24, Pages [7] G. Agreda, J. Gaviria. "EvaluaciónExperimental the Capacity of IEEE 82.11b support for VoIP." Converging technologies applied to mobile computing. Memories I2ComM 26. Pp [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 23, Benalmadena, Malaga (Spain), September 23. [9] RFC 3261, SIP: Session Initiation Protocol. [1] Wireshark It is available in [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, 28. UBICOMM'8. Publication Date: Sept Oct On page (s):

31 Multicast TV over WLAN in a University Campus Network Alejandro Canovas 1, Fernando Boronat 2, Carlos Turro 3, Jaime Lloret 4 Polytechnic University of Valencia EPSG - IGIC Institute 1,2,4 ASIC-UPV 3 Ctra. Nazaret-Oliva, S/N Camino de Vera, S/N Grao de Gandia Valencia alcasol@epsg.upv.es 1, {fboronat 2,jlloret 4 }@dcom.upv.es, turro@cc.upv.es 3 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 82.11, 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 24 ([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 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-

32 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 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 82.11b 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 4, 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

33 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 82.11g wireless network, with over 5 APs deployed, which provides 1% 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 32x24 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 Gbps Figure 2. Network communications for UPVTV 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 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 networks. We use location from the network side, as proposed by the authors in [1], 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. 1Gbps IV. OUTLINE OF MULTIMEDIA HANDOVER SCHEME 1 Gbps 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 As stated previously, our proposal solves packet loss on hard IEEE roaming by combining location and network control in order to achieve near-seamless multicast roaming without requiring any 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

34 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 82.11n 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 32x24 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.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 AP1 Figure 4. Message flow for our proposal 5. EVALUATION Figure 5. Measurement scenario