Repositorio Institucional de Documentos

Tesis de Fin de Máster Máster en Ingeniería de Sistemas e Informática Curso 2010/2011 Construcción de Mapas de Cobertura para Comunicaciones Inalámbricas Coverage Map Building for Wireless Communications Carlos Ernesto Rizzo Bobbio Director: Dr. José Luis Villarroel Salcedo Departamento de Informática e Ingeniería de Sistemas Escuela de Ingeniería y Arquitectura Universidad de Zaragoza Noviembre de 2011 ii Construcción de Mapas de Cobertura para Comunicaciones Inalámbricas RESUMEN Conocer ciertas características sobre cómo es la propagación de la señal en determinados entornos es de vital importancia para el uso efectivo de una red de comunicaciones inalámbrica. Dependiendo de la complejidad del medio podemos utilizar como guía uno o varios modelos de propagación, pudiéndose llegar a buenas aproximaciones sobre el comportamiento de la señal. Bien sea para desarrollar modelos (empíricos o deterministas) o validarlos, se requieren mediciones experimentales. En otros casos no se dispone de un modelo de propagación, por lo que la única opción radica en tomar mediciones prácticas. Cualquiera sea el caso, a través de la representación de estas mediciones en función de la posición obtenemos lo que se suele llamar un mapa de comunicaciones o mapa de cobertura. Situados en este contexto, en este trabajo se desarrollaron herramientas para la construcción de mapas de comunicaciones a gran escala y a pequeña escala. Pensando en una solución modular, se desarrollaron diversos módulos para el meta sistema operativo ROS y se implementaron en un vehículo real todoterreno, y en un robot Pioneer P3AT. Se realizaron pruebas en un ambiente de especial interés para el grupo RoPeRT (Robotics, Perception and Real Time) de la Universidad de Zaragoza: el túnel ferroviario de Somport, que conecta Francia con España. Se obtuvo un mapa de cobertura a gran escala de una sección de especial interés, de unos 2.5 km de largo con cambio de pendiente, y uno más detallado a menor escala de una sección de 1 Km, donde aparecen atenuaciones importantes. Se compararon los resultados con un modelo de propagación basado en “Ray Tracing” (trazado de rayos), desarrollado por Valenzuela (1993). Se obtuvieron similitudes como la existencia de un notable fading, pero a la vez diferencias que dan importancia a las mediciones realizadas, como la ubicación de este fading y diversas atenuaciones que no aparecen en las simulaciones. Se verificó la repetibilidad de estos fenómenos realizando diversos experimentos, inclusive en días diferentes, cuestión que no se ha sido tratada con importante énfasis en la literatura. También se encontró que, debido a variaciones transversales, aplicando una diversidad espacial muy superior a la de las tarjetas comerciales, podemos mejorar la calidad de señal en la mayoría del trayecto estudiado. Los resultados obtenidos pueden ser utilizados tanto para el despliegue óptimo de redes inalámbricas, hasta inclusive para el desarrollo de técnicas de navegación para equipos multi-robot manteniendo la comunicación. iii Coverage Map Building for Wireless Communications ABSTRACT Knowing certain characteristics about signal propagation in specific environments is crucial in order to make effective utilization of a wireless network. In some cases, one or several propagation models may be used, which can provide good approximations to signal behavior. To develop these models (either empirically or in a deterministic way) and evaluate its reliability, experimental measurements are required. In other cases, due to environment complexity, a specific model isn’t available and practical measurements must be done. In any case, representing these measurements as a function of position, we obtain what is usually called a communication map or coverage map. In this context, this work aims the development of tools for building communication maps. Based on a modular solution, various modules were developed for meta-operating system ROS, and were implemented over a real all-terrain vehicle, and a Pioneer P3AT robot. Tests were made on a special scenario of interest for University of Zaragoza research group RoPeRT (Robotics, Perception and Real Time): the Somport railway tunnel, which connects Spain with France. At first, a large scale coverage map of a zone of interest was obtained (about 2.5 Km long with a change in slope). Then, a smaller and more detailed map within this zone was built, where important fadings and phenomena appeared. Results were compared to a straight tunnel Ray Tracing model developed by Valenzuela (1993). We found important similarities, like the existence of a notable fading, but at the same time important differences, such as the exact location of this fading and other fadings that don’t appear in simulations. Repeatability of these phenomena was proved by performing different experiments, even on different days. This has not been significantly addressed in the revised literature. We also found that because of transversal variations, applying large scale diversity for antennas, good quality signal can be achieved in almost the entire studied area. Results obtained in this work can be use from optimal deployment of a wireless network, up to the development of multi-robot navigation strategies under communication constrains. iv AGRADECIMIENTOS Primeramente a mis padres, Tony y Chelo, a mis hermanos, Gabriel y Carolina; y a Eugenia Valentina, por el apoyo emocional y familiar muy necesitado, en visitas y por Skype, brindado a lo largo de todo el proyecto. En especial, a los profesores Carlos Sagues y José Luis Villarroel. Gracias a ustedes me encuentro en estos momentos aquí, y sin su tutoría el proyecto no se hubiese podido llevar a cabo. Al grupo RoPeRT, por brindarme la oportunidad de desarrollar este proyecto y cuyos comentarios y sugerencias fueron vitales para la realización exitosa del mismo. A José Luis, Danilo, Domenico y Luis, por toda la ayuda brindada en el famoso Túnel y aguantar horas de frío, oscuridad y polvo para ayudarme a probar las herramientas. A Daniel Cestari y Antonio Baez, por su ayuda técnica en ciertas partes. Finalmente, pero no menos importante, a mis compañeros especiales de siempre: Den, Mily, Ceci, Vane, Tony, Diego, Juan, Elías, Daniel, Alejandro “El Micri”, El Gordo Dassy; a los que no he visto mucho pero han servido de gran apoyo… Mariale; y a los nuevos: Roberto, Rosita, Boris, Danilo, Dome, Pablo, Luis, Mayte, El César, Henry, El otro Luis, Agus… Ha sido un placer conocerlos y gracias a ustedes este año ha sido genial. v Financiamiento El trabajo aquí presentado es financiado y se desenvuelve en el marco del proyecto “SISTEMAS MULTI-ROBOT EN APLICACIONES DE SERVICIO Y SEGURIDAD - TESSEO” (ref. DPI2009-08126), dentro del subprograma FPI-MICINN, referencia de la ayuda BES-2010- 038377. Funding The work presented has been supported by the FPI-MICINN Project “TEams of robots for Service and Security missiOns - TESSEO” (ref. DPI2009-08126) under research grant BES- 2010-038377. vi Contents List of Figures ………………………………………………………………………………………………………………... vii Chapter 1. Introduction ……………………………………………………………………………….................... 1 Chapter 2. Related Work ………………………………………………………………………………................... 3 2.1 Link Quality Metrics ………………………………………………………………......................... 3 2.2 Propagation models ………………………………………………………………......................... 4 Chapter 3. Test Scenario: The Somport Tunnel ……………………………………………………………… 7 Chapter 4. Tool Development for Coverage Map Building …………………………………………...... 12 4.1 Large Scale Model Tool ………………………………………………………........................... 12 4.1.1 Communication Module …………………………………………………......... 13 4.1.2 Odometry Module………………………………………………………............... 14 4.1.3 Test Platform ……………………………………………………......................... 14 4.2 Small Scale Model Tool ……………………………………………………............................. 15 4.2.1 Navigation module ……………………………………………………................. 15 4.2.2 Teleoperation Module ……………………………………………………............ 17 4.2.3 SLAM Module ……………………………………………………......................... 18 4.2.4 Test Platform ……………………………………………………......................... 19 Chapter 5. Results ……………………………………………………...................................................... 20 5.1 Large Scale Model ……………………………………………………..................................... 20 5.2 Small Scale Model ………………………………………………..………………....................... 23 5.3 Comparison between Models …………..………………………………….......................... 25 5.4 Maps Obtained……………………………………………….............................................. 27 5.5 Influence of vaults and galleries over propagation ……………………….……….......... 27 5.6 Spatial Diversity Analysis ……………………………………………………........................... 30 Conclusions ……………………………………………………................................................................ 33 Future Work …………………………………………………….............................................................. 35 References ……………………………………………………................................................................. 36 APPENDIX A: Repeatable Fadings in the “Fadings Zone” ………………………………................. 39 APPENDIX B: Influence of vaults over propagation ……………………………….......................... 45 APPENDIX C: Influence of galleries over propagation ………………………………....................... 49 Appendix D: Coverage Maps for the Fadings Zone ……………………………………….................. 52 Appendix E: Physical Maps for the Fadings Zone ……………………………………………................ 56 vii List of Figures Figure 1. Longitudinal and Cross section of the Somport Tunnel ................................................. 7 Figure 2. Small Vaults in the tunnel (a), and galleries (b). ............................................................ 8 Figure 3. Longitudinal fadings derived from Ray Tracing Model ................................................ 10 Figure 4. Transversal variation derived from Ray Tracing Model. .............................................. 10 Figure 5. Three Dimensional view for Longitudinal and Transversal variations in tunnels. ....... 11 Figure 6. Large Scale Tool Modules: Communication module: Transmitter (a) and Receivers (b). Odometry module (c), and antenna platform (d) .................................................................. 15 Figure 7. Online goal planning algorithm .................................................................................... 16 Figure 8. Basic SLAM algorithms structure ................................................................................. 18 Figure 9. Final assembly for Pioneer P3AT. ................................................................................. 19 Figure 10. RSSI (dBm) vs Position (m). Journey 1 and 2. PTx=1dBm. .......................................... 21 Figure 11. RSSI (dBm) vs Position (m). Journey 3. PTx=-19dBm.................................................. 22 Figure 12. “Zone of Interest” and ”Fadings Zone”. ..................................................................... 23 Figure 13. RSSI (dBm) vs Position (m). Journey 4 and 5. PTx=1 dBm. ......................................... 24 Figure 14. RSSI (dBm) vs Position (m). Journeys 1 to 5, and Ray-Tracing model. ....................... 25 Figure 15. RSSI (dBm) vs Position (m). Repeatable fadings location for Large Scale and Small Scale models. ........................................................................................................................... 27 Figure 16. First 500 meters of the Fadings Zone (a), and Last 500 m of this zone (b) ................ 27 Figure 17. Influence of vaults over propagation in the second half of the fading Zone. ............ 28 Figure 18. Influence of galleries over antenna A (a) and antenna B (b) in the second half of the Fadings Zone. .......................................................................................................................... 29 Figure 19. Commercial application for antenna diversity. .......................................................... 30 Figure 20. Received power for each antenna in the 4th Journey and differences between them. ................................................................................................................................................. 31 Figure 21. RSSI scale for coverage maps. .................................................................................... 31 Figure 22. Coverage map for one antenna. ................................................................................ 32 Figure 23. Coverage map with Spatial Diversity.......................................................................... 32 1 Chapter 1 Introduction Communication networks are crucial for the proper performance of many systems, where an opportune and adequate information exchange is always needed. In many situations, a wired architecture is not feasible, either because is not possible or convenient. This is the case for multi-robot cooperative teams, VANETs (Vehicular Ad-Hoc Networks), communication systems in hostile environment, among others. In these cases, a wireless solution is needed, and this type of networks has become popular in the last years because of its versatility and ease of implementation and deployment. Among these technologies, WiFi, which operates under the 802.11 standards suite, has become one of the most popular. One reason is because it works on the ISM band (Industrial, Scientific and Medical), which requires no license. Besides, equipment needed to deploy this type of networks has become less expensive because of its massive utilization and commercialization. This has made WiFi preferable over other standards, such as WiMAX, despite they have different characteristics (El-Sayed et al. 2008). To make optimal utilization of these wireless networks it’s desirable to have knowledge of both the nature of the signal as well as the environment where it is spreading. A practical manner to do this is creating coverage maps or communication maps, where signal power as a function of position is represented for a specific environment. Several studies have been made about signal propagation. Antenna polarization, operation frequency, material characteristics and geometry of the environment have all been taken into account. Nevertheless, there are few papers that describe the tools used to derive these studies, and according to our literature revision, there is no such tool designed to facilitate the construction of these maps, and do so autonomously. In human-made case, it’s a consuming and prone error job. This is the reason why in most of the cases a sensor network is deployed, or a robot is used but with previous knowledge of the place (Holland et al. 2006; Lun-Wu Yeh et al. 2009). These type of solutions are not suitable for some environments. For instance, if the coverage area is too large, it may result economically unfeasible to deploy a sensor network. Because of hostile conditions human presence may be compromised, so it is better to be 2 performed in an autonomously or semiautonomously manner. Also, previous map of the site is not always available, so if you add the ability to Simultaneous Localization and Mapping (SLAM), the solution would be much more complete. In such a context, the aim of this work is to develop a tool that fulfills the above requirements for building coverage maps, and test it on a real scenario. To do this, we explored ways to acquire and analyze characteristics of the wireless signal, as well as a study of propagation phenomena with emphasis in a specific environment: tunnels. After testing the tools, a large scale coverage map of a special zone in the Somport railway tunnel was obtained (about 2.5 Km long with a change in slope). Then, a smaller and more detailed map within this zone was built, where important fadings and phenomena appeared. Results were compared to an implementation of a straight tunnel Ray Tracing model, developed by Valenzuela (1993). Important similarities appeared, like the existence of a notable fading, but at the same time important differences, such as the exact location of this fading and other fadings that don’t appear in simulations. Repeatability of these phenomena was proved by performing different experiments, even on different days. This has not been significantly addressed in the revised literature. We also found that because of transversal variations, applying large scale diversity for antennas, good quality signal can be achieved in almost the entire studied area. The utilization of these tools could influence areas such as the study of phenomena as consistency and repeatability of the signal in certain environments, propagation models development, efficient and optimal network deployment, communication studies in VANETs, optimization and upgrade of coverage maps, localization with communication metrics, and even suggesting navigation techniques in multi-robot teams under connectivity constrains. This work is structured as follows. In the next section, a state of the art revision is presented, as well as the theoretical bases about signal propagation with the aim of coverage map building. In section 3, the test scenario is described, along with a tunnel propagation model. Later, in section 4, we present the tools developed in order to build communication maps (at a large scale and a small scale). In section 5, results obtained after testing both tools in the Somport tunnel are presented, followed by an analysis of the phenomena observed. Finally, we present our conclusions and future work. 9 Gri = Square root of transmit and receive antenna gain product in direction of ith ray. r = Path length of LOS direction. ri = Path length of ith ray. Ri = Reflection coefficient of the ith ray. And, 𝜑𝑖=2𝜋𝛥𝑙𝑖 𝜆 (2) Where Δli is the distance between the LOS and the ith path. If the ray is horizontally polarized respect to the surface, the reflection coefficient is given by 𝑅ℎ𝑖 =cos(𝜑𝑖)− (𝜀−𝑠𝑖𝑛2(𝜑𝑖))1 2 cos(𝜑𝑖)+ (𝜀−𝑠𝑖𝑛2(𝜑𝑖))1 2 (3) In the other hand, if the ray is vertically polarized respect to the surface, the reflection coefficient is given by 𝑅𝑣𝑖 =𝜀cos(𝜑𝑖)− (𝜀−𝑠𝑖𝑛2(𝜑𝑖))1 2 𝜀cos(𝜑𝑖)+ (𝜀−𝑠𝑖𝑛2(𝜑𝑖))1 2 (4) Where 𝜑𝑖 is the angle of incidence for the ray, and 𝜀= 𝜀𝑟𝑖 −𝑗60𝜎𝑖 𝜆𝑖 (5) Being εri and σi the relative permittivity and conductivity of the media. After implementing the algorithm, we made a simulation with the “Zone of Interest” values: 5.5 m wide, 2.5 Km in length, operating frequency of 2.4 GHz. A maximum number of 6 reflections was considered. The relative permittivity εr = 5 and relative conductivity σ =0.01 s/m were selected, as values widely used in the literature. At first it can be seen longitudinal variations, depicted in figure 3 for a transversal distance of three meters from the right wall. The transmitter is located at the beginning (0 m in length), at 3 meters from the right wall, and at 2.5 meters from the left wall. 10 Figure 3. Longitudinal fadings derived from Ray Tracing Model. During the first 600 meters we can see several narrow fadings. After that area, more notable wider fadings appear. The signal power follows a logarithmic decay. Regarding the transversal variations, they are illustrated in figure 4 for a longitudinal distance of 1000 m, as an example. The transmitter is located in the same place as in the above example. Figure 4. Transversal variation derived from Ray Tracing Model. It can be observed that cross-section variations are also of great importance, with a difference greater than 20 dB between local maximums and minimums in simulations. This should be taken into account when designing the tools for communication mapping. 11 Finally, in figure 5, we depict an example in three dimensions, where transversal and longitudinal variations can be observed together within 20 meters in length (X position from 900 to 920 meters from the transmitter). Again, the transmitter is located at 3 meters from the right wall. Figure 5. Three Dimensional view for Longitudinal and Transversal variations in tunnels. In summary, in the case of tunnel scenarios, longitudinal and transversal variations are of great importance, and should be taken into account to design tools for coverage mapping. In the next section we present the tools developed for creating communication maps, both in a large scale and in a small scale. 12 Chapter 4 Tool Development for Coverage Map Building A specific propagation model for the Somport tunnel is not available. In addition, it has galleries, vaults and a change in slope that would hamper the development of one. Therefore, building a coverage map with experimental measurements appears to be the most feasible solution in order to study propagation in this environment. The atmosphere in the tunnel is hostile: low temperatures, darkness, dust and air currents, which make it difficult for humans to work comfortably. Its length makes it not economically feasible to deploy a sensor network to study propagation phenomena. It seems to be the best option to place sensors on a mobile platform and move it. There isn´t a physical map of the tunnel available. By building this, we would be able to create the coverage map associated to the real map, as well as determine the effect of vaults and galleries over propagation. The first goal was to create a large scale coverage map for the “Zone of Interest”. Later, it became necessary to build a model on a smaller scale in certain regions of interest, and a second tool was developed. 4.1 Large Scale Model Tool The first solution considered was to place wireless sensors over the robotics platforms available in the lab, Pioneer P3AT robots, but due to the length of the tunnel and the speed limit of these robots, it would take a huge amount of time to build the map. For this reason, a real vehicle was used. Because of its modularity and integration with other solutions for RoPeRT group, it was decided to work on ROS meta-operating systems, which runs over Linux. We developed a communication module, responsible for frame broadcasting, reception and processing. In 13 order to associate signal quality to position, an odometry module was designed. Navigation is by a human operator for being the case of a real vehicle. 4.1.1 Communication Module In order to obtain signal quality in certain points we need at least a transmitter and a receiver, as well as a quality metric to represent and analyze. Regarding the quality metrics, RSSI and PDR were selected. To represent the received power we chose the RSSI, as it’s reported by the network card. Because it can be obtained the RSSI of received packets only, we need an additional metric to know how many frames have been received. The PDR was implemented for this purpose, by adding a sequence number in the frame header. Knowing how many packets have been sent and lost, PDR can be computed and used to validate RSSI measurements. Besides, there is a strong correlation between PDR and RSSI, as stated by Vlavianos et al. (2008) and Dhananjay et al. (2003). In this manner, we are combining two metrics to obtain an idea about link reliability and signal quality. Concerning the hardware, for the emitting node, a PcEngines ALIX3D3 board was selected. It is equipped with an Atheros™ based chipset wireless card, and it’s sealed and designed for hostile environment operation. It runs Linux operating system, and was programmed to broadcast frames that later on will be received by a node to be analyzed. For the receiving node, a laptop running meta-operating system ROS over Linux was chosen, equipped with an Atheros™ network card. “ROS (Robot Operating System) provides libraries and tools to help software developers create robot applications. It provides hardware abstraction, device drivers, libraries, visualizers, message-passing, package management, and more. ROS is licensed under an open source, BSD license.” (http://www.ros.org/wiki/). Previously, a driver was developed to obtain the RSSI and floor noise values from the network card. A ROS executable communicates with the network card to obtain these values, and received power in dBm is computed through the sum of these in the case of Atheros cards (Bardwell 2002). In table 1 can be observed a summary of the main characteristics for both nodes, and they are shown in figure 6. Transmitter Receiver Model PcEngines ALIX3D3, Atheros wireless card Dell Laptop, Intel Core 2 Duo T9300 2.5 GHz Processor, 2 GB RAM memory, Atheros wireless card Operating System Linux ROS over Linux Working frecuency 2.142 Ghz Antenna polarization Vertical Frame size 250 bytes Transmission speed 6 Mbps Frame periodicity 5 ms Table 1. Transmitter and Receiver nodes characteristics. 14 4.1.2 Odometry Module To associate signal quality to position, an odometry module was designed. Regarding the hardware, it’s mainly composed by an infrared light emitter-receiver. Reflective tape was placed over the rear wheels. Therefore, we can perceive an electronic pulse each time the tape passes in front of the sensor. The traveled distance is calculated based on the wheel circumference length, the number of reflective tape bands, and hence the number of received pulses. In figure 6 we can observe the final assembly for this module. 4.1.3 Test Platform All of these modules were integrated on a real vehicle, in this case, an all terrain. In order to study transversal variations of the signal we decided to place 3 antennas, located at the ends and center of the vehicle. To avoid the interference the vehicle may cause, a platform was installed in the roof, as observed in figure 6. Three laptops were used. One is in charge for both odometry and communication modules, while the others are responsible for communication modules only. Data is synchronized through the sequence number in the emitted frames. 15 (a) (b) (c) (d) Figure 6. Large Scale Tool Modules: Communication module: Transmitter (a) and Receivers (b). Odometry module (c), and antenna platform (d) 4.2 Small Scale Model Tool With the aim of studying certain zones in more detail, such as the effect of vaults and galleries over propagation, arises the need to build a physical map of the surrounding environment and associate link quality to that map. Because the length in this case is smaller, it’s feasible to use Pioneer P3AT robots. Again, a modular solution was proposed. The communication module is the same as the one described above. Nevertheless, the navigation is not responsibility of a human operator now, and to this end a navigation module was implemented. Finally, to build the environment map, a SLAM (Simultaneous Localization and Mapping) module was used. All of these modules were designed for ROS operating system. 4.2.1 Navigation module This module is composed by two layers: local navigation and reactive navigation. For the local navigation, an algorithm was developed to plan local goals based on the distance to 16 the tunnel walls. For reactive navigation, an implementation in ROS of the Dynamic Window Approach, described by Fox et al. (1997), was used, with some minor changes in the objective function. a) Online goal planning To obtain a reliable coverage map and make a precise study about transversal variations of the signal, the robot must be able to travel in a straight line, at a constant distance from walls or following the center of the tunnel. Because of the width of the tunnel is not always constant, as in the case where a vault is found, the algorithm was designed to detect the walls and filter galleries and vaults. The program was structured in 4 phases: 1. Wall Detection: based on the information from a laser scan sensor, a program receives these measurements and, using the Hough Transform, detects the straight lines derived from these. These happen to be the tunnel walls, galleries and vaults. 2. Vault filtering: once the lines are detected, a filtering process is applied to eliminate the small vaults and detect the pair of tunnel walls. This is done with information about the number of votes obtained by the Hough Transform. 3. Goal line: after obtaining the pair of walls, a line is generated in a way that passes through the center of both lines, or at a certain constant distance from the walls. The navigation route converges to this line. 4. Goal setting: once the goal line is determined, a goal is placed over it, at a certain distance from the robot (in the longitudinal dimension). Trigonometric calculations are computed so that the goal is located according the actual robot coordinate system. This process is repeated every second (f = 1 Hz), a value set in simulations, yielding good results in the real implementation. In this manner, we keep the robot traveling in a straight line along the tunnel. The whole algorithm can be visualized in figure 7. Figure 7. Online goal planning algorithm 17 b) Reactive navigation In order to avoid obstacle collision, an implementation of the Dynamic Window Approach (Fox et al. 1997) was used. This approach was considered the most appropriate for this environment because it avoids oscillatory behaviors in corridors or tunnels. This problem appears using the potential field approach, as stated by Fox et al. (2007). Besides, this method considers the vehicle dynamics and is suitable for robots traveling at “high speeds”. The original method describes the procedure to obtain linear and angular speed in two phases: 1. Search space, where possible velocities are defined. These velocities are those which are collision free, and able to be reached within a short interval of time (given the robot accelerations). 2. Optimization, where the objective function G(v, ω) = α * heading + β * distance+ γ * velocity (6) Is maximized, being “heading” the orientation of the robot, “distance” the distance to the closest obstacle, and “velocity” the speed of the robot. For the meta-operating system ROS, there is an implementation of this approach (http://www.ros.org), with a difference in the objective function to maximize. In this case, the function is given by: G(v, ω) = α * (path) + β * (goal) + γ * (obstacle_cost) (7) Where “path” is the distance to the local navigation route, “goal” is the distance to the local goal, and “obstacle_cost” is a measure about how much the controller should attempt to avoid obstacles. To determine the best α, β and γ, three values were assigned to each one: high, medium and low. A model of the Pioneer P3AT was tuned in Stage™ and simulations with all 27 possible combinations were made in a corridor scenario. Then, the parameters in the real robot where tuned around simulation results. The values obtained are depicted in table 2. Parameter Value α 0.4 β 0.3 γ 0.03 Table 2. Parameters used for ROS DWA implementation 4.2.2 Teleoperation Module In order to achieve a semi-autonomous or guided navigation for the robot to be used in other scenarios, the system was equipped with a PS3JOY module (http://www.ros.org), which allows to drive the robot using a commercial PlayStation™ game controller. 18 4.2.3 SLAM Module With the purpose of associating signal quality with the position of vaults and galleries, or reference in a better way communication loss risk zones, it would be useful to build a physical map of the environment. To this end, a SLAM (Simultaneous Localization and Mapping) module was implemented in the system. In a basic sense, SLAM algorithms deal with map representation and the estimation technique (for the actual localization and map creation). A summary of these can be observed in figure 8. Figure 8. Basic SLAM algorithms structure Nowadays, algorithms based on particle filters as estimation technique are becoming important. There is an implementation in ROS, called SLAM-gmapping, capable to realize SLAM with this estimation technique. It uses the “Rao-Blackwellized particle filter” approach, described by Grisetti et al. (2005) and Grisetti et al. (2007). This approach creates the map in 4 phases: 1. Sampling: new particles are generated according to the particles of the previous step and the proposed distribution. 2. Weighting: each particle is weighted according to actual and previous information. 3. Re-sampling: particles (or substitutes) are drawn according to its weight. After this process, all particles weight the same. 4. Map estimation: for each particle a map is estimated, which is processed according to its observation history. 25 For these experiments, the worst PDR obtained was greater than 95%, which gives a high degree of reliability for measurements presented. Again, both graphs appear to have the same pattern, but in this case we obtained a difference less than 5 dB between both Journeys. In the next section, a comparison on the “Fadings Zone” between all experiments conducted will be performed. 5.3 Comparison between Models Representing the fading zone for all Journeys (1 to 5) and the Straight Tunnel Model, figure 14 is obtained. Figure 14. RSSI (dBm) vs Position (m). Journeys 1 to 5, and Ray-Tracing model. Qualitatively, it can be observed the same pattern in all experiments conducted. In this study, we will remark repeatable phenomena and not the differences between the models (for example, highlighting repeatable fadings and not all of them). For instance, it can be noticed a 26 considerable fading in all models. In the experiments performed, this fading is located around 1450 m from the transmitter, while in the model implemented in section 3 with the Somport Tunnel values, it’s located at 1850 m. However, this model includes direct LOS between transmitter and receiver, a condition not fulfilled in this “Zone of Interest”. Also, vertical reflections, vaults and galleries were ignored. Hence the importance of creating an experimental coverage map. Other phenomenon observed in the experiments but not in the Ray Tracing model is two zones, composed of two fadings each. These fadings have a separation between them of about 75 to 100 m. The zones are located around 1300 and 1700 m respectively. In figure 15 we depict these phenomena for Journey 1 and 4, in order to demonstrate that it appears on both large scale and small scale models. The notable fading that appears in both plots of figure 15 is the fading explained above (located at 1450 m). Complete analysis about all Journeys is presented in Appendix A. Figure 15. RSSI (dBm) vs Position (m). Repeatable fadings location for Large Scale and Small Scale models. 27 5.4 Maps Obtained Because of processing and storage space limitations, SLAM module couldn’t be used online in the whole Journey, so Laser and Odometry data was recorded while navigating through the “Fadings Zone” and then, the SLAM algorithm was ran offline with these measurements. Maps obtained for the first and second half of this zone are presented in figure 16. They are depicted in a bigger scale in Appendix E, where vaults and galleries can be appreciated in a better way. (a) (b) Figure 16. First 500 meters of the Fadings Zone (a), and Last 500 m of this zone (b) And in table 6, data obtained from both Odometry and Maps are compared In Fadings Zone Robot’s Travel Time (sec) 1549 Robot´s Avarage Speed (m/s) 0.64 Odometry Map Length (m) 988 SLAM obtained Map Length (m) 1013 Error 25 meters ; 2.46% Table 6. Odometric vs Obtained Maps data. Odometry data is used to associate RSSI values to position. SLAM maps obtained are used to associate this data to the real physical map. It can be seen a difference of about 25 meters (over the 1000 meters traveled) between the robot’s odometry and the maps, which represents an approximate error of about 2.5%, not very significant for our referencing purposes. 5.5 Influence of vaults and galleries over propagation Once the maps were obtained, we analyzed the influence of vaults and galleries over propagation, if any. By matching the map with signal quality as a function of position and highlighting the place where small vaults take place (for one antenna in the second half of the “Fadings Zone”), figure 17 is obtained. 28 Figure 17. Influence of vaults over propagation in the second half of the fading Zone. Remaining antennas are shown in Appendix B for the whole Journey. In this case, it’s difficult to define a clear influence pattern. No evident fading or gain point can be seen due to vaults. This may be because antennas were placed at 2 meters in height, while vaults are only 1.5 meters high. Analysis varying antenna height is left as a future work. In this case, results serve to dismiss a significant effect of vaults over propagation. Similarly, by matching galleries location with signal quality as a function of position for two antennas, we obtained figure 18. 29 (a) (b) Figure 18. Influence of galleries over antenna A (a) and antenna B (b) in the second half of the Fadings Zone. 30 Remaining antennas are shown in Appendix C for the whole Journey. In figure 18 (a), it can be seen a 4 dB attenuation where the galleries take place. Nevertheless, this occurred only for one of the three antennas, so results are not conclusive as to associate these attenuations to the galleries. 5.6 Spatial Diversity Analysis Spatial diversity is one mechanism or technique used in communication systems with the aim of overcoming multipath propagation distortions. It consists on placing multiple antennas on the receiver, strategically spaced between them. The idea is to select the antenna with the best signal at any time (Rappaport, 2002, chapter 6). Commercially, there are solutions available to apply this type of mechanisms, for example, to routers or WiFi repeaters. According to or product review, antennas are separated from millimeters, up to a wavelength distance (which is 12.5 cm for 2.4 GHz devices). In figure 19 we depict one commercial antenna array for applying spatial diversity to a WiFi repeater. Figure 19. Commercial application for antenna diversity. Nevertheless, in the model developed in section 3, it can be seen that in order to obtain a good quality signal, separation between antennas must be greater than commercial cards diversity systems. In the transversal section depicted in figure 3, we can notice that the distance between a local maximum and a minimum is at least 50 centimeters. Hence, spatial diversity should be in that order of distance. This is the reason why a diversity of about 60 cm was considered. In figure 20 we depict the received power for the three antennas in the whole “Fadings Zone” (Journey 4), in order to denote the RSSI difference between all of them. 31 Figure 20. Received power for each antenna in the 4th Journey and differences between them. It can be seen areas where the difference is up to 17 dBm, which may break a wireless link and cause communication loss, for example, with a base station. On this basis, we proceed to build a coverage map for this section of interest. At first, signal is discretized in intervals depending on its quality, and a color is assigned according to it so to get a visual appreciation of the hazardous areas in terms of position. This can be seen in figure 21. Figure 21. RSSI scale for coverage maps. Firstly, we build a coverage map taking the RSSI measurements from one antenna in the second half in the Fadings Zone. Map obtained is depicted in figure 22. 32 Figure 22. Coverage map for one antenna. Later, a coverage map taking into account spatial diversity was built; this is, taking the best value among the three signals. Map is depicted in figure 23. Figure 23. Coverage map with Spatial Diversity. It’s clearly evident the importance of spatial diversity. In our study, we can assert that when applying a spatial diversity in the order of 60 cm (almost 5 wavelengths distance) there aren’t hazardous zones (represented in red in the maps), and dangerous zones can be converted into safer regions. However, a more detailed study about diversity must be performed in order to determine the optimal distance between antennas. The full map for the Fadings Zone is presented in Appendix D. 33 Conclusions In this work we have developed tools that allow study of propagation phenomena at a large and small scale. In the large scale case, odometry and communication modules for meta-operating system ROS were developed, and then implemented on a real all-terrain vehicle. For small scale maps, communication, navigation and SLAM modules were implemented on a Pioneer P3AT robot. For navigation, due to the peculiarity of the environment, we developed an algorithm based on Hough Transform. This would allow autonomous navigation in straight lines at a constant distance from walls in any parallel wall environment, such as corridors and tunnels. Because of the modularity of the solution, it can be easily improved by adding modules or upgrading the existing ones. Also, it can be easily migrated to other type of solutions with minor changes. Both tools were tested in a real environment, an area of interest in the Somport railway tunnel. There is no specific model for this tunnel, and has some features that could hinder the development of one. A large and small scale models based on measurements were obtained and then compared to a Straight Tunnel Ray-Tracing model, obtaining similarities and differences. First, an important fading was found in all maps. In the case of the built maps, at 1450 meters, while in the model appears at 1820 meters. Using our tools we also discovered a pattern of two fadings, separated approximately 100 meters. This pattern appears again 400 meters away from its first appearance. This phenomenon is not reported by the Ray-Tracing model, and may be due to all dismissed information when implementing the model (such as ignoring vertical reflections, vaults and galleries), or more important, by implementing a Line Of Sight model when in the tunnel this condition was not fulfilled. In any case, the model was implemented only to obtain an idea about what phenomena to find. With the aid of physical map building, we were able to dismiss any significant influence of vaults and galleries over propagation. Also, we noticed the importance of spatial diversity in this type of environments. It was found that with a spatial diversity far beyond commercial network cards (60 cm between antennas), good quality signal can be obtained and communication loss hazardous zones can be avoided in the studied area. Finally, repeatability of phenomena, an issue that has not been significantly addressed in the revised literature, was proved by performing different types of experiment, with different tools, in different days, and at different transmission powers, obtaining similar results from all of them. 34 As a general conclusion, we can assert that when involving robotics and communications in tunnel environments, either for surveillance purposes, propagation studies, hostile areas exploration and so on, both transversal and longitudinal variations of signal quality must be taken into account. Regarding diversity, we demonstrated that when applying a “big scale” spatial diversity (beyond commercial network cards diversity), signal quality can be significantly improved in the studied scenario. Finally, coverage maps can help to have an overall idea about propagation in the environment, easily and visually, with not many details. Repeatability of propagation was demonstrated; hence, coverage maps obtained can be useful for several situations. 41 JOURNEY 3 Antenna A: Figure A4. Repeatable fadings for Antenna A in Journey 3. Antenna B: Figure A5. Repeatable fadings for Antenna B in Journey 3. 42 Antenna C: Figure A6. Repeatable fadings for Antenna C in Journey 3. JOURNEY 4 Antenna A: Figure A7. Repeatable fadings for Antenna A in Journey 4. 43 Antenna B: Figure A8. Repeatable fadings for Antenna B in Journey 4. Antenna C: Figure A9. Repeatable fadings for Antenna A in Journey 4. 44 And a summary where we depict location and distance between fadings, in meters, is depicted in table A.1. 1st fading 2nd fading Separation 1st – 2nd 3rd fading 4th fading Separation 3rd - 4th Separation 2nd - 3rd Journey 1 1221 1283 62 1725 1790 65 442 Journey 1 1194 1294 100 1732 1797 65 438 Journey 1 1176 1275 99 1709 1756 47 434 Journey 3 1175 1265 90 1716 1784 68 451 Journey 3 1222 1301 79 1717 1787 70 416 Journey 3 1201 1314 113 1694 1740 46 380 Journey 4 1209 1289 80 1769 1857 88 480 Journey 4 1212 1297 85 1781 1860 79 484 Journey 4 1191 1273 82 1784 1869 85 511 Table A1. Repeatable Fadings Location in Every Journey. All units are in meters. 45 Appendix B: Influence of Vaults over Propagation In this case, we highlighted the places were vaults take place, in order to find a pattern of fadings or gain points, if any. Six graphs are presented: one for each of the three antennas, during the first half and second half of the 4th Journey. As a reminder, SLAM module was implemented in the Small Scale tool, which was used in the “Fadings Zone” and not in the entire “Zone of Interest”. For this reason, vaults influence over propagation is analyzed only in this zone (1 Km in length). For the first half of the Fadings Zone, graphs obtained are depicted in figures Antenna A: Figure B1. Vaults location for Antenna A in the first half of Journey 4. 46 Antenna B: Figure B2. Vaults location for Antenna B in the first half of Journey 4. Antenna C: Figure B3. Vaults location for Antenna C in the first half of Journey 4. 47 For the second half of the Fadings Zone, graphs obtained are depicted in figures Antenna A: Figure B4. Vaults location for Antenna A in the second half of Journey 4. Antenna B: Figure B5. Vaults location for Antenna B in the second half of Journey 4. 48 Antenna C: Figure B6. Vaults location for Antenna C in the second half of Journey 4. It couldn’t be found a pattern of influence from vaults over propagation. Casually, some fadings appear where a vault is located, but in the other cases signal remains equal or has a gain. So no specific phenomena are attributed to vaults. Regarding the magnitude of these fadings and gain points, they are from less than 1 dB up to almost 10 dB. Again, these differences are attributed to multipath propagation, instead of vault influence. 49 Appendix C: Influence of galleries over propagation We have highlighted the places were galleries take place, in order to find a pattern of fadings or gain points, if any. Three graphs are presented: one for each of the three antennas, during the second half of the 4th Journey. The reason of analyzing these effects only in the second half of the Fadings Zone is because no gallery appeared in the first half. As a reminder, SLAM module was implemented in the Small Scale tool, which was used in the “Fadings Zone” and not in the entire “Zone of Interest”. For this reason, galleries influence over propagation is analyzed only in this zone (1 Km in length). For the second half of the Fadings Zone, graphs obtained are depicted in figures Antenna A: Figure C1. Galleries location for Antenna A in the second half of Journey 4. 50 Antenna B: Figure C2. Galleries location for Antenna B in the second half of Journey 4. Antenna C Figure C3. Galleries location for Antenna C in the second half of Journey 4. 57 And zooming the maps, both vaults and galleries can be appreciated clearly, depicted in figure E.2 and E.3 respectively. Figure E.2. Vaults in the Physical Map. Figure E.3. Gallery in the Physical Map.