Publication: Detección y estimación en redes de sensores inalámbricas con centro de fusión
Loading...
Identifiers
Publication date
2011-01-27
Defense date
2011-01-27
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
El presente proyecto trata de analizar las redes de sensores inalámbricas y la posible detección y estimación descentralizada a partir de la información obtenida por los sensores de dichas redes. Para ello se ha considerado un escenario en el cual los sensores están distribuidos de modo uniforme por toda el área monitorizada, en la que se pretende detectar la presencia de un blanco móvil. Concretamente, se ha implementado en Matlab, mediante el uso de una máquina de vectores soporte (SVM), un detector global de presencia que coincide con el centro de fusión de datos de la red. La regla de decisión global de este centro se genera en base a las decisiones locales tomadas por cada nodo-sensor. La necesidad de tomar las decisiones con carácter local, y de este modo obligar a la inferencia descentralizada, tiene como origen las limitaciones que se imponen a la capacidad del canal de comunicaciones entre los sensores y el centro de fusión, tratando de reducir al mínimo (hasta 1 bit) la información que cada nodo puede enviar de forma paralela al mencionado centro. Por ello, cada sensor dispone de una regla de decisión local establecida bajo el criterio de Neyman–Pearson, que nos garantiza unos niveles programables de probabilidad de error a partir de los cuales se obtienen umbrales de potencia con los que comparar la potencia recibida en un entorno con ruido Gaussiano. Posteriormente se modifica el escenario para dar la oportunidad a cada sensor de abstenerse de enviar información si considera que su información es confusa, aumentando indirectamente la información que recibe el centro de fusión. Una vez obtenidos y evaluados los detectores globales, el proyecto se completa con una primera aproximación a la estimación, tratando de comprobar en qué medida es posible estimar la posición y recrear la trayectoria una vez detectado el blanco. __________________________________________________________________________________________________________________
The main goal of this project is analyzing Wireless Sensor Networks and how decentralized detection and estimation can be performed in them thanks to the sensed information. In order to achieve this goal, an environment where sensors are uniformly distributed inside the supervised area has been considered. Detecting a potential mobile target crossing mentioned area is the ultimate purpose in this scenario. In particular a global detector which corresponds to the data fusion center has been implemented in Matlab using Support Vectors Machines (SVMs). In this way a global decision rule is generated based on local decisions made by each sensor-node. Due to the established limits on the communications channel capacity between the sensors and the fusion center, local decisions are required, as well as decentralized inference. The scheme is designed to try to reduce it as much as possible (up to 1 bit) information sent in a parallel way from each node to the mentioned fusion center. Hence, each single node performs its decision according to a local detection rule obtained under Neyman-Pearson criterion. This criterion guarantees fixed probability of error levels from which power thresholds are derived and used to be compared with the received power in a Gaussian noise environment. Later on, this scenario is modified to provide sensors the opportunity of refraining from transmitting information if they consider that their local information is confusing. This means that amount of information received by the fusion center increases indirectly. Once global detectors have been obtained and evaluated, the project is completed with a first approach to estimation. The purpose now is taking advantage of previous work to check whether it´s possible or not to estimate the target´s position and reconstruct the followed path when the target has already been detected.
The main goal of this project is analyzing Wireless Sensor Networks and how decentralized detection and estimation can be performed in them thanks to the sensed information. In order to achieve this goal, an environment where sensors are uniformly distributed inside the supervised area has been considered. Detecting a potential mobile target crossing mentioned area is the ultimate purpose in this scenario. In particular a global detector which corresponds to the data fusion center has been implemented in Matlab using Support Vectors Machines (SVMs). In this way a global decision rule is generated based on local decisions made by each sensor-node. Due to the established limits on the communications channel capacity between the sensors and the fusion center, local decisions are required, as well as decentralized inference. The scheme is designed to try to reduce it as much as possible (up to 1 bit) information sent in a parallel way from each node to the mentioned fusion center. Hence, each single node performs its decision according to a local detection rule obtained under Neyman-Pearson criterion. This criterion guarantees fixed probability of error levels from which power thresholds are derived and used to be compared with the received power in a Gaussian noise environment. Later on, this scenario is modified to provide sensors the opportunity of refraining from transmitting information if they consider that their local information is confusing. This means that amount of information received by the fusion center increases indirectly. Once global detectors have been obtained and evaluated, the project is completed with a first approach to estimation. The purpose now is taking advantage of previous work to check whether it´s possible or not to estimate the target´s position and reconstruct the followed path when the target has already been detected.
Description
Keywords
Redes de sensores inalámbricas, Redes de comunicación, Radiocomunicaciones