Publication: Design and optimization of wireless sensor networks for localization and tracking
Loading...
Identifiers
Publication date
2014-05-13
Defense date
2014-04
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Knowledge of the position of nodes in a WSN is crucial in most wireless
sensor network (WSN) applications. The gathered information needs to be
associated with a particular location in a specific time instant in order to
appropiately control de surveillance area. Moreover, WSNs may be used for
tracking certain objects in monitoring applications, which also requires the
incorporation of location information of the sensor nodes into the tracking
algorithms. These requisites make localizacion and tracking two of the most
important tasks of WSN.
Despite of the large research efforts that have been made in this field,
considerable technical challenges continue existing in subjects areas like data
processing or communications. This thesis is mainly concerned with some
of these technical problems. Specifically, we study three different challenges:
sensor deployment, model independent localization and sensor selection.
The first part of the work is focused on the task of sensor deployement.
This is considered critical since it affects cost, detection, and localization accuracy
of a WSN. There have been significant research efforts on deploying
sensors from different points of view, e.g. connectivity or target detection.
However, in the context of target localization, we believe it is more convenient
to deploy the sensors in views of obtaining the best estimation possible
on the target positioning. Therefore, in this work we suggest an analysis of
the deployment from the standpoint of the error in the position estimation.
To this end, we suggest the application of the modified Cram´er-Rao
bound (MCRB) in a sensor network to perform a prior analysis of the system
operation in the localization task. This analysis provides knowledge
about the system behavior without a complete deployment. It also provides
essential information to select fundamental parameters properly, like
the number of sensors. To do so, a complete formulation of the modified
information matrix (MFIM) and MCRB is developed for the most common
measurement models, such as received signal strength (RSS), time-of-arrival
(ToA) and angle-of-arrival (AoA). In addition, this formulation is extended
for heterogeneous models that combine different measurement models. Simulation
results demonstrate the utility of the proposed analysis and point
out the similarity between MCRB and CRB.
Secondly, we address the problem of target localization which encompasses
many of the challenging issues which commonly arise in WSN. Consequently,
many localization algorithms have been proposed in the literature each one oriented towards solving these issues. Nevertheless, it have seen
tahta the localization performance of above methods usually relies heavily
on the availability of accurate knowledge regarding the observation model.
When errors in the measurement model are present, their target localization
accuracy is degraded significantly.
To overcome this problem, we proposed a novel localization algorithm
to be used in applications where the measurement model is not accurate or
incomplete. The independence of the algorithm from the model provides
robustness and versatility. In order to do so, we apply radial basis functions
(RBFs) interpolation to evaluate the measurement function in the entire
surveillance area, and estimate the target position. In addition, we also
propose the application of LASSO regression to compute the weigths of the
RBFs and improve the generalization of the interpolated function. Simulation
results have demonstrated the good performance of the proposed
algorithm in the localization of single or multiples targets.
Finally, we study the sensor selection problem. In order to prolong the
network lifetime, sensors alternate their state between active and idle. The
decision of which sensor should be activated is based on a variety of factors
depending on the algorithm or the sensor application. Therefore, here we
investigate the centralized selection of sensors in target-tracking applications
over huge networks where a large number of randomly placed sensors are
available for taking measurements.
Specifically, we focus on the application of optimization algorithms for
the selection of sensors using a variant of the CRB, the Posterior CRB
(PCRB), as the performance-based optimization criteria. This bound provides
the performance limit on the mean square error (MSE) for any unbiased
estimator of a random parameter, and is iteratively computed by
a particle filter (in our case, by a Rao-Blackwellized Particle Filter). In
this work we analyze, and compare, three optimization algorithms: a genetic
algorithm (GA), the particle swarm optimization (PSO), and a new
discrete-variant of the cuckoo search (CS) algorithm. In addition, we propose
a local-search versions of the previous optimization algorithms that
provide a significant reduction of the computation time. Lastly, simulation
results demonstrate the utility of these optmization algorithm to solve a
sensor selection problem and point out the reduction of the computation
time when local search is applied. ---------------------------------------------------
Las redes de sensores se presentan como una tecnología muy interesante que ha atraído considerable interés por parte de los investigadores en la actualidad [1, 109]. Recientes avances en electrónica y en comunicaciones inalámbricas han permitido de desarrollo de sensores de bajo coste, baja potencia y multiples funciones, de reducido tamaño y con capacidades de comunicación a cortas distancias. Estos sensores, desplegados en gran número y unidos a través de comunicaciones inalámbricas, proporcionan grandes oportunidades en aplicaciones como la monitorización y el control de casas, ciudades o el medio ambiente. Un nodo sensor es un dispositivo de baja potencia capaz de interactuar con el medio a través de sus sensores, procesar información localmente y comunicar dicha información a tus vecinos más próximos. En el mercado existe una gran variedad de sensores (magnéticos, acústicos, térmicos, etc), lo que permite monitorizar muy diversas condiciones ambientales (temperatura, humedad, etc.) [25]. En consecuencia, las redes de sensores presentan un amplio rango de aplicaciones: seguridad en el hogar, monitorización del medio, análisis y predicción de condiciones climáticas, biomedicina [79], etc. A diferencia de las redes convencionales, las redes de sensores sus propias limitaciones, como la cantidad de energía disponible, el corto alcance de sus comunicaciones, su bajo ancho de band y sus limitaciones en el procesado de información y el almacenamiento de la misma. Por otro parte, existen limitaciones en el diseño que dependerán directamente de la aplicación que se le quiera dar a la red, como por ejemplo el tamaño de la red, el esquema de despliegue o la topología de la red..........
Las redes de sensores se presentan como una tecnología muy interesante que ha atraído considerable interés por parte de los investigadores en la actualidad [1, 109]. Recientes avances en electrónica y en comunicaciones inalámbricas han permitido de desarrollo de sensores de bajo coste, baja potencia y multiples funciones, de reducido tamaño y con capacidades de comunicación a cortas distancias. Estos sensores, desplegados en gran número y unidos a través de comunicaciones inalámbricas, proporcionan grandes oportunidades en aplicaciones como la monitorización y el control de casas, ciudades o el medio ambiente. Un nodo sensor es un dispositivo de baja potencia capaz de interactuar con el medio a través de sus sensores, procesar información localmente y comunicar dicha información a tus vecinos más próximos. En el mercado existe una gran variedad de sensores (magnéticos, acústicos, térmicos, etc), lo que permite monitorizar muy diversas condiciones ambientales (temperatura, humedad, etc.) [25]. En consecuencia, las redes de sensores presentan un amplio rango de aplicaciones: seguridad en el hogar, monitorización del medio, análisis y predicción de condiciones climáticas, biomedicina [79], etc. A diferencia de las redes convencionales, las redes de sensores sus propias limitaciones, como la cantidad de energía disponible, el corto alcance de sus comunicaciones, su bajo ancho de band y sus limitaciones en el procesado de información y el almacenamiento de la misma. Por otro parte, existen limitaciones en el diseño que dependerán directamente de la aplicación que se le quiera dar a la red, como por ejemplo el tamaño de la red, el esquema de despliegue o la topología de la red..........
Description
Keywords
Wireless sensor networks, Localization, Tracking