ETH Zurich :
Computer Science :
Pervasive Computing :
Distributed Systems :
Student Projects :
Adaptive Model Selection Library for TinyOS (M)
The availability of small and cheap wireless sensing devices increased significantly in the past few years and large scale sensor network deployments begin to appear. Such a large number of sensors deployed in the real-world allow to accurately monitor a variety of physical phenomena, like weather conditions (temperature, humidity,
), noise levels, traffic jams or room occupancy in public buildings. Reporting this sensor data to a central sink using the sensor nodes on-board radio represents a significant communication overhead and may rapidly exhaust nodes batteries. The development of adequate data collection techniques able to reduce the amount of data sent throughout the network is therefore recognized as a key factor for allowing long-term, unattended network operation.
In the context of our research in wireless sensor networks we developed an efficient, light-weigth algorithm to reduce the amount of data a single sensor node needs to report to the central sink. Our approach relies on time series prediction techniques to make a sensor node send collected sensor readings to the sink only if the prediction data, know by both node and sink, deviates from the actual reading by more than a pre-specified error threshold. To achieve high data reduction rates, we developed an adaptive model selection scheme (AMS) that allows to select, at each instant in time, the model that offers the highest achievable communication savings. First experimental results demonstrated the feasibility and efficiency of our approach, which has already been published by a renouwned international journal*.
The main goal of this master thesis consists in implementing the above described algorithm on a commercially available wireless sensor networks platform (Tmote Sky) and evaluating its performances in a real-world, multi-hop sensor network deployment. To this scope, the data collection algorithm will be implemented as an application on top of the tinyOS open source operating system, the de-facto standard OS for wireless sensor network platforms. Applications in tinyOS are written in nesC, a simple C dialect, and ad-hoc java tools enable a users PC to easily receive and send data to and from the networked sensors. An elegant, modular implementation of the algorithm could be included as a openly available library in the tinyOS distribution. Furthermore, the algorithm evaluation, performed on a real-world sensor network deployment, will provide a basis for further scientific publications.
*Yann-Aël Le Borgne, Silvia Santini, Gianluca Bontempi: Adaptive Model Selection for Time Series Prediction in Wireless Sensor Networks, International Journal for Signal Processing, Special Issue on Information Processing and Data Management in Wireless Sensor Networks, Elsevier, December 2007.
Student/Bearbeitet von: Philipp Küderli
Contact/Ansprechpartner: Silvia Santini