Performing Data Aggregation on Wireless Sensor Motes


Abstract (Summary)
A Wireless Sensor Network (WSN) consists of a large number of sensor nodes dispersed over a target area for monitoring and these sensing devices are interconnected in the form of an ad-hoc network using wireless radios. A sensor node senses a physical parameter and transmits the same in a multi-hop fashion towards a single collector node known as the sink or Base Station (BS). At the sink, collected data from all the sensors is analyzed and appropriate action is taken. Again, most of the physical parameters obtained by the environmental sensors exhibit a property of spatial-correlation, that is, smooth variation over space. This property of sensors can be exploited to remove redundant data sensed by neighboring sensors. This can be done by employing a fast and robust data aggregation protocol at intermediate sensors that combines its own sensed data with the data reported from its neighbors, before transmitting the final compressed packet to the BS. In this thesis, a tree-based aggregation scheme for WSNs, TREG (Tree-based Regression) is implemented on wireless sensor motes and its effectiveness is evaluated. The experimental results obtained in the TinyOS programming environment validate the results obtained previously from the software simulation. TREG builds a distributed spanning (aggregation) tree where each tree node performs multivariate polynomial regression and instead of raw data, transmits only the coefficients of a polynomial P, reducing the volume of data considerably. Since, computation consumes negligible energy as compared to data transmission, a primary objective of WSN programming is to conserve energy at each sensor through additional computation that could result in reduced radio communication. In TREG, some of the sensors only sense data and report them to their nearest tree nodes. Each of the tree nodes then implements an aggregation scheme in a distributed manner, thereby reducing the overall communication overhead as compared to a strictly centralized scheme. The sensor motes which are tree nodes, have been programmed to perform their own polynomial regression in nesC, a component-oriented variant of the C language, used by the TinyOS operating system. One of the main challenges in implementing a high level scheme in low level programs is the extremely resource constrained hardware of motes. The current generation of sensor mote typically has an 8-bit micro-controller, few KB of Program ROM and Data RAM respectively. Thus, any scheme must not overstretch the limited processing and memory storage of the mote hardware. Some of the main challenges encountered during the course of the implementation are detailed in this thesis.
Bibliographical Information:


School:University of Cincinnati

School Location:USA - Ohio

Source Type:Master's Thesis

Keywords:aggregation computational overhead mote polynomial tinyos


Date of Publication:01/01/2008

© 2009 All Rights Reserved.