Print

Mayer, Linda M.

Computer Science

2004

M.S.

Advances in integrated circuit (IC) fabrication, micro-electromechanical devices (MEMs), wireless communications, and digital signal processing (DSP) have facilitated the development of miniature intelligent sensing nodes which are wirelessly linked. Each node is composed of sensing, communication, processing, and power devices. When linked together, the nodes form a Wireless Sensor Network (WSN). These WSNs are groups of smart, self-organizing sensor nodes which can be applied to industrial automation, battlefield surveillance, border surveillance, habitat monitoring, structural monitoring, autonomous navigation and traffic control without the cost and tediousness of wires. Energy usage remains a challenge to WSN s since being untethered requires the use of batteries which have limited energy. Even with tiny devices, every computation and transmission as well as reception depletes the energy source until operation is no longer possible. Decreasing the sensor sampling rate may help to slow energy consumption within a network. However, applications such as target tracking require a minimum sampling rate to achieve desirable performance. In this work, we apply a traditional tracking algorithm, the Kalman filter, to WSNs using a MAT LAB simulation to track four trajectories in uniformly-distributed and randomly-distributed networks of nodes. The contributions of this research determine the accuracy and robustness of several Kalman filter techniques when applied to a WSN. In particular, we consider the tracking performance given two constraints unique to these networks. First, we reduce the rate at which tracking data is collected in order to reduce energy usage. Second, we reduce node density to represent sensor failure that occurs as the network ages. Overall, we find that our use of adaptive methods does not significantly improve tracking performance during these network constraints, and thus may not justify the additional computations in an already energy constrained system. We conclude noting several aspects of this work that can be further researched and improved.