Wirelesses sensor networks are fulfilling their promise of ubiquitous computing through their ease of deployment and reductions in size, weight, power, and cost. A fast growing application for these small form-factor and computationally powerful wireless devices is in environmental monitoring, health monitoring, infrastructure monitoring, battlefield target tracking, and other applications of Body-Area-Networks and Personal-Area-Networks. Due to the ubiquity of these devices and their computational abilities, applications in the signal processing domain using networks of these devices are being conceived, such as adaptive filtering and beamforming. The signal processing tasks involved, however, are computationally intensive and stress the energy resources of any single computational unit. A single computational unit, thus, may not have the necessary power, memory, and computational resources to complete the often-repetitive signal processing tasks, when the problem space is considerably larger than the computational devices.
Collaborative distributed processing on resource-constrained wireless systems has the potential to increase the aggregate computational capabilities of a network, speed up the computation, and decrease the energy consumption per node. Thus, exploring distributed signal-processing mechanisms on such systems is necessary.
This research effort develops a kernel for distributed computing and signal processing on resource-constrained wireless sensor networks. The kernel is formed of distributed versions of the LU decomposition, QR decomposition, and fast Fourier transform. It also includes a smart compact tagging mechanism for efficient, collision-free transmission; an energy and channel-aware formulation for task mapping and allocation for extending the network's lifetime; and a user-friendly scalable simulator for task scheduling and mapping.
The contributions of this research are at the algorithm and the middleware levels. At the algorithm level, two new algorithms were developed to meet the need for scalability. The algorithms are: (1) a tile-based distributed QR decomposition, and (2) a distributed FFT with low communication overhead. At the middleware level, there are three contributions: (1) a new communication framework for extending the network's lifetime whose aim is not only minimizing the energy consumption, but also balancing it; (2) a heuristic algorithm which follows the proposed tenet; and (3) a task mapping simulator to decrease the time for successful development and testing of task mapping and allocation algorithms on the systems under development and deployment.