Energy consumption trade-offs in power constrained networks

The main goal of this work is to understand the energy consumption trade-offs between computation and communication in power constrained networks in general, and in visual sensor networks, in particular. In order to do so, we need to evaluate and model: energy spent to process data by sensors (e.g., how much energy vision algorithms require, etc.), as well as energy required for communications. The first step is to develop an accurate model for energy consumption due to communications. Such energy model must be as close to reality as possible, taking into account all radio states, i.e., energy spent not only while transmitting and receiving a packet, but also while in idle, overhearing, or sleep modes. We have developed such a model and used it to instrument the QualNet and GloMoSim network simulators. We have used QualNet/GloMoSim instrumented with our model to compare energy consumption of power-aware MAC protocols, multi-hop ad-hoc wireless networks routing protocols, and to evaluate an analytical model of energy consumption in single-hop IEEE 802.11 ad-hoc networks. Next, we look at a model that also includes processing and sensing tasks, besides communication. Since we were targeting power constrained networks, we had to look into a typical platform for wireless mobile applications: a laptop. This step provided a good opportunity to define our methodology, which determines the energy consumption of basic tasks. From this point on, we focused on Visual Sensor Networks. We designed and implemented a wireless camera network testbed, Meerkats, which is based on the Stargate platform. We followed the same methodology developed for the laptops' testbed to characterize the energy consumption of the Meerkats' node. Along this work, we also extended and validate the on-board battery monitoring capability on the Stargate. Following this step, we tailored the tasks to the ones representative of activities carried out by wireless camera networks targeting surveillance applications. Given this set of tasks (which we call elementary tasks), a quantitative power consumption and temporal analysis of set of duty cycles that apply to the Meerkats testbed was done. Then we used what we learned from the duty cycle analysis along with a set of experiments on the Meerkats testbed to understand how a framework for lifetime prediction can be build. We proposed a simple deterministic lifetime prediction model based on task composition, which we validate using the experiments executed on the Meerkats testbed. Finally, we consider a set of possible duty cycles given requirements for the Meerkats testbed and analyze how they affect the node lifetime. (Abstract shortened by UMI.)