An emerging class of sensor network applications involves the acquisition of high-resolution signals using low-power wireless sensor nodes. Examples include monitoring acoustic, seismic, and vibration waveforms in bridges, industrial equipment, volcanoes, and animal habitats. These systems share the goal of acquiring high data-rate (100 Hz or higher), high-fidelity data across the network, subject to severe constraints on radio bandwidth and energy usage.
Given these constraints, it is typically not possible to acquire continuous waveforms from all nodes. As a result, applications strive to acquire the most “interesting” signals, such as a marmot call or earthquake, and avoid wasting resources on “uninteresting” signals. Currently, these resource-management decisions are made on an ad hoc basis for each application, often resulting in suboptimal solutions that can consume excessive bandwidth or lose data. We argue that all of these applications would benefit from a general approach to managing resources that optimizes the application-specific value of the signals acquired by the network. The goal of this thesis is to develop such an approach.
Optimizing reliable data acquisition requires a coordinated approach to managing both limited energy capacity and severely constrained radio bandwidth. Depending on the sampling rate and resolution, downloading signals may take longer than real time; while sensor node radios obtain single-hop throughput of about 100 Kbps, the the best reliable protocols achieve less than 8 Kbps for a single transfer over multiple hops. Likewise, to achieve long lifetimes in the field, the energy cost of downloading a signal from the network must be carefully considered. The fundamental challenge is how to best direct these limited network resources to deliver the most valuable data to the application.
This proposal presents Lance, a general approach to bandwidth and energy management for reliable signal collection in wireless sensor networks. In Lance, each node acquires data at potentially high rates. For each signal region, the node generates a concise summary, which are periodically sent to the base station where they are combined with other application-specific knowledge and used to assign a value to each signal region across the network. Because energy usage and battery lifetime is a major concern for long-term sensor network deployments, Lance incorporates a cost estimator that predicts the energy cost for reliably downloading each signal from the network. Using these cost estimates in conjunction with measurements of available resources allows Lance to accurately target specific system lifetimes by ensuring that data extraction adheres to an energy schedule. Finally, once a cost and value have been assigned to each signal, Lance aims to maximize the value returned to the application through an optimization process that considers the cost to download each signal region against the resources available in the network at that time. The processes of value assignment, cost estimation and the resulting optimization problem are areas in which this thesis will make research contributions.
Lance incorporates a general framework for managing bandwidth and energy that decouples the mechanism of prioritizing resource allocation from the application-specific policies used to assign priorities. This is accomplished through user-supplied policy modules that permit a range of sophisticated prioritization policies to be tailored for specific applications. Policy modules allow the network’s behavior to be significantly altered at the base station, without reprogramming the sensor nodes themselves.
The structure of this proposal is as follows. The first part, consisting of this introduction, Section 2 motivating this work, and Section 3 laying out the fundamental research questions, aims to introduce our architecture, justify the need for research in this area, and provide a high-level overview of our approach. The second part delves into much more detail, including sections on cost, value and constraints (4); a section outlining the optimization problem (5); and a section providing an overview of my proposed architecture (6). Finally, the last part lays out a clear plan for moving forward, including an assessment of metrics and evaluation strategy in Section 7, an overview of work to date in Section 8, related work in Section 9, a timeline for future progress in Section 10 and conclusion in Section 11.