Research Project Title
Design and Analysis of Sensor Networks for Statistical Inference Applications

Principal Investigator
Venugopal Veeravalli

Unit # 28
Project Overview

The problem of detecting an abrupt change in a system based on stochastic observations of the system was first studied by Page fifty years ago in the context of quality control. The standard formulation of this change-point detection problem is a “single sensor/channel” formulation where there is one sequence of observations whose distribution changes at some unknown point in time. The goal is to detect this change as soon as possible, subject to false alarm constraints. Under the assumption that the observations are independent and identically distributed (i.i.d.) with known distributions before and after the change, this problem is well understood and has been solved under a variety of criteria since the seminal work by Page.

Our goal in this research scenario is to develop new procedures for change detection in distributed multisensor systems, and to provide an analytical framework to predict their performance in terms of the tradeoff between detection delay and frequency of false alarms. To address this goal, we propose to analyze several generalizations of the change detection problem that arise in context of sensor networks. We will investigate a variety of models for the change process: only one (or a subset) of the sensors changes, they all change at the same time, or they change at different times. We will also include various scenarios for communication with the fusion center, from the centralized one where the sensors send sufficient statistics, to the decentralized one where they send quantized observations or local decisions. We will study the role of feedback from the fusion center and intersensor communication. We will investigate schemes for conserving energy at the sensors such as switching the sensors between active/sleep modes and censoring their observations. Our strategy for design and analysis will accommodate general statistical models for the observations, and allow for different degrees of model uncertainty. Asymptotic analyses in the limits of large number of sensors as well as small false alarm rates will be key to designing sensor systems for this research scenario.