Distributed consensus and linear functional calculation in networks

Abstract

We study the problem of performing sensor fusion and distributed consensus in networks, where the objective is to calculate some linear function of the initial sensor values at some or all of the sensors. We utilize a linear iteration where, at each time-step, each sensor updates its value to be a weighted average of its own previous value and those of its neighbors. We show that this approach can be viewed as a linear dynamical system, with dynamics that are given by the weight matrix for the linear iteration, and with outputs for each sensor that are captured by the subset of the state vector that is measurable by that sensor. We then cast the fusion and consensus problems as that of observing a linear functional of the initial state vector using only local measurements (that are available at each sensor). When the topology of the network is time-invariant, we show that the weight matrix can be chosen so that each sensor can calculate the desired function as a linear combination of its measurements over a finite number of time-steps.