Distributed averaging algorithms resilient to communication noise and dropouts

In this paper, we consider the problem of distributed average computation over communication networks whose channels are non-ideal, but noisy and/or intermittent. Channel intermittency captures randomness of network interconnections and packet-drop links. Based on input-output properties of feedback systems, we propose novel iterative algorithms that incorporate a networked feedback compensator to mitigate effects of the unreliable communication on distributed averaging. The new algorithms are time-invariant and do not suffer from the random walk behavior to additive noise of other average consensus algorithms. Moreover, the use of the link state information at the receiver leads to a new algorithm, which computes averages approximately correctly in the presence of intermittent communication and additive noise, under certain conditions.