Distributed Kalman Filtering Using Weighted Averaging

This paper addresses the problem of distributed Kalman filtering, with
focus on limiting the required communication bandwidth.
By distributed we refer to a scenario when all nodes in the network desire an
estimate of the full state of the observed system and there is no
centralized computation center. Communication only takes place
between neighbors and only a fixed number of times each sample. To
reduce bandwidth requirements of individual nodes, estimates
instead of measurements are communicated. A new estimate is
then formed as a weighted average of the neighbouring estimates. The
weights are optimized to yield a small estimation error covariance in
stationarity. The minimization can be done off line thus allowing
only estimates to be communicated. The advantage of communicating
estimates instead of measurements becomes more evident when the number
of nodes exceeds the size of the state vector to be estimated. The
algorithm is applied to one
simple second order system and temperature sensing network.