A method for estimating a system state, in a decentralized network having a plurality of nodes, each node being confined for receiving and sending information, and for processing information. Each node being connected to neighboring nodes of the network. At each node the method includes: (i) maintaining a set of particles and associated weights, which represent an estimate of the system state, (ii) representing the estimated system state as a mixture of Gaussian distributions in a channel filter, and communicating the mixture to neighboring nodes, and (iii) a neighboring node receiving the mixture in a channel filter that contains a similar Gaussian representation of its own estimate of system state, and dividing the incoming mixture by the existing mixture, for updating the estimate of the system state that is maintained at the node.