Decentralised data fusion in 2-tree sensor networks

This paper describes an algorithm for decentralised estimation and data fusion in `2-tree´ networks. In earlier approaches, exact decentralised estimation which avoids double-counting information has generally only been possible in singly-connected or 1-tree networks, or by using conservative fusion. The problem with tree networks is that they are very fragile; a single node failure causes separation of the network. In contrast, a 2-tree network is a denser network in which two nodes must fail for the network to become separated, but which are still much sparser, and therefore more scalable, than fully-connected networks. This paper describes an algorithm for decentralised estimation in 2-tree networks that correctly accounts for common information in communication, allowing consistent and scalable operation. The method presented is scalable, i.e.: storage and communication sizes do not grow with the size of the network. Results show the correct operation of the algorithm on complete 2-tree, 1-tree and mixed 1&2-tree networks, and show robustness of the 2-tree network against node and link failures.