An Information-Theoretic Approach to Data-Validation

This paper describes a method of data validation for decentralized data fusion systems. For decentralized systems we consider networks of local processors without any central processor or common communication facility: Communication only occurs between neighbouring nodes and all computation is performed locally. Each local processor or node takes local observations and, using the information form of the Kalman filter, calculates local information contributions. These are then communicated and a local estimate of a common state is computed. The system state estimates can be altered significantly by erroneous sensor data. To avoid this undesirable effect, an algorithm which decides whether to assimilate or reject data is required. This algorithm will compute a validation gate or region using the state estimates and communicated terms. This gate wili be compared with a given threshold (which sets the maximum gate size), the observations inside this region will be considered valid and those outside will be discarded. The gate is equivalent to the one obtained in the conventional Kalman Filter, so a similar criteria for the threshold selection can be followed. It is also shown that in the case of a distributed model, validation of information generated at a node can be performed at any other node, regardless of the local observation model.