Multirate multisensor data fusion for linear systems using Kalman filters and a neural network

In this paper the data fusion problem for asynchronous, multirate, multisensor linear systems is studied. The linear system is observed by multiple sensor systems, each having a different sampling rate. Under the assumption that the state space model is known at the scale of the highest time resolution sensor system, and that there is a known mathematical relationship between the sampling rates, a comprehensive state space model that includes all sensor systems is presented. The state vector is estimated with a neural network that fuses the outputs of multiple Kalman filters, one filter for each sensor system. The state estimate is shown to perform better than other data fusion approaches due to the new neural network based sensor fusion approach.