Distributed multiple-model fusion with transformed measurements

This paper deals with estimation fusion for a Markovian jump-linear system (MJLS) and proposes a distributed fusion scheme, in which local sensors send their transformed measurements to the fusion center and the fusion center fuses them with a multiple-model (MM) filter. A specific linear transformation for local measurements is studied and it is shown that the distributed minimum mean-squared error (MMSE) fusion with these transformed measurements has the same performance as the centralized MMSE fusion when full-rate communication is employed. The reduced-rate communication case is also considered for the systems with very limited communication capacity. Moreover, approximate algorithms are presented for practical application. Illustrative numerical results are provided to show the performance of the fusion methods.