Optimal Linear Estimation Fusion— Part VII: Dynamic Systems

In this paper, we first present a general data model for discretized asynchronous multisensor systems and show that errors in the data model are correlated across sensors and with the state. This coupling renders most existing “optimal” linear fusion rules suboptimal. While our fusion rules of Part I are valid and optimal for this general model, we propose a general, exact technique to decouple the two types of correlation of the errors so that other existing rules can be applied after decoupling. Then, we discuss several theoretically important issues unique to fusion for dynamic systems. The first is the role of prior information in the static case versus that of prediction in the dynamic case. We present two general, best linear unbiased estimation fusers with and without prior information respectively. Other issues discussed include optimality of existing linear fusion rules as well as two commonly used fusion schemes, and the effect of feedback.