Full order distributed particle filters for intermittent connections: Feedback from fusion filters to local filters improves performance

In [1, 2], we proposed a consensus/fusion based distributed implementation of the particle filter (CF/DPF) for non-linear systems with non-Gaussian excitation and intermittent communication connectivity. To recap, the CF/DPF implemented two filters at each node: (i) A localized particle filter based only on the host node´s observations, and; (ii) A separate consensus-based filter (fusion filter) to fuse together the local filters´ densities for estimating the global posterior in a distributed fashion. At each sensor node, the fusion filter provides the overall state estimates. The paper extends the CF/DPF framework by incorporating feedback from the fusion filter back to the local particle filter - a proposed enhancement to the original CF/DPF. No additional communication overhead is needed for the feedback modified CF/DPF (FCF/DPF), which exhibits improved overall performance over the CF/DPF in highly noisy Monte-Carlo simulations.