A novel message passing algorithm for online Bayesian filtering: Turbo filtering

In this manuscript a novel online technique for Bayesian filtering, dubbed turbo filtering, is illustrated. In particular, it is shown that this filtering method, which can be interpreted as an extension of marginalized particle filtering, results from the application of the sum-product rule to a factor graph representing a mixed linear/nonlinear state-space model. Simulation results for a specific state-space model evidence that turbo filtering can outperform marginalized particle filtering in terms of both accuracy and complexity.