Federated Tobit Kalman Filtering Fusion With Dead-Zone-Like Censoring and Dynamical Bias Under the Round-Robin Protocol

This paper is concerned with the multi-sensor filtering fusion problem subject to stochastic uncertainties under the Round-Robin protocol (RRP). The uncertainties originate from three sources, namely, censored observations, dynamical biases and additive white noises. To reflect the dead-zone-like censoring phenomenon, the measurement observation is described by the Tobit model where the censored region is constrained by prescribed left- and right-censoring thresholds. The bias is modeled as a dynamical stochastic process driven by a white noise in order to reflect the random behavior of possible ambient disturbances. The RRP is employed to decide the transmission sequence of sensors so as to alleviate undesirable data collisions. The filtering fusion is conducted via two stages: 1) the sensor observations arriving at its corresponding estimator are first leveraged to generate a local estimate, and 2) the local estimates are then gathered together at the fusion center in order to form the fused estimate. The local estimator implements a Tobit Kalman filtering algorithm on the basis of an enhanced Tobit regression model, whilst the fusion center realizes a filtering fusion algorithm in accordance with the well-known federated fusion principle. The validity of the fusion approach is finally shown via a simulation example.

Automated summary

This paper investigates the multi-sensor filtering fusion problem under stochastic uncertainties with the use of the Round-Robin protocol (RRP). Various sources of uncertainties, such as censored observations, dynamical biases, and additive white noises, are considered and modeled in the study. The fusion process involves a two-stage approach where sensor observations are first used to generate local estimates before being fused at a central fusion center.