Event-Triggered Consensus Robust Filter with Noise Outliers for Distributed Sensor Networks

This article proposes a distributed robust nonlinear consensus filter for distributed state estimation (DSE) in distributed sensor networks with heavy-tailed noise containing unknown outliers. Specifically, both the measurement likelihood probability density function (PDF) and the prior PDF in the proposed filter are modeled as hierarchical Gaussian (HG) models. By adaptively adjusting the covariance through squared scaling parameters, the approach effectively handles nonstationary noise. The DSE is implemented within a Bayesian framework, where the data fusion strategy, based on information filtering, interprets the consensus theory of local PDFs under the HG distribution from the perspective of the Kullback-Leibler weighted averaging. Furthermore, the single-step consensus of local likelihood and prior PDFs is simplified into the fusion of information vectors and information covariance matrices. The parameters of the HG model are iteratively updated by the variational Bayesian method to improve robustness against heavy-tailed noise. To reduce the computational burden in distributed networks, an event-triggered communication mechanism is incorporated into the design of the proposed consensus filter. Finally, simulations conducted in a bearing-only target tracking DSNs demonstrate the effectiveness and superiority of the proposed robust consensus filter.