A Fast Distributed Variational Bayesian Filtering for Multisensor LTV System With Non-Gaussian Noise.

Area of Science:

• Control Systems Engineering
• Signal Processing
• Statistical Inference

Background:

• Designing distributed robust estimators for multisensor linear time-varying systems with non-Gaussian noise is challenging.
• Key issues include achieving accuracy, robustness to outliers, and minimizing computation and communication costs.

Purpose of the Study:

• To propose a fast distributed variational Bayesian (VB) filtering algorithm for state and noise distribution estimation.
• To evaluate the algorithm's performance across incremental, diffusion, and consensus-based sensor networks.

Main Methods:

• Modeling non-Gaussian measurement noise using Student's t-distribution.
• Employing a variational Bayesian (VB) approach for iterative state and parameter estimation.
• Implementing an interaction scheme for fusing local parameters to achieve global optimality.
• Developing a sensor selection criterion based on Cramér-Rao lower bound to reduce computational burden.

Main Results:

• The proposed distributed VB filtering algorithm effectively estimates system states and noise distributions.
• The algorithm demonstrates increased robustness against node or link failures compared to centralized methods.
• Achieved lower computation costs with acceptable estimation performance and communication load.

Conclusions:

• The developed fast distributed VB filtering algorithm offers a robust and efficient solution for multisensor systems with non-Gaussian noise.
• The method provides a favorable trade-off between accuracy, robustness, and resource utilization in distributed sensing networks.