Distributed fusion estimation for multisensor systems with non-Gaussian but heavy-tailed noises.

Area of Science:

• Signal Processing
• Statistical Inference
• Control Systems

Background:

• Heavy-tailed noise, often modeled by the Student's t-distribution, is prevalent in practical systems.
• Existing t-distribution-based filters lack an information filter form, and data fusion for dynamic systems with such noise is underexplored.

Purpose of the Study:

• To develop information filter and data fusion algorithms for dynamic systems subject to heavy-tailed noise.
• To generalize classical Kalman filter algorithms for improved performance in non-Gaussian noise environments.

Main Methods:

• Derivation of t-distribution-based information filter using multivariate t-distribution and variational Bayesian estimation.
• Development of centralized batch fusion, distributed fusion, and suboptimal distributed fusion algorithms.
• Parameter approximation for computational complexity reduction in suboptimal distributed fusion.

Main Results:

• The proposed algorithms generalize classical Kalman filter-based methods.
• The derived distributed fusion algorithm is equivalent to centralized batch fusion.
• Theoretical and experimental analyses demonstrate the feasibility and effectiveness of the algorithms, particularly in target tracking.

Conclusions:

• The novel t-distribution-based information filter and fusion algorithms effectively handle heavy-tailed noise in dynamic systems.
• The developed methods offer a robust alternative to traditional Kalman filtering approaches, enhancing system performance and efficiency.