Kullback-Leibler Divergence Based Distributed Cubature Kalman Filter and Its Application in Cooperative Space Object...

Area of Science:

• Control Systems Engineering
• Signal Processing
• Information Theory

Background:

• Distributed Bayesian filtering is complex in sensor networks due to limited sensor access to global likelihood functions and potential weak state observability.
• Existing methods struggle with nonlinear dynamics and measurement mappings in decentralized systems.

Purpose of the Study:

• To design a distributed Bayesian filter for nonlinear dynamics and measurement mapping using Kullback-Leibler divergence.
• To address challenges of limited sensor information and weak observability in sensor networks.
• To develop a robust algorithm for cooperative tracking problems.

Main Methods:

• The distributed Bayesian filter problem was reformulated as a maximization of posterior probability optimization problem.
• A global cost function was decomposed into local cost functions solvable by individual sensors.
• Kullback-Leibler divergence was utilized for neighbor communication and global estimate approximation.
• A distributed cubature Kalman filter (DCKF) was proposed based on the developed structure.

Main Results:

• The proposed DCKF algorithm effectively handles varying communication topologies within the sensor network.
• The algorithm successfully mitigates issues arising from weak observability in certain sensors.
• Simulations confirmed the algorithm's efficacy in a cooperative space object tracking scenario.

Conclusions:

• The developed distributed Bayesian filter design offers a robust solution for nonlinear filtering in sensor networks.
• The proposed DCKF algorithm enhances cooperative sensing capabilities, particularly in challenging environments.
• This approach provides a foundation for advanced distributed estimation and tracking in complex systems.