Adaptive Gaussian Mixture Learning in Distributed Particle Filtering

We consider the problem of adaptive Gaussian mixture learning in posterior-based distributed particle filtering, in which posteriors are approximated as Gaussian mixtures for wireless communication. We develop a hierarchical clustering algorithm to learn from weighted samples a Gaussian mixture with an adaptively determined number of components. Different from existing work, the proposed algorithm embeds a kernel density estimation-based clustering algorithm in each recursive step of hierarchical clustering to adaptively split a cluster. We use the hierarchical clustering result as an initial guess for the expectation-maximization algorithm to obtain a local maximum likelihood solution. Numerical examples show that the proposed method leads to higher accuracy in distributed particle filtering and is more efficient in both computation and communication than other methods.