A Distributed Diffusion Kalman Filter In Multitask Networks

Abstract

The Distributed Diffusion Kalman Filter (DDKF) algorithm in all its magnitude has earned great attention lately and has shown an elaborate way to address the issue of distributed optimization over networks. Estimation and tracking of a single state vector collectively by nodes have been the point of focus. In reality, however, there are several multi-task-oriented issues where the optimal state vector for each node may not be the same. Its objective is to know many related tasks simultaneously, rather than the typical single-task problems. This work considers sensor networks for distributed multi-task tracking in which individual nodes communicate with its immediate nodes. A diffusion-based distributed multi-task tracking algorithm is developed. This is done by implementing an unsupervised adaptive clustering process, which aids nodes in forming clusters and collaborating on tasks. For distributed target tracking, an adaptive clustering approach, which gives agents the ability to identify and select through adaptive adjustments of combination weights nodes who to collaborate with and who not to in order to estimate the common state vector. This gave rise to an effective level of cooperation for improving state vector estimation accuracy, especially in cases where a cluster's background experience is unknown. To demonstrate the efficiency of our algorithm, computer simulations were conducted. Comparison has been carried out for the Diffusion Kalman Filter multitask with respect to the Adapt then combine (ATC) diffusion schemes utilizing both static and adaptive combination weights. Results showed that the ATC diffusion schemes algorithm has great performance with the adaptive combiners as compared to static combiners.