Centralized and distributed Multi-Sensor Filtering Fusion (MSFF) methodologies differ in their approach to data fusion and handling of censored data and network constraints.
Centralized MSFF involves a fusion center that collects all sensor measurements, processes them, and produces an optimal state estimate. This method offers optimal performance but incurs high communication and computational costs, and is vulnerable to communication failures.
Distributed MSFF, on the other hand, involves local estimators that generate state estimates and send them to a fusion center. This method reduces communication and computational overheads but may result in suboptimal fusion performance.
Both methods address censored data by using Tobit Kalman filters (TKF), which can handle non-Gaussian measurement noises. For network constraints, centralized MSFF can be designed to accommodate fading measurements, packet dropouts, and redundant channels. Distributed MSFF can utilize probabilistic methods to analyze stability and performance under packet delays and transmission delays.