Date of Award
Engineering and Applied Science - Electrical
This dissertation focuses on the systemic design of proper estimation as well as fusion techniques and data-driven communication schemes to infer the state of a dynamic discrete-time linear system over a wireless network. The goal is to effectively extract and share key information over networks to enhance estimates of system states. Such co-design research is drawn by the need to leverage limited resources (communication bandwidth, energy and computational power) for a multitude of data and data sources in networked systems.
We first build the overarching structure by synthesizing communication and estimator/fuser, and research the collective behaviors of the system.
In the design phase, we jointly construct the data-driven communication scheme and the estimator/fuser based on a single sensor, centralized and distributed estimation fusion architecture, respectively.
Based on this structure, we design data-driven communication schemes both based on measurement innovation and estimation innovation. Moreover, we propose a new metric based on cumulative estimate innovation to smartly pick key information. We derive the corresponding (approximate) minimum mean square error (MMSE) estimators/fusers and the MMSE-optimal weighted least square (WLS) fusers in the closed-form representations. Further, our fusers have guaranteed stability.
These data-driven communication schemes take into account multiple network structures and the interaction between modules of communication and estimation. They utilize the importance of innovation for estimation and can achieve a trade-off between communication costs and estimation performance. The proposed estimators and fusers optimally use information in the triggering decisions to boost estimation performance.
Further, with the aim of optimally calibrating the tradeoff between estimation quality and communication expenses, we treat limited communication resources quantitatively in the joint design and formulate the problem in the framework of optimal estimation by incorporating these limited resources. Through constructing an auxiliary state vector, the optimization problem on the expected total discounted cost---including estimation error and weighted communication cost---over the infinite horizon is shown to be representable as a Markov Decision Process (MDP) problem. An iterative algorithm is proposed to find the optimal cost and optimal policy. The optimal policy has a peculiar degree-of-freedom reduction property.
Besides all these theoretical contributions, our work carries weight of practical applications. The proposed data-driven communication schemes under our metric have easy setup and can seamlessly integrate different dynamic systems, because the optimal decision policy is algorithmically attainable. The stability study of estimators and fusers ensures their applicability in practical control systems.
The designed data-driven communication schemes and estimators/fusers can provide optimal solutions for state estimation over sensor networks; also they are reliable and easy to implement across multiple sensor network architectures.
Bian, Xiaolei, "State Estimation and Data Fusion with Data-driven Communication" (2023). University of New Orleans Theses and Dissertations. 3060.