Efficient resource allocation schemes for remote state estimation systems are investigated in this thesis. In remote state estimation problems, sensors are distributively deployed to measure the states of several interested physical processes. The measurements taken by the sensors are preprocessed and scheduled to be sent to a remote estimator via a wireless network, and the statistical inference is then performed by the remote estimator based on received data packets. Three closely related problems are studied, and they share the following common features:

(1) The data transmission processes are subject to the uncertainty due to the wireless communication.

(2) The network resources, like the bandwidth and the energy, are limited.

(3) The computation of obtaining the optimal...[ Read more ]

Efficient resource allocation schemes for remote state estimation systems are investigated in this thesis. In remote state estimation problems, sensors are distributively deployed to measure the states of several interested physical processes. The measurements taken by the sensors are preprocessed and scheduled to be sent to a remote estimator via a wireless network, and the statistical inference is then performed by the remote estimator based on received data packets. Three closely related problems are studied, and they share the following common features:

(1) The data transmission processes are subject to the uncertainty due to the wireless communication.

(2) The network resources, like the bandwidth and the energy, are limited.

(3) The computation of obtaining the optimal data transmission policies is formidable.

We study the problems where the network resources are shared among several independent subsystems, and mainly focus on obtaining efficient solutions. Two different aspects are considered: a decentralized setting or a centralized setting, an offline optimization or an online stochastic learning. In the first problem, a decentralized setting where the transmission decisions of any subsystem solely relies on its own state and the network state, and the data transmission should be scheduled to avoid the potential congestion. In the second problem, we consider the scenario where the remote estimator serves as a central scheduler, which collects the whole information from all subsystems and makes the transmission decisions for all subsystems. In the third problem, we consider the online learning scenario where the network statistics are estimated from the online information. The common objective of the above three considered problems is to obtain an efficient transmission allocation policy which achieves the desired trade-off between the estimation performance and the resource consumption. For the decentralized setting, the sufficient and necessary condition for the stability of the remote estimator is derived, and the optimal scheduling policy can be effectively computed. For the centralized setting with known channel statistics, a heuristic is developed to achieve the near optimality. For transmission scheduling with unknown network statistics, the stochastic learning schemes are developed to adaptively schedule the data transmission. Several possible future directions are also discussed.