The ability to accurately estimate the state of a system and control its behavior is crucial for a wide range of applications, from managing power grids to automating industrial processes. With the rise of internet of things (IoT), state estimation and control systems can now utilize sensors and control agents distributed across wide areas, connected wirelessly for ease of maintenance and enhanced coverage. The rise of the IoT-based state estimation and control systems, however, presents new challenges. IoT systems, with their geographically distributed sensors and control agents connected through often unreliable wireless networks, demand new approaches to state estimation and control. Traditional methods, while effective in other contexts, often result in poor stability performance wh...[ Read more ]

The ability to accurately estimate the state of a system and control its behavior is crucial for a wide range of applications, from managing power grids to automating industrial processes. With the rise of internet of things (IoT), state estimation and control systems can now utilize sensors and control agents distributed across wide areas, connected wirelessly for ease of maintenance and enhanced coverage. The rise of the IoT-based state estimation and control systems, however, presents new challenges. IoT systems, with their geographically distributed sensors and control agents connected through often unreliable wireless networks, demand new approaches to state estimation and control. Traditional methods, while effective in other contexts, often result in poor stability performance when applied directly to the dynamic and unpredictable nature of IoT environments. Part I of this thesis lays the groundwork for understanding state estimation and control in the context of IoT. We begin by outlining the key characteristics of state estimation and control for IoT systems, then delve into traditional state estimation and control methods, highlighting their limitations when applied to the unique challenges posed by IoT scenarios.

Part II focuses on state estimation for IoT systems. We begin by addressing the crucial issue of sensor measurement quantization. To achieve zero-access latency, high spectral efficiency, and desirable state estimation performance at the remote estimator, we propose a novel radix-partition-based over-the-air aggregation multiple access scheme. To further enhance computional complexity performance in unreliable network conditions, we introduce a novel fixed-gain-based filtering scheme, where the state estimation solution can be optimized offline based on closed-form stability criteria. Next, we tackle the challenge of sensor measurement asynchronization. Our solution enhances system observability by leveraging a novel oversampling-based datapath design at both the sensor and estimator sides. This design also enables sparse structure of the observation matrix and allows for a computationally efficient 2-Dimensional (2D) graph-based state estimation algorithm that achieves maximum-a-posteriori (MAP) estimation via message passing on loop-free graphs. Finally, we consider remote state estimation system design with multiple input channels of sensor measurements and limited resource elements (REs) in the wireless interface. Our proposed over-the-air aggregation-based random access scheme allows each sensor input channel to randomly access the finite RE pool. This scheme induces a sparse observation matrix, enabling a low-complexity 2-D graph-based state estimation approach that achieves exact MAP solutions at the remote estimator.

In Part III, we shift our focus to remote control for IoT systems. First, we explore optimal controller design for systems with a single control agent. We formulate the optimal control problem for IoT systems over wireless channels and analyze the existence of an optimal solution using contractive analysis. Utilizing the Bellman technique, we derive the structural form of the optimal control solution. A structured stochastic approximation (SA) algorithm is then developed to learn the optimal control solution online by understanding the structured kernel of the reduced-state value function. As optimal control requires knowledge of plant dynamics, we extend our approach to include simultaneous system identification and control. This is achieved through a novel online normalized-stochastic-gradient-descent (NGD)-based algorithm. Second, we extend to consider optimal control in IoT systems with multiple control agents. The problem is modeled as a stochastic non-zero-sum game. We obtain the Nash equilibrium (NE) by learning the kernels for a tuple of structured reduced-state value functions through an online SA-based algorithm. Additionally, we address optimal controller design without prior knowledge of plant dynamics. We achieve optimal control asymptotically using the NSGD-based online algorithm. Third, recognizing fully coupling in the optimal controller design for multi-agent IoT systems, we explore decentralized scheduling and control for multi-agent IoT systems. We frame this challenge as a drift-plus-penalty problem and derive a closed-form decentralized scheduling and control solution. This allows each control agent to independently determine when and how to stabilize the dynamic systems, based solely on its local channel state and plant state.

Part IV explores the application of control techniques in IoT-based furnace temperature control systems. In the systems, the temperature controller is wirelessly connected to the furnace for ease of maintenance and installation. We model the furnace dynamics by considering linear heat conduction and convection, as well as nonlinear heat radiation. Based on this thermal model, we formulate furnace temperature tracking control problems and derive the structure of the optimal control solutions using the Bellman technique and the homotopy perturbation method. To exploit the structured properties of the optimal control solutions, we develop two efficient temperature control algorithms: a structured stochastic approximation-based algorithm for scenarios involving linear heat conduction and convection, and a structured reinforcement learning (RL) algorithm for more complex dynamics that include non-linear heat radiation. Simulation results demonstrate that our proposed temperature control scheme achieves superior tracking performance compared to existing state-of-the-art solutions.