Sensor Scheduling In Multi-Process Systems And Applications

Currently sensor scheduling receives more and more attention. In certain appli-cations of sensor networks, inevitably there exist communication constraints and interference between different sensors. In these cases, only one or few of sensors in the networks can be active at each step. This kind of constraints may lead a system into bad performance. Therefore, how to schedule the sensors is the key to improve the system's performance. In fact, an optimal sensor scheduling algorithm can not only improve the system's performance, but also reduce energy consumption. Therefore, the study of sensor scheduling is meaningful from the perspective of real applications.This thesis focuses on sensor scheduling algorithms in multi-process systems. In a multi-process system, there are several independent processes. For each process, we have a Kalman filter or extended Kalman filter to observe it. It is also assumed that, at each step only one or few of the processes can be observed due to practical constraints. For such kind of systems, we need to find sensor scheduling schemes to improve the estimation quality.Firstly, we discuss the cases when the systems are linear. In our setup, each process is observed by a Kalman filter and at each time step only one Kalman filter could obtain observation due to practical constraints. To solve the problem, two novel notions, permissible consecutive observation loss (PCOL) and least consecu-tive observation (LCO), are introduced as criteria to describe feasible observation sequences for a process ensuring desired estimation qualities. Next, two methods, namely, threshold method and periodic method, are proposed to calculate PCOL and LCO for each process. Based on the derived PCOL and LCO requirements, we develop two algorithms that are applicable to different situations:Sxy algorithm from the pinwheel problem for the case of LCO= 1 and tree search algorithm for general cases. Also, to reduce the computational complexity of tree search algorithm, several useful pruning conditions are obtained. Both rigorous analysis and simulation results are provided to validate the approaches.After discussing the linear cases, we also investigate the non-linear cases, where we use extended Kalman filter to observe the processes. To make the algorithms more realizable, we study a special realization of multi-process systems:an active-mode ultrasonic indoor multi-target tracking system. It is assumed that several moving targets are moving on the plane and may interfere with each other if they chirp simultaneously. A myopic scheduling scheme together with a grouping ap-proach are proposed to improve tracking accuracy and avoid interference. Two metric functions are used in myopic scheduling, namely, Minimal Estimation Error (MEE) and Maximal Average Error Deduction (MAED). Simulation results show that MEE may cause over-partiality which leads to bad tracking accuracy for some targets, and MAED can effectively eliminate such over-partiality. Myopic schedul-ing schemes with both MEE and MAED are realized on the Cricket platform with active architecture. Experimental results show that the myopic schemes with both MEE and MAED have better tracking performance than a periodic scheme.