This thesis focuses on remote state estimation problem where an estimator uses data sent by a sensor to estimate the state of a linear discrete-time system. We use an event-based approach to improve the remote estimation quality subject to limited sensor-estimator communication rate. This restriction is due to either limited communication bandwidth or a limited energy budget of the sensor.

Two types of sensors are considered in this thesis work. For the sensor with sufficient computational capability to run a Kalman filter locally, we first propose two online sensor-to-estimator communication schedulers, under which a significant communication rate reduction is achieved while only a tolerable estimation performance is sacrificed. However, both schedulers reduce the communicatio...[ Read more ]

This thesis focuses on remote state estimation problem where an estimator uses data sent by a sensor to estimate the state of a linear discrete-time system. We use an event-based approach to improve the remote estimation quality subject to limited sensor-estimator communication rate. This restriction is due to either limited communication bandwidth or a limited energy budget of the sensor.

Two types of sensors are considered in this thesis work. For the sensor with sufficient computational capability to run a Kalman filter locally, we first propose two online sensor-to-estimator communication schedulers, under which a significant communication rate reduction is achieved while only a tolerable estimation performance is sacrificed. However, both schedulers reduce the communication rate by at most a half. To tackle the situation where communication resource can be severe, we propose a stochastic online sensor scheduler subject to any given communication rate, and provide a closed-form expression on the minimum mean-squared error (MMSE) estimate. To find the best scheduler, we formulate an optimization problem and relax it, from a non-convex problem, to a generalized geometric programming, and show that it can be solved with an acceptable computational complexity. For all the aforementioned schedulers, we prove they outperform the optimal offline scheduler.

For the sensor without computational capability to process its measurement data, we propose a different event-based scheduler. A corresponding MMSE estimator is derived. By adopting an approximation technique from nonlinear filtering, a simple form of an accurate MMSE estimator is provided, from which a clear relationship between the sensor-to-estimator communication rate, the remote estimation quality, and the event-triggering threshold is obtained.