Data-driven state estimation under limited communication resources

Remote state estimation in networked control systems always consumes too much sensor battery power and communication bandwidth. Under power and communication constraint, we seek a desirable tradeoff" between communication rate and estimation performance in terms of estimation error covariance. We propose two data-driven sensor scheduling strategies to achieve that goal. We prove that under our strategies the minimum mean squared error (MMSE) estimator is a Kalman-like filter which maintains linearity. We give the explicit MMSE estimator under each strategy. In the end we conduct numerical experiment to show the superiority of our design.