Sensor scheduling is a well studied problem in signal processing and control with numerous applications.Despite its successful history, most of the related literature assumes the knowledge of the underlying probabilistic model of the sensor measurements such as the correlation structure or the entire joint probability density function.Herein, a framework for sensor scheduling for remote estimation is introduced in which the system design and the scheduling decisions are based solely on observed data. Unicast and broadcast networks and corresponding receivers are considered. In both cases, the empirical risk minimization can be posed as a difference-of-convex optimization problem and locally optimal solutions are obtained efficiently by applying the convex-concave procedure. Our results are independent of the data's probability density function, correlation structure and the number of sensors.
I. INTRODUCTIONSensor scheduling is a classical problem in signal processing and control with a very rich history. The traditional static sensor scheduling problem consists of selecting a subset of k sensors among a group of n sensors such that the expected distortion between the random state-of-the-world and its estimate is minimized [1]. The fact that we are selecting k out of n sensors already indicates that this problem is of combinatorial nature and typically hard to solve. This class of problems has many applications in engineering, especially in sensor networks in which the number of sensors allowed to communicate with a remote fusion center is limited due to bandwidth constraints. In an extreme case, the sensor scheduling problem consists of choosing one among n sensors, and transmitting only its measurement across the network.Consider the system described in the block diagram of Fig. 1, where n sensor-estimator pairs share a common wireless network, which can operate either in unicast or broadcast modes. The system operates as follows. Each of the n sensors observe a distinct random variable and reports it to the scheduler. The scheduler selects a single random variable according to a scheduling decision rule and transmits it over the network. If the network is in unicast mode, only the intended estimator receives the sensor's observation and the remaining estimators observe an erasure symbol. If the network is in broadcast mode, all the sensors receive the same transmitted measurement.Upon observing the network output, each receiver forms its estimate according to an estimation policy. The goal of the system designer is to select scheduling and estimation policies such that the mean-squared error (MSE) between M. M. Vasconcelos and U. Mitra are with the