Data-driven sensor scheduling for remote estimation in wireless networks
Abstract : 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.
? We consider a periodic superframe structure common to many existing wireless sensor network protocols. A superframe repeated every sampling interval is divided into timeslots. We assume only one pointto-point link is activated at a time.
? This made the problems simpler since scheduled sensor data can be reached to a remote estimator or the corresponding controller before the next sampling time.
? However, the number of timeslots in a single superframe is limited in existing protocols such as WirelessHART and ISA-100.
? We are planning to consider larger networks in which sensor data may not be able to reach an estimator or a controller in a single superframe.
? 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.
? 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.
? This problem lies in the category of team decision problems with non-classical information structure, which are in general very difficult to solve due to a coupling between the scheduling and estimation policies known as signaling.
• An important and surprising feature of our proposed scheduling policy is that it does not require any knowledge about the freshness of each individual information update, despite the fact that the accuracy of real-time estimation depends on data freshness.
• We also conduct comprehensive simulations to evaluate the performance of our proposed scheduling policy.
• Although some industrial standards have been proposed so far, it is not yet commonly applied to industrial process control.
• A stochastic sensor selection algorithm is proposed in when a plant is monitored by multiple sensors but only one of them can access the estimator at every time instance.
? Another line of work is to use concentration inequalities to obtain performance guarantees as a function of the size of the datasets used for training.
? Unfortunately, due to the lack of convexity and without knowledge on the probabilistic model, we cannot guarantee that the solutions found by our algorithms are in fact optimal, but we provide a learning framework which provides a systematic way to train and validate the performance of the data-driven design.
? Due to the massive number of devices and the very high demand for communication resources, the scheduler selects the pieces of information that are most relevant for a given task and discard the others, keeping the network data flow under control but at the same time achieving a good system performance.
|