Scheduling Observers Over a Shared Channel with Hard Delivery Deadlines
We abstract the core logical functions from applications that require ultra-low-latency wireless communications to provide a novel definition for reliability. Real-time applications -- such as intelligent transportation, remote surgery, and industrial automation -- involve a significant element of control and decision making. Such systems involve three logical components: observers (e.g. sensors) measuring the state of an environment or dynamical system, a centralized executive (e.g. controller) deciding on the state, and agents (e.g. actuators) that implement the executive's decisions. The executive harvests the observers' measurements and decides on the short-term trajectory of the system by instructing its agents to take appropriate actions. All observation packets (typically uplink) and action packets (typically downlink) must be delivered by hard deadlines to ensure the proper functioning of the controlled system. In-full on-time delivery cannot be guaranteed in wireless systems due to inherent uncertainties in the channel such as fading and unpredictable interference; accordingly, the executive will have to drop some packets. We develop a novel framework to formulate the observer selection problem (OSP) through which the executive schedules a sequence of observations that maximize its knowledge about the current state of the system. To solve this problem efficiently yet optimally, we devise a branch-and-bound algorithm that systematically prunes the search space. Our work is different from existing work on real-time communications in that communication reliability is not conveyed by packet loss or error rate, but rather by the extent of the executive's knowledge about the state of the system it controls.