Goal-Oriented Communication: When is Semantic Feedback Useful?

The aim is to acquire knowledge about fundamental limits, performance evaluation, and design principles for goal-oriented wireless communication operating in the short-packet regime. These are systems where the communicating entities collaborate to reach a goal, implying that only data that is relevant towards this goal is meaningful to communicate. Focus is on the important special case of remote estimation, i.e., when the goal is to estimate a random process at one or several monitor nodes (network edge devices) based on noisy measurements acquired at a potentially large number of sensor nodes and where communication between monitors and sensors is over a shared wireless channel. Our approach involves devising an accurate, rigorous, and easy-to-use analysis framework based on finite-blocklength information theory for such systems.We hypothesize that the key for making such systems resource-efficient is to design the feedback policy carefully, i.e., to determine when and what information the monitor nodes should feedback to the sensor nodes,such that the sensor nodes can determine when to measure the random process and how to compress the measurement and its timestamp to as few bits as possible,subject to that the estimation performance is sufficiently good.In particular, since the cost (energy, bandwidth, etc.) for feedback can be significant, we aim to provide a comprehensive answer to the question: "for given scenario, how much feedback is cost-effective?"