Minimizing the Age of Incorrect Information With Continual Belief Learning

In the context of 6G, Status Update Systems have become a pervasive component. Typically comprising a sender and a receiver, the system functions as follows: the sender observes a remote process and transmits status updates via an unreliable wireless channel. The sender’s objective is to optimize the relevance and timeliness of the information received by the receiver by minimizing the Age of Incorrect Information (AoII), defined as the duration since the receiver had correct information regarding the observed process. AoII is a metric that captures both the timeliness of status updates and their semantic content. However, measuring AoII at the sender necessitates knowledge of the remote process’s state at any given moment, which is only attainable if the sender constantly senses. This poses a significant challenge, particularly when sensing a new status update is energy-intensive, given the fact that the senders are small devices, often powered by energy-harvesting techniques. To address this, we propose a novel approach, Continual Belief Learning, to optimize the AoII under energy constraints. We derive a belief distribution over all possible AoII values, propose a corresponding update procedure for this distribution, and use it to learn the best sensing and transmission strategies at the sender. We validate the performance of our approach through detailed numerical simulations, using measurement data from the SKAB dataset. The simulations demonstrate the superiority of Continual Belief Learning, achieving gains of up to approximately 40% when compared to established reference schemes.