Information Freshness Optimization and Semantic-aware Status Updating

Future Internet-of-Things (IoT) applications will increasingly rely on sharing time-sensitive information for monitoring and control, which is referred to as status updating. Examples include autonomous vehicles, digital health, cyber-physical systems, smart factories, and smart homes. Recently, the age of information (AoI) was proposed as a metric to measure the freshness of information at the intended destination. In addition to AoI, other metrics such as the age of incorrect information (AoII) and distortion measures were introduced to address the shortcomings of AoI by taking into account the semantics of data transfer. The objective of this thesis is to develop novel sampling and scheduling algorithms to optimize information freshness and semantic-aware communication, constrained by limited network resources.

First, the thesis addresses the problem of minimizing the (time) average AoI in a relaying status updating system with stochastic arrival sources. Using tools from constrained Markov decision processes (CMDP), Lyapunov optimization theory, and deep reinforcement learning (DRL), several scheduling policies are developed and their performance is numerically demonstrated. Additionally, some structural analysis of the derived policies is provided.

Then, the thesis addresses the problem of average query AoI (QAoI) minimization in a heterogeneous status updating system, i.e., coexistence of the generate-at-will and stochastic arrival sources, subject to individual constraints on sampling and transmission costs. Using the CMDP approach, an optimal joint sampling and scheduling policy is developed. Furthermore, to strike a balance between performance and complexity, a near-optimal low-complexity policy is devised.

The thesis also addresses the problem of real-time remote tracking of a partially observable Markov source in an energy harvesting system with both sampling and transmission costs. To this end, two semantic-aware performance metrics, namely, a generic distortion measure and AoII, are proposed. For each metric, joint sampling and scheduling policies are developed using partially observable MDP (PODMP) and its belief MDP formulation. Particularly, the belief is expressed as a function of AoI, and then by bounding the AoI, a finite-state MDP problem is cast and solved.