On the Role of Age and Semantics of Information in Remote Estimation of Markov Sources

TL;DR

This paper introduces a semantics-aware remote estimation framework for Markov sources, optimizing transmission policies by considering both the age and significance of information to improve estimation accuracy.

Contribution

It formulates a constrained Markov decision process incorporating semantics and age, deriving optimal policies and a novel algorithm for efficient computation.

Findings

Optimal mixture policy exists and is characterized by a closed-form expression.

Threshold-based transmission policies depend on estimation error and AoI.

Incorporating semantics and age improves estimation performance significantly.

Abstract

This paper studies semantics-aware remote estimation of Markov sources. We leverage two complementary information attributes: the urgency of lasting impact, which quantifies the significance of consecutive estimation error at the transmitter, and the age of information (AoI), which captures the predictability of outdated information at the receiver. The objective is to minimize the long-run average lasting impact subject to a transmission frequency constraint. The problem is formulated as a constrained Markov decision process (CMDP) with potentially unbounded costs. We show the existence of an optimal simple mixture policy, which randomizes between two neighboring switching policies at a common regeneration state. A closed-form expression for the optimal mixture coefficient is derived. Each switching policy triggers transmission only when the error holding time exceeds a threshold that depends on both the instantaneous estimation error and the AoI. We further derive sufficient conditions under which the thresholds are independent of the instantaneous error and the AoI. Finally, we propose a structure-aware algorithm, Insec-SPI, that computes the optimal policy with reduced computation overhead. Numerical results demonstrate that incorporating both the age and semantics of information significantly improves estimation performance compared to using either attribute alone.