Exploiting Data Significance in Remote Estimation of Discrete-State Markov Sources

TL;DR

This paper develops a novel framework for remote estimation of Markov sources with different priorities, introducing new metrics to quantify the impact of errors and designing threshold-based policies to optimize trade-offs between accuracy and communication costs.

Contribution

It introduces the Age of Missed Alarm and Age of False Alarm metrics, formulates a countably infinite-state MDP, and derives threshold policies for optimal remote estimation with prioritized error impacts.

Findings

Threshold policies with distinct age-based thresholds are optimal.

Finite-state approximations converge exponentially fast to the original MDP.

Numerical results validate the effectiveness of the proposed policies.

Abstract

We consider semantics-aware remote estimation of a discrete-state Markov source with both normal (low-priority) and alarm (high-priority) states. Erroneously announcing a normal state at the destination when the source is actually in an alarm state (i.e., missed alarm) incurs a significantly higher cost than falsely announcing an alarm state when the source is in a normal state (i.e., false alarm). Moreover, consecutive estimation errors may cause significant lasting impacts, such as maintenance costs and misoperations. Motivated by this, we introduce two new metrics, the Age of Missed Alarm (AoMA) and the Age of False Alarm (AoFA), to capture the lasting impacts incurred by different estimation errors. Notably, these two age processes evolve interdependently and distinguish between different error types. Our goal is to design a transmission policy that achieves an optimized trade-off between lasting impact and communication cost. The problem is formulated as a countably infinite-state Markov decision process (MDP) with an unbounded cost function. We show the existence of a simple switching policy with distinct thresholds for each age process and derive closed-form expressions for its performance. For symmetric and non-prioritized sources, we show that the optimal policy reduces to a threshold policy with identical thresholds. For numerical tractability, we propose a finite-state approximate MDP and prove that it converges exponentially fast to the original MDP in the truncation size. Finally, we develop an efficient search algorithm to compute the optimal switching policy and validate our theoretical findings with numerical results.