On Remote Estimation with Multiple Communication Channels

TL;DR

This paper investigates remote estimation with multiple channels, revealing that traditional threshold policies may be suboptimal and proposing a threshold-in-threshold policy under certain assumptions, supported by numerical analysis.

Contribution

It introduces a new problem setting with a noisy and a perfect channel, and proves the optimality of threshold-in-threshold policies under specific conditions.

Findings

Symmetric threshold policies can be suboptimal in this setting.

Threshold-in-threshold policies are optimal under certain assumptions.

Numerical results reveal surprising properties inherited from classical settings.

Abstract

This paper considers a sequential estimation and sensor scheduling problem in the presence of multiple communication channels. As opposed to the classical remote estimation problem that involves one perfect (noiseless) channel and one extremely noisy channel (which corresponds to not transmitting the observed state), a more realistic additive noise channel with fixed power constraint along with a more costly perfect channel is considered. It is shown, via a counter-example, that the common folklore of applying symmetric threshold policy, which is well known to be optimal (for unimodal state densities) in the classical two-channel remote estimation problem, can be suboptimal for the setting considered. Next, in order to make the problem tractable, a side channel which signals the sign of the underlying state is considered. It is shown that, under some technical assumptions, threshold-in-threshold communication scheduling is optimal for this setting. The impact of the presence of a noisy channel is analyzed numerically based on dynamic programming. This numerical analysis uncovers some rather surprising results inheriting known properties from the noisy and noiseless settings.