Typicality with Feedback

TL;DR

This paper proves that in a feedback-enabled discrete memoryless channel, the output distribution remains close to the channel law, and applies this to establish fundamental limits in integrated sensing and communication systems.

Contribution

It introduces a general result on output types in feedback systems and applies it to derive capacity limits for integrated sensing and communication models.

Findings

Output distribution remains close to the channel law under feedback.

Fundamental limits of the feedback system match those of the open-loop system.

Results hold when the sum of error and distortion probabilities is less than one.

Abstract

The main objective of this paper is to analyze a closed-loop feedback system where a transmitter probes a discrete memoryless channel (DMC) and can adapt its inputs based on the previous channel outputs. We prove that, regardless of the transmitter's strategy, the conditional type of the outputs given the inputs remains close to the DMC transition law $P_{Y|X}$. This general result enables the study of fundamental limits in certain adaptive systems. As an application, we establish a converse result for an integrated sensing and communication (ISAC) model. In this setting, the transmitter also functions as a radar receiver, aiming to simultaneously transmit a message over the channel and estimate the channel state from the backscattered feedback signals. We show that the fundamental limits of the closed loop system are the same as of the open-loop system where the transmitter can use the feedback signal to estimate the state but not to produce adaptive channel inputs. This result holds as long as the sum of the admissible-average-decoding-error-probability, denoted $\epsilon$, and the admissible-excess-distortion-probability, denoted $\delta$, is below $1$, i.e., $\delta +\epsilon < 1$.