Learning an Adversary's Actions for Secret Communication

TL;DR

This paper explores secure communication over a wiretap channel with an active adversary, proposing a joint learning and transmission scheme that adapts to the adversary's actions to enhance security.

Contribution

It introduces a novel scheme where the transmitter learns and adapts to the adversary's actions, achieving near-hindsight optimal rates in certain channel models.

Findings

Achievable rates close to those with perfect knowledge of adversary's actions.

The proposed scheme effectively learns and adapts to adversary's behavior.

Physical-layer security can be significantly improved by learning the adversary's actions.

Abstract

Secure communication over a wiretap channel is investigated, in which an active adversary modifies the state of the channel and the legitimate transmitter has the opportunity to sense and learn the adversary's actions. The adversary has the ability to switch the channel state and observe the corresponding output at every channel use while the encoder has causal access to observations that depend on the adversary's actions. A joint learning/transmission scheme is developed in which the legitimate users learn and adapt to the adversary's actions. For some channel models, it is shown that the achievable rates, defined precisely for the problem, are arbitrarily close to those obtained with hindsight, had the transmitter known the actions ahead of time. This initial study suggests that there is much to exploit and gain in physical-layer security by learning the adversary, e.g., monitoring the environment.