Coded Power Control: Performance Analysis

TL;DR

This paper introduces coded power control (CPC), embedding information into power levels for interference channels, and provides a theoretical framework to analyze its performance in a multi-decision-maker setting with asymmetric information.

Contribution

It develops a general information-theoretic framework for decision-making with asymmetric information and imperfect monitoring, applying it to analyze CPC in interference channels.

Findings

Derived a general information constraint for decision-makers with asymmetric information.

Characterized the performance limits of CPC policies in interference channels.

Extended Shannon-theoretic tools to scenarios with imperfect monitoring and asymmetric knowledge.

Abstract

In this paper, we introduce the general concept of coded power control (CPC) in a particular setting of the interference channel. Roughly, the idea of CPC consists in embedding information (about a random state) into the transmit power levels themselves: in this new framework, provided the power levels of a given transmitter can be observed (through a noisy channels) by other transmitters, a sequence of power levels of the former can therefore be used to coordinate the latter. To assess the limiting performance of CPC (and therefore the potential performance brought by this new approach), we derive, as a first step towards many extensions of the present work, a general result which not only concerns power control (PC) but also any scenario involving two decision-makers (DMs) which communicate through their actions and have the following information and decision structures. We assume that the DMs want to maximize the average of an arbitrarily chosen instantaneous payoff function which depends on the DMs' actions and the state realization. DM 1 is assumed to know non-causally the state (e.g., the channel state) which affects the common payoff while the other, say DM 2, has only a strictly causal knowledge of it. DM 1 can only use its own actions (e.g., power levels) to inform DM 2 about its best action in terms of payoff. Importantly, DM 2 can only monitor the actions of DM 1 imperfectly and DM 1 does not observe DM 2. The latter assumption leads us to exploiting Shannon-theoretic tools in order to generalize an existing theorem which provides the information constraint under which the payoff is maximized. The derived result is then exploited to fully characterize the performance of good CPC policies for a given instance of the interference channel.