Delayed Feedback Capacity of Stationary Sources over Linear Gaussian Noise Channels

TL;DR

This paper investigates the capacity of linear Gaussian noise channels with delayed feedback, modeling the noise as ARMA processes and deriving the maximal information rate using Kalman-Bucy filtering techniques.

Contribution

It reformulates the channel as an intersymbol interference model and characterizes the feedback capacity with delayed feedback using Kalman filtering and spectral analysis.

Findings

Delayed feedback is equivalent to instantaneous feedback in a transformed channel.

A conditional Gauss-Markov source achieves the feedback capacity.

The maximal information rate is derived via the Riccati equation of the Kalman-Bucy filter.

Abstract

We consider a linear Gaussian noise channel used with delayed feedback. The channel noise is assumed to be a ARMA (autoregressive and/or moving average) process. We reformulate the Gaussian noise channel into an intersymbol interference channel with white noise, and show that the delayed-feedback of the original channel is equivalent to the instantaneous-feedback of the derived channel. By generalizing results previously developed for Gaussian channels with instantaneous feedback and applying them to the derived intersymbol interference channel, we show that conditioned on the delayed feedback, a conditional Gauss-Markov source achieves the feedback capacity and its Markov memory length is determined by the noise spectral order and the feedback delay. A Kalman-Bucy filter is shown to be optimal for processing the feedback. The maximal information rate for stationary sources is derived in terms of channel input power constraint and the steady state solution of the Riccati equation of the Kalman-Bucy filter used in the feedback loop.