Feedback Capacity of a Class of Symmetric Finite-State Markov Channels

Abstract

We consider the feedback capacity of a class of symmetric finite-state Markov channels. Here, symmetry (termed “quasi-symmetry”) is defined as a generalized version of the symmetry defined for discrete memoryless channels. The symmetry yields the existence of a hidden Markov noise process that depends on the channel's state process and facilitates the channel description as a function of input and noise, where the function satisfies a desirable invertibility property. We show that feedback does not increase capacity for such class of finite-state channels and that both their nonfeedback and feedback capacities are achieved by an independent and uniformly distributed (i.u.d.) input. As a result, the channel capacity is explicitly given as a difference of output and noise entropy rates, where the output is driven by the i.u.d. input.