 
Summary: Feedback Can Increase the CapacityCost Function
of Discrete Channels with Memory y
Fady Alajaji and Tom Fuja
Electrical Engineering Department and Institute for Systems Research
University of Maryland, College Park, MD 20742
ABSTRACT
We consider a binary additive noise channel, where the
noise process is a stationary mixing Markov process. Ad
ditive noise channels are symmetric in the sense that
equiprobable input vectors maximize their block mutual
information. We incorporate average power constraints on
the possible input sequences of the additive channel, thus
``destroying'' its symmetry. We show that output feed
back increases the capacitycost function of this channel.
This is achieved by establishing a tight upper bound to
the capacitycost function of the channel with no feedback
(CNFB (fi)) and a lower bound to the capacitycost func
tion with feedback (CFB (fi)). The lower bound to CFB (fi)
is then shown numerically to exceed the upper bound to
CNFB (fi). An upper bound to CFB (fi) is also obtained,
