MAC with feedback is amongst the best understood multi-terminal channel coding
with feedback systems. Its capacity region is given by a multi-letter directed
information expression and this region is known to be strictly larger than the
no-feedback capacity region. However, no single letter characterization of the
capacity region is known.

Almost all achievable schemes for MAC with feedback use block Markov
superposition coding. Recently, Venkataramanan and Pradhan have proposed a
series of achievable block Markov superposition coding schemes that improve the
achievable rate region by increasing the number of phases in the block Markov
superposition codes.

In this paper, we study the structural properties of optimal block Markov coding
schemes by formulating them as real-time communication over MAC with feedback
and decoding with finite delay.