Capacity of Random Channels with Large Alphabets

Abstract: We consider discrete memoryless channels with input alphabet size $n$ and output alphabet size $m$, where $m=$ceil$(\gamma n)$ for some constant $\gamma&gt;0$. The channel transition matrix consists of entries that, before being normalised, are independent and identically distributed nonnegative random variables $V$ and such that $E[(V \log V)<sup>2]&lt;\infty$. We prove that in the limit as $n\to \infty$ the capacity of such a channel converges to $Ent(V) / E[V]$ almost surely and in $L<sup>2$, where $Ent(V):= E[V\log V]-E[V] \log E[V]$ denotes the entropy of $V$. We further show that, under slightly different model assumptions, the capacity of these random channels converges to this asymptotic value exponentially in $n$. Finally, we present an application in the context of Bayesian optimal experiment design.