Ratio List Decoding

Abstract

We extend the notion of list decoding to {\em ratio list decoding} which involves a list decoder whose list size is specified as a function of the number of messages $M_n$ and the block length $n$. We present necessary and sufficient conditions on $M_n$ for the existence of code sequences which enable reliable list decoding with respect to the desired list size $L(M_n,n)$. It is shown that the ratio-capacity, defined as the supremum of achievable normalized logarithms of the ratio $r(M_n,n)=M_n/L(M_n,n)$ is equal to the Shannon channel capacity $C$, for both stochastic and deterministic encoding. Allowing for random list size, we are able to deduce some properties of identification codes, where the decoder's output can be viewed as a list of messages corresponding to decision regions that include the channel output. We further address the regime of mismatched list decoding, in which the list constitutes of the codewords that accumulate the highest score values (jointly with the channel output) according to some given function. We study the case of deterministic encoding and mismatched ratio list decoding. We establish similar necessary and sufficient conditions for the existence of code sequences which enable reliable mismatched list decoding with respect to the desired list size $L(M_n,n)$, and we show that the ratio-capacity with mismatched decoding is equal to the mismatch capacity. Focusing on the case of an exponential list size $L_n=e^{n Θ}$, its comparison with ordinary mismatched decoding shows that the increase in capacity is by $Θ$ bits per channel use for all channels and decoding metrics. Several properties of the average error probability in the setup of mismatched list decoding with deterministic list size are provided.