Speed-Up Method for Optimal Quantizations of Binary-Input Discrete Memoryless Channels with Monge Property

摘要

This study examines quantizations for outputs of binary-input discrete memoryless channels (B-DMCs) by concatenating a DMC, so-called a quantizer, to the output of the B-DMC. In this study, we define a generalized information measures for B-DMCs, which contains the mutual information, Gallager's reliability function E_0, the α-mutual information, and the f-divergence. We call such generalized information measures the φ-information, and we evaluate the loss of quantizations by using the φ-information. The main result of this study is to reduce the computational complexity for finding an optimal quantizer by using the Monge property.