Optimal quantization of B-DMCs maximizing α-mutual information with monge property
Yuta Sakai, Ken‐ichi Iwata · 2017
This study examines quantization for outputs of binary-input discrete memoryless channels (B-DMCs) by concatenating its output with another DMC, so-called a quantizer. As an objective function of channel quantization, we employ the α-mutual information of a B-DMC, which connects to more powerful coding theorem than the ordinary mutual information. Showing a Monge property of the α-mutual information, we propose an optimal quantizer design algorithm for given B-DMC in polynomial time complexity with respect to the output alphabet size and the quantized level. Since the proposed method employs the SMAWK algorithm due to the Monge property, our algorithm is faster than a naive dynamic programming.