A Recursive quantizer design algorithm for binary-input discrete memoryless channels

Abstract

The optimal quantization of the outputs of binary-input discrete memoryless channels is considered, whereby the optimal quantizer preserves at least a constant α-fraction of the original mutual information, with the smallest output cardinality. Two recursive methods with top-down and bottom-up approaches are developed; these methods lead to a new necessary condition for the recursive quantizer design. An efficient algorithm with linear complexity, based on dynamic programming and the new necessary optimality condition, is proposed.