Locally Optimal Message Quantization for Binary LDPC Decoding

摘要

For decoding of LDPC codes, the concatenation of an arbitrary discrete memory less channel followed by a quantizer is of interest. For a restricted quantizer alphabet size, it is shown that the maximum of the mutual information between the channel input and the quantizer output can be found by concave programming. The optimal quantizer is shown to be deterministic. For the binary-input case, a dynamic programming method is given which can effectively find a good quantizer.