Quantization of Log-Likelihood Ratios to Maximize Mutual Information

We propose a quantization scheme for log-likelihood ratios which optimizes the trade-off between rate and accuracy in the sense of rate distortion theory: as distortion measure we use mutual information to determine quantization and decision levels maximizing mutual information for a given rate over a Gaussian channel. This approach is slightly superior to the previously proposed idea of applying the Lloyd-Max algorithm to the dasiasoft bitpsila density associated to the L-values. A further data rate reduction can be achieved with entropy coding, because the optimum quantization levels based on mutual information are used with pronounced unequal probabilities.