Joint optimization of signal constellation bit labeling for bit-interleaved coded modulation with iterative decoding

We optimize signal constellation and bit labeling for bit-interleaved coded modulation with iterative decoding (BICM-ID). Target is to minimize bit error rate floor and signal to noise ratio for the turbo cliff. Various optimal non-orthogonal 16-QAM mappings are presented. An improvement of 0.2 dB is shown compared to state of the art orthogonal QAM-constellation with approx. the same error floor. To obtain these results, we derive the probability density function of extrinsic L-values of the demapper for perfect a priori knowledge in closed form. This allows fast computation of mutual information in the EXIT chart.