Hybrid neural coded modulation: Design and training methods

We propose a hybrid coded modulation scheme which composes of inner and outer codes. The outer-code can be any standard binary linear code with efficient soft decoding capability (e.g. low-density parity-check (LDPC) codes). is designed using deep neural network (DNN) takes the channel bits outputs modulated symbols. For training DNN, we to use loss function that inspired by generalized mutual information. resulting constellations are shown outperform conventional quadrature amplitude (QAM) based coding for order 16 64 5G LDPC