Research on Autoencoders for Higher Order Modulation

Aiming at the problem that previous autoencoders (AEs) were generally unable to handle higher-order modulated signals with a input encoded by one-hot, a strategy of segmented encoding of the input data and a parallel-convolutional neural network (PCNN) model are used to apply the AE to high order modulation. The simulation results show that the proposed scheme can handle 4096QAM signals and has an acceptable classification accuracy which is beyond the traditional algorithm. In addition, the Gate Recurrent Unit (GRU) layers are introduced into autoencoder for channel coding and result a huge gain.