Deep Learning Based Constellation Rearrangement Design in Hybrid ARQ System
Cao Bo, Ming Jiang, Chunming Zhao, Longhao Zou · 2022
This paper proposes a constellation rearrangement autoencoder (CoRe-AE) for hybrid automatic repeat request (HARQ) system through a learning-based neural network. Two identical autoencoder networks are used in the CoRe-AE scheme respectively corresponding to the first data transmission and the retransmission. It can both optimize the bit mapping and the geometric shaping of the constellation points during the two transmissions. It is shown that the bit error rate (BER) performances of our constellations learned by CoRe-AE are noticeably better than those of the regular square quadrature amplitude modulation (QAM) and some well-known geometric shaped constellations with conventional constellation rearrangement (CoRe) scheme in HARQ systems.