A Novel PAPR Reduction of OFDM Based on Deep Learning

The high peak-to-average power ratio (PAPR) in orthogonal frequency division multiplexing (OFDM) systems is still an open issue. In this paper, we propose a new low- complexity Mapping and Demapping Network (MD-Net) based on deep learning for OFDM systems to improve the PAPR and BER performance. And the proposed MD-Net corresponds a new transceiver architecture, in which it adopts multiple convolution and full connection layers to realize adaptive con- stellation mapping at transmitter and demapping at receiver both in frequency domain. Compared with some benchmark methods, the simulation results demonstrate that the proposed method achieves comparable PAPR and BER performance, and lower training complexity.