Neural Network Assisted Active Constellation Extension for PAPR Reduction of OFDM System

Abstract

One of the major drawbacks of Orthogonal Frequency Division Multiplexing (OFDM) is its high Peak-to-Average Power Ratio (PAPR) problem, which may result in nonlinear signal distortion and thus significantly reduces the efficiency of the power amplifier. In this paper, we propose a novel Neural Network assisted Active Constellation Expansion (NN-ACE) method to reduce the PAPR of OFDM symbols. The extension vector of ACE is learned by an autoencoder to reduce the PAPR while keeping the signal power increment low. Moreover, a compromise between PAPR-reduction and power-increment can be adjusted by a weight factor in the loss function according to different requirements. The proposed neural network based ACE scheme is proved to be efficient of achieving lower PAPR and thus reduce the bit error rate (BER) in a nonlinear channel model.