Joint PAPR Reduction Technology Enabled by Neural Network-based Coding and Companding

Multi-carrier modulation technology is an advanced modulation method commonly used in broadband communication systems. However, multi-carrier signals will produce a very large Peak to Average Power Ratio (PAPR) in the time domain, which will lead to non-linear damage and seriously affects the performance of system. In this paper, a joint PAPR reduction technology enabled by neural network-based coding and companding is proposed. The frequency-domain coding is implemented by a fully connected layer, which greatly reduces the complexity and difficulty of network training. A small proportion of spread spectrum is introduced to provide coding gain, and the time-domain companding network uses a nonlinear convolutional neural network to reduce PAPR. The effect of the scheme is shown through simulation under various parameter conditions, and compared with various schemes. The simulation results show that the scheme can reduce PAPR by 5 dB while introducing rare distortion. Finally, the scheme is experimentally verified. The experimental results show that the Bit Error Ratio (BER) is reduced by 75% compared with the clipping scheme when the PAPR is reduced by 5 dB, which verifies the feasibility of the scheme.