A Low-PAPR Scheme for Millimeter-Wave Communication Systems Based on Neural Network

High peak-to-average power ratio (PAPR) is a significant drawback in millimeter-wave communication systems, leading to excessive power consumption. Reducing PAPR can enhance the efficiency of power amplifiers, thereby positively impacting the overall performance of millimeter-wave communication systems. In this paper, we propose a novel approach to mitigate PAPR using an autoencoder-based framework. Our autoencoder is constructed from a complex convolutional neural network (CNN). Through unsupervised multi-task learning, this scheme effectively lowers the system’s PAPR while maintaining superior bit error rate (BER) performance. Additionally, the signals encoded by the autoencoder are clipped to comply with the maximum power constraints of the transmitting antenna. Simulation results demonstrate that our proposed method achieves better PAPR and BER performance compared to conventional approaches.