A New Approach to Dealing with the PAPR Problem for OFDM Systems via Deep Learning

Deep Learning is a newly emerging tool in the field of signal processing and brings a promising way to solve some tough problems. In this paper, a novel method called deep learning based clipped reconstruction network (DL-Clip) is proposed for OFDM systems as an effective technique to overcome the PAPR problem, which is a prominent weakness for the OFDM waveform. Specifically, the clipping operation de-creases the high PAPR by restricting the signal amplitude below the predetermined threshold at the transmitter, while DL-Clip exploits the correlation in the frequency domain to reconstruct the signal. Simulation results indicate that the proposed method has obvious advantages in reducing PAPR and can save 1 dB of transmit power while achieving the same BER as compared to theoretical OFDM, which effectively reduces energy consumption and transmitting costs, thus satisfying the requirements of green communication.