A novel PAPR reduction scheme based on nonlinear real-valued support vector regression for CO-OFDM systems

Coherent optical orthogonal frequency division multiplexing (CO-OFDM) systems suffer from the high peakto-average ratio (PAPR) of orthogonal frequency division multiplexing (OFDM) signals, which leads to the back-off of laser power and fiber nonlinear effects, thereby deteriorating the bit error rate (BER) performance and limiting the transmission distance. Iterative clipping and filtering (ICF) scheme, as a direct, simple and effective PAPR reduction scheme that only processes signals at the transmitter, but requires multiple fast Fourier transform (FFT)/inverse fast Fourier transform (IFFT), resulting in high computational complexity. We propose a novel scheme using the machine learning (ML) technique, which is based on nonlinear real-valued support vector regression (NRSVR) and trained on ICF scheme. The simulation results show that compared with ICF scheme, the proposed scheme has a similar ??? ?????? ???0 for 10-4 complementary cumulative distribution function (????????????), an additional 15 km of single mode fiber (SMF) transmission for 16 dB optical signal-to-noise ratio (????????????) and 10-3 ?????????, and a significant time complexity reduction. Compared with other ML-based schemes, the proposed scheme also achieves better performance and lower computational complexity.