A Novel PAPR Reduction Scheme Based on Joint Traditional Algorithm and Machine Learning for CO-OFDM Systems

The high peak-to-average power ratio (PAPR) deteriorates the performance of coherent optical orthogonal frequency division multiplexing (CO-OFDM) systems. Machine learning (ML) has evolved into a powerful technology to reduce PAPR with its flexible learning ability. We propose a joint scheme which adopts iterative partial transmission sequence (IPTS) scheme cascading a feedforward neural network (FNN) trained on iterative
$\pmb {\mu }$
-law companding and filtering (IMCF) algorithm. The proposed scheme greatly reduces the complexity of partial transmission sequence (PTS) scheme, which has excellent PAPR reduction effect and BER performance. Our method achieves a 5.03 dB PAPR reduction for
$10\pmb {^{-4}}\,\,\pmb {CCDF}$
and an additional transmission distance of 200 km SMF for
$10\pmb {^{-3}}\,\,\pmb {BER}$
, compared with the original OFDM signal. Meanwhile, the computational complexity of proposed scheme is reduced by 32.4%, compared with IPTS cascading IMCF (IPTS-IMCF) scheme.