Supervised RNN-PTS framework for PAPR reduction and performance optimization in optical OTSM-VLC systems

This paper presents a supervised recurrent neural network–based partial transmit sequence (S–RNN–PTS) approach for reducing – the – peak-to-average power ratio (PAPR) in optical orthogonal time–sequence multiplexing (OTSM) visible-light communication (VLC) systems. A high PAPR in intensity–modulation/direct–detection (IM/DD) links introduces severe nonlinear distortion owing to LED nonlinearity, degrading power efficiency, and link reliability. The proposed framework integrates supervised learning with the PTS structure to enable search-free phase prediction at a fixed inference complexity, while preserving the conventional OTSM architecture. The performance was evaluated through Monte Carlo simulations using M−QAM−modulated OTSM waveforms over IM/DD VLC channels with nonlinear LED characteristics. The results demonstrate a PAPR reduction of up to 7 dB compared to conventional PTS schemes, and an SNR improvement of approximately 8 dB at a bit-error-rate (BER) of 10 −3 for large subcarrier configurations. These gains enhance the spectral compactness and robustness against nonlinear distortion, making the S–RNN–PTS framework highly suitable for high-speed, energy-efficient OTSM–VLC systems.