Optical OTFS waveform PAPR analysis for high order modulation employing CNN, DNN, and AE machine learning algorithms under a variety of channel scenarios

Optical orthogonal time–frequency space (OTFS) modulation is a future technique for optical wireless communication with high mobility. Nevertheless, the high peak-to-average power ratio (PAPR) of OTFS modulation severely degrades the system performance, particularly in the case of high-order modulation formats. This paper proposes an algorithm for machine-learning (ML)-based PAPR reduction dedicated to optical OTFS under varying channel conditions, such as turbulence and multipath fading. The proposed method utilizes deep learning models to maximize signal processing and suppress peak-power variations, while ensuring signal integrity. The simulations result prove that the proposed method attains a PAPR reduction of about 4 dB and 3.8 dB for 256-QAM and 2.2 dB and 1.6 dB for 64-QAM under a Rayleigh and Rician channel at a Complementary Cumulative Distribution Function (CCDF) of 10-5, better than conventional PAPR reduction methods. Power Spectral Density (PSD) analysis verifies that ML-based techniques, such as deep neural networks (DNN), convolutional neural networks (CNN), and autoencoders (AE), are spectrally efficient with negligible out-of-band radiation of -1070 and -1470 for 256QAM with diverse channel conditions. Moreover, Bit Error Rate (BER) performance tests demonstrate an SNR improvement of 8.2 dB and 3.9 at a BER of 10-5, guaranteeing error-free data transmission for diverse mobility scenarios. Furthermore, the proposed methods are compared with partial transmission schemes (PTS), selective mapping (SLM), tone reservation (TR), companding, clipping, and filtering (C&F). The numerical results emphasize the capability of ML to improve PAPR performance without PSD and BER performance. The results are significant for future optical wireless networks, where high data rates must be sustained and nonlinear distortion minimized.