Autoencoder-driven PTS for power optimization of optical-NOMA waveform amplifiers

This paper proposes an autoencoder-based partial transmit sequence (AE-PTS) framework for peak-to-average power ratio (PAPR) reduction and power efficiency enhancement in optical non-orthogonal multiple access (Optical-NOMA) systems with nonlinear optical amplifiers. The encoder learns compact representations of multi-user waveforms, while a differentiable PTS module generates candidate phase-rotated subblocks. A lightweight neural selector optimally minimizes peak power, and the decoder reconstructs compliant optical waveforms under real-valued intensity and non-negativity constraints. Unlike conventional techniques, AE-PTS jointly optimizes phase selection and waveform representation, effectively reducing PAPR, mitigating nonlinear distortion, and improving BER. Extensive MATLAB simulations for 64, 128, and 256 sub-carrier optical-NOMA confirm its robustness. At a CCDF of 10 −3 , AE-PTS achieves PAPR values of 1.8 dB, 2.8 dB, and 5.2 dB, corresponding to gains of up to 10 dB over baseline optical-NOMA and 2–5 dB over conventional PTS. Capacity analysis shows AE-PTS reaching ∼200 at 50 dB SNR, compared with 100 for baseline, while training accuracy results demonstrate stable convergence across sub-carrier sizes. These simulation results establish AE-PTS as a scalable and energy-efficient solution for future optical-NOMA networks.