Neural network-based peak power and throughput analysis of optical OFDM waveforms with 1024-QAM for high-capacity computer networks

Abstract This work presents a neural network-based framework for peak power suppression and throughput enhancement in optical OFDM systems employing 1024-QAM modulation for high-capacity computer networks. Optical OFDM signals inherently suffer from high peak-to-average power ratio (PAPR), which leads to nonlinear distortions in the optical channel and degrades bit error rate (BER) performance. To address this challenge, a convolutional neural network (CNN)-based scheme is proposed and evaluated under different subcarrier configurations. The analysis demonstrates that the proposed method consistently achieves significantly lower PAPR values, recording only 1.8 dB, 3.5 dB, and 5.5 dB at a complementary cumulative distribution function (CCDF) of 10 −3 for 64, 256, and 512 subcarriers, respectively. In comparison, conventional schemes such as SLM, PTS, SVM, and RNN exhibit considerably higher PAPR levels. Similarly, BER performance analysis shows that the proposed method achieves target BER of 0.001 at 7 dB and 9 dB SNR for 256 and 512 subcarriers, representing up to 13 dB improvement over traditional techniques. These results confirm the robustness, scalability, and efficiency of the proposed NN framework in reducing peak power and improving signal reliability. The findings highlight its potential for integration into next-generation optical communication and computer networking systems requiring high spectral efficiency and energy efficiency.