Throughput analysis of optical NOMA waveform through RNN and CNN neural networks with 256-QAM

Abstract Optical non-orthogonal multiple access (NOMA) is a critical communication technology that allows multiple users to access the same frequency spectrum at the same time, greatly improving spectral efficiency. This paper explores neural network-based encryption methods to protect NOMA systems with high throughput. In particular, convolutional neural networks (CNNs) and recurrent neural networks (RNNs) are used to study and improve encryption mechanisms. CNNs learn spatial features of the data, whereas RNNs are good at modeling temporal dependencies. Simulation outcomes show that RNN-based encryption provides better performance in dynamic scenarios with increased security against eavesdropping and lower bit error rate (BER). Under a Rayleigh fading channel, RNN encryption lowers the BER at SNR to 6.1 dB, which is better than CNN (8.1 dB), MIMO-NOMA (10.2 dB), OFDM (13 dB), and OTFS (15 dB). In the same way, in Rician channel conditions, RNN attains 5 dB BER, outperforming CNN (7.1 dB), MIMO-NOMA (9.1 dB), OFDM (12.1 dB), and OTFS (10.1 dB). RNN also enhances power spectral density (PSD), reducing it to −810 dB from −700 dB of CNN, −590 dB of MIMO-NOMA, and −400 dB and −510 dB of OFDM and OTFS, respectively. These advancements emphasize the compromises between complexity, throughput, and present a promising avenue for optical NOMA systems using deep learning-based methods.