Performance Investigation of OAMSK Modulated Wireless Optical System Over Turbulent Ocean Using Convolutional Neural Networks

The adaptive demodulators based on convolutional neural networks (CNNs) are proposed for an M-ary orbital angular momentum-shift keying (OAMSK) modulated underwater wireless optical communication (UWOC) system. The impacts of different modulation orders, oceanic turbulence strengths, seawater types, signal to noise ratios (SNRs), transmission distance, and scales of CNNs are investigated. For the proposed system model, oceanic turbulence and pass loss caused by absorption and scattering are emulated by Monte Carlo (MC) methods. By comparing four CNNs, AlexNet and GoogLeNet inception v4 are adopted as the key part of M-ary OAMSK adaptive demodulating modules. After being trained by 100 000 OAM intensity specimens under unknown levels of oceanic turbulence, the suitable network parameters are achieved. Results show that the average bit error rate performances of both CNN-based demodulators outperform the traditional conjugate demodulator by several orders of magnitude. Moreover, deeper and more complex CNN would be more effective in enhancing the ABER performance under the strong oceanic turbulence condition. Furthermore, for both CNN-based and conjugate demodulators, the ABER of the system with noise will approach the saturation values when the instantaneous SNR is approximately 26 dB larger than the pass loss. This work benefits the design of OAMSK modulation-based UWOC system.