A Demodulator for UWOC-OAM-SK Systems Based on Convolutional Neural Networks

In order to identify the different modes of vortex beams in the underwater OAM shift keying (OAM-SK) system more conveniently and accurately, a deep learning technique is introduced, and an improved convolutional neural networks (CNN) model is proposed as the demodulator of the communication system. Based on the vortex beam transmission characteristics and the fundamental theory of ocean turbulence, a random phase screen is generated by the power spectrum inversion method with low frequency compensation to simulate the transmission of vortex beam in different ocean turbulence channels. The recognition rate of this CNN model as a system demodulator for the damaged light intensity distribution maps of LG beams, BG beams and superposed state LG beams (S-LG) after passing through ocean turbulence is investigated. The experimental results show that the CNN model has strong generalization ability and can effectively identify different kinds of vortex beams and its modes. The recognition rate can reach more than 97% for superposition state LG beams after 100m transmission in high intensity turbulent environment. The model has some reference value for the research design of OAMSK system.