Experimental recognition of vortex beams in oceanic turbulence combining the Gerchberg–Saxton algorithm and convolutional neural network

In underwater wireless optical communication (UWOC), vortex beams carrying orbital angular momentum (OAM) can improve channel capacity but are vulnerable to oceanic turbulence (OT), leading to recognition errors. To mitigate this issue, we propose what we believe to be a novel method that combines the Gerchberg-Saxton (GS) algorithm-based recovery with convolutional neural network (CNN)-based recognition (GS-CNN). Our experimental results demonstrate that superposed Laguerre-Gaussian (LG) beams with small topological charge are ideal information carriers, and the GS-CNN remains effective even when OT strength C n2 is high up to 10^-11 K ² m ^-2/3. Furthermore, we use 16 kinds of LG beams to transmit a 256-grayscale digital image, giving rise to an increase in recognition accuracy from 0.75 to 0.93 and a decrease in bit error ratio from 3.98×10^-2 to 6.52×10^-3 compared to using the CNN alone.