Convolutional Neural Network-Assisted Optical Orbital Angular Momentum Recognition and Communication

The identification of orbital angular momentum (OAM) modes with high-accuracy and -speed is always a difficult issue in practically applying optical vortex beams (OVs). In this work, we propose and experimentally investigate a convolutional neural network (CNN) method for optical OAM mode identification and shift-keying (SK) communications. The CNN model, including convolution and pooling layers, was designed to extract mode information from the diffraction patterns produced by diffracting the OVs with a cylindrical lens. After trained with loads of studying samples, the CNN model has a good generation ability in recognizing the OAM modes of OVs ranging from -15 to 15. The recognition accuracy reaches 99% with the turbulence intensity of C
_n
²
= 1 × 10
^-13
m-2/3
, Δz = 50 m. Even under the turbulence of C
_n
²
= 1 × 10
^-12
m
^-2/3
, Δz = 50 m, the accuracy still exceeds 89%. By mapping and encoding a Lena gray image with the size of 100 × 100 pixels to two OAM channels, the OAM-SK signals with 900 modulation orders were successfully demodulated by the CNN model, and the image was well recovered after transmission. With an I5-8500 Central Processing Unit, this recognition process only takes 1 ×10
^-3
s per mode. It is anticipated that the CNN methods might provide an effective way for identifying OAM modes with high-accuracy and -speed, which may have great potentials in OAM communication, quantum information processing, and astronomical application, etc.