Convolutional Neural Network Based Atmospheric Turbulence Compensation for Optical Orbital Angular Momentum Multiplexing

Abstract

Atmospheric turbulence in free-space will distort the helical phase-front of vortex beams (VBs) and cause mode diffusion, seriously hindering the practical application of optical orbital angular momentum (OAM) communications. Here, we propose and experimentally investigate a convolutional neural network (CNN)-based atmospheric turbulence compensation method for OAM multiplexing communication. Taking advantage of signal processing, we design a CNN model that can automatically extract the characteristic parameters from the distorted intensity distribution of VBs. After supervisory training, the CNN model possesses a strong generalization ability and can efficiently predict the equivalent turbulence phase screen. Under the influence of the turbulence with $\boldsymbol{C}_n^{\bf{2}} = {\bf{5}} \times {\bf{1}}{{\bf{0}}^{ - {\bf{13}}}}{\textbf{m}^{ - {\bf{2}}/{\bf{3}}}}$ and $\Delta {\bf{z}}{\kern 1pt} = {\kern 1pt} {\bf{{50}}}\,{\mathbf{m}}$, the mode purity of the distorted VB improves from 26.91% to 93.12% through the compensation. By constructing an OAM multiplexing communication link with the bit-rate of 100 Gbit/s and employing the CNN model to equalize the OAM channels, the bit-error-rates are decreased by three orders of magnitude, and the measured crosstalk is reduced from -23.15 dB to -29.46 dB. Moreover, the constellations converge obviously at the signal-to-noise ratio of 20 dB, and the error-vector-magnitude decreases from 0.3337 to 0.1622. These results indicate that the CNN model can well compensate the atmospheric turbulence induced distortion in VBs, and may open new avenues for improving the performance of OAM communications.