Atmospheric turbulence compensation with sensorless AO in OAM-FSO combining the deep learning-based demodulator

Abstract

This study proposed a hybrid framework that is made up of a sensorless adaptive optics (AO) and a convolutional neural network (CNN) to demodulate the multiplexed orbital angular momentum (OAM) signals. The model-based sensorless AO is applied to compensate for the distorted wavefront caused by the atmospheric turbulence in the OAM free-space optical (FSO) communication. The CNN demodulator is trained to achieve the testing accuracy of 99.15% for the testing set without the AO compensation. The newly introduced model-based sensorless AO system helps CNN to identify the information-carrying diffraction patterns of OAM beams at the receiver more accurately. We used the Monte Carlo method to testify the improved performance on the bit error rate (BER) when applied sensorless AO in combination with the CNN demodulator. Amounts of numerical results indicate the superiority of the proposed framework. The purity of OAM modes would be significantly improved, thus BERs under strong turbulence decrease from 10−2 to 10−4, which meet the requirements of the FSO system.