Performance comparison of two OAM shift keying CNN demodulating schemes

Abstract

The technique of orbital angular momentum (OAM) increases the channel capacity in free space optical (FSO) communications. Lately, the convolutional neural network (CNN)-based demodulating for the OAM shift keying (OAM-SK) draws great attention. Unfortunately, the atmospheric turbulence (AT) causes a critical challenge of wavefront distortion for OAM-SK-FSO links. In this study, we analyze two different methods to resist the negative influence caused by the AT in the OAM-SK-FSO link. The first method is to compensate for the wavefront by using the adaptive optics system (AOS) directly. On the contrary, the second method is to increase the CNN demodulator's recognition rate for all ATs by deeply digging into the OAM image dataset. An AT-detecting based multiple CNN (ATDM-CNN) demodulator, rather than the costly physical devices AOS, is proposed to achieve a qualified OAM modes recognition rate. The AT detector detects the AT strength, and an AT-determined demodulator is activated to recognize the incident OAM modes. The two methods are compared by simulation in different AT cases. Satisfactory results indicate the improvement in the recognition rate for each method. The advantages and disadvantages of these two methods are also listed.