OAM modes classification and demultiplexing via Fourier optical neural network

Abstract

Here, we present a free-space optical communication system, adept at managing alignment deviations and the challenges posed by the atmospheric turbulence in the transmission of spatially structured laser beams, particularly those carrying orbital angular momentum (OAM). The core of our system is tilizing a hybrid convolutional neural network (HCNN). This network is uniquely characterized by the first layer designed as a Fourier optics convolutional layer, implemented optically through a 4f system. This system is dynamically controlled by high-resolution, kilohertz-fast reprogrammable digital micromirror device (DMD). The primary innovation is employing this optical-filtering-based CNN for the training and recognition of multiplexed OAM beams with their unique intensity patterns. The experiments were conducted under conditions simulating atmospheric turbulence based on modified von Kármán model. The system demonstrated promising accuracy, achieving 78.25% under weak turbulence condition and 68.43% in scenarios of strong turbulence.