Identification of hybrid orbital angular momentum modes with deep feedforward neural network

• Published in: Results in physics Vol. 15; p. 102790
• Main Authors: Huang, Zebin, Wang, Peipei, Liu, Junmin, Xiong, Wenjie, He, Yanliang, Zhou, Xinxing, Xiao, Jiangnan, Li, Ying, Chen, Shuqing, Fan, Dianyuan
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.12.2019; Elsevier
• Subjects: 1-dimension fork grating; Feedforward neural network; Hybrid orbital angular momentum modes; Vortex beams; Hybrid orbital angular momentum modes; 1-dimension fork grating; Vortex beams; Feedforward neural network
• ISSN: 2211-3797; 2211-3797
• DOI: 10.1016/j.rinp.2019.102790

•An FNN based hybrid OAM modes identification method was proposed.•This method can reach the accuracy of 97% with five superimposed OAM modes, and the identification process takes only 0.09 ms.•Under the influence of atmospheric turbulence with Cn2=1×10-15m-2/3, the identification accuracy of the FNN still exceed 98% at the transmission distance of 1000 m.•A 120-ary OAM-SK communication link was constructed, and the signals were demodulated by the FNN model with only 4.3×10-3 pixels error rate. The identification of hybrid orbital angular momentum (OAM) modes with high-accuracy and -speed is always a challenge in practically applying optical vortex beams (VBs). In this paper, we propose and investigate a deep feedforward neural network (FNN) method for identifying the hybrid OAM modes of VBs. The FNN model with 15 input neurons and 7 hidden layers was constructed. And an improved1-dimension fork grating (1-D FG) was designed to diffract VBs and produce feature parameters which are used as the input of the FNN. After supervised training, the deep FNN model can identify arbitrarily combined hybrid OAM modes with a wide detection range owing to its non-linear operations to neurons and massive iterations. Besides, this FNN model has better robustness to atmospheric turbulence. The results show that the identification accuracy reaches 97% with five superimposed modes. Under the influence of atmospheric turbulence with Cn2=1×10-15m-2/3, the accuracy still exceeds 98% at the transmission distance of 1000 m. With an Intel Core i5-4590 CPU and NVIDIA GeForce GTX 750 GPU, this identification process takes only 0.09 ms. Furthermore, we constructed a 120-ary OAM shift-keying communication link, and the signals were demodulated by the FNN model with only 4.3×10-3 pixels error rate. It is anticipated that the FNN might pave an effective way to detect hybrid OAM modes, which may have great potentials in increasing communication capacity and modulation ability.