A CNN-based method for non-parallel misaligned OAM modal recognition of arrays

In this paper, a modal identification method combining phase compensation and machine learning is proposed for the modal identification challenge in the case of non-parallel misalignment of transceiver arrays in orbital angular momentum (OAM) communication. Firstly, the method is based on the equivalence principle to utilize the spatial phase difference to obtain the compensated weighted phase of the receiving array elements and establish the receiving array equivalent model, through which the equivalence mechanism is used to recover the vortex phase wavefront distorted due to the non-alignment of the transceiver arrays. And secondly, the compensated phase and amplitude data are integrated into a two-channel dataset, which is used for the training of a convolutional neural network (CNN) model. Lastly, the modal identification is carried out using the trained network. The paper also describes the traditional modal identification method for amplitude detection (AD), and verifies the accuracy and stability of the proposed identification method through comparative analysis of simulation examples. The simulation results confirm that the proposed phase compensation method can effectively recover the vortex phase wavefront of the signal with the advantage of low complexity. Under OAM indexed coding modulation, the proposed modal recognition method achieves significant improvement in bit error rate (BER) performance and shows superior robustness compared with the traditional method. Characterized by low complexity, high accuracy and high stability, this method provides an efficient modal reception identification solution for OAM short-range high-speed communication in the transceiver array non-parallel misalignment scenario.