涡旋光束携带的轨道角动量（Orbital Angular Momentum，OAM）可作为一种空间维度资源实现信息的传输，目前已在OAM通信领域被广泛研究。衍射深度神经网络（Diffractive Deep Neural Network，D2NN）各衍射层上的透过率函数可对光束的振幅和相位进行操纵，训练完成后D2NN可对涡旋光束的OAM模式进行高准确率的识别。目前研究大多是基于混合型D2NN对光轨道角动量进行研究，但是混合型D2NN模型训练过程非常复杂，也难以完成实际制备。
本文提出一种基于振幅型D2NN和相位型D2NN的OAM模式检测方法，对单OAM态和OAM复用态的D2NN解码解调进行了研究。主要研究内容：
（1）设计了一种基于D2NN的涡旋光轨道角动量检测方法。模拟生成包含涡旋光束特性（振幅、相位）的光束训练集，再使用基于D2NN的OAM光束检测模型检测不同OAM光束的拓扑荷值。通过理论研究和仿真结果分析了这种OAM检测方法，并做出相应改进。
（2）研究了单OAM态的D2NN解码解调。在OAM通信系统发射端，采用本征、径向和高阶模式的单OAM态拉盖尔-高斯光束，在接收端，采用振幅型与相位型的D2NN解调模型对OAM光束进行解调。比较了每种单OAM态D2NN解调系统的解调准确率和误码率，对振幅型D2NN解调系统与相位型D2NN解调系统之间、各振幅型解调系统之间和各相位型解调系统之间进行比较分析，并给出理论解释。
（3）研究了OAM复用态的D2NN解码解调。建立四相OAM相干复用的振幅型D2NN解调模型，根据相干OAM解调系统的解调性能，提出一种改进的OAM模式选择策略进一步降低误码率。并将基于振幅型D2NN的相干OAM解调系统与基于相位型D2NN的相干OAM解调系统进行比较，实验结果表明：在湍流条件的干扰下，基于振幅型D2NN的相干OAM解调系统性能相较于相位型D2NN解调系统，解调准确率最高可提高24.22%，误码率最高可降低0.2438。

The orbital angular momentum (OAM) carried by the vortex beam can be used as a spatial dimension resource to realize information transmission, has been widely researched in the field of OAM communication at present. The transmittance function on each diffraction layer of the diffraction deep neural network (D2NN) can manipulate the amplitude and phase of the beam. After training, D2NN can accurately identify the OAM mode of the vortex beam. Currently, most research is based on the study of optical orbital angular momentum using hybrid D2NN. The training process of hybrid D2NN models is very complex and it is difficult to complete actual manufacturing.
An OAM pattern detection method based on amplitude-D2NN and phase-D2NN are proposed in this thesis. Based on this OAM mode detection method, D2NN demodulation of single OAM state and OAM multiplexing state was researched. The main research content is as follows:
(1) A vortex optical orbit angular momentum detection method based on D2NN is designed. The beam training set containing vortex beam characteristics (amplitude and phase) is simulated, and then the OAM beam detection model based on D2NN is used to detect the topological charge of different OAM beams. This OAM detection method was analyzed through theoretical research and simulation results, and corresponding improvements were made.
(2) The D2NN decoding and demodulation of a single OAM state is studied. At the transmitter of the OAM communication system, a single OAM state Laguerre-gaussian (LG) beam using eigenmode，radial mode and high mode. At the receiver, two types of amplitude-D2NN and phase-D2NN demodulation models, were used to demodulate the OAM beams. Compared the demodulation accuracy and bit error rate of each single OAM state D2NN demodulation system, and analyzed the comparison between amplitude-D2NN demodulation system and phase-D2NN demodulation system, as well as between each amplitude type demodulation system and each phase type demodulation system, and provided theoretical explanations.
(3) The D2NN decoding and demodulation of OAM multiplexing states is studied. Established an amplitude-D2NN demodulation model for four phase OAM coherent multiplexing, based on the demodulation performance of coherent OAM demodulation systems, an improved OAM mode selection strategy is proposed to further reduce the bit error rate. And a comparison was made between the coherent OAM demodulation system based on amplitude-D2NN and the coherent OAM demodulation system based on phase-D2NN. The experimental results showed that under the interference of turbulent conditions, the coherent OAM demodulation system based on amplitude-D2NN can improve demodulation accuracy by up to 24.22% and reduce bit error rate by up to 0.2438 compared to phase-D2NN demodulation system.