【作者】 李丽红；

【导师】 马荣；

【作者基本信息】 山西大学 ， 通信工程， 2025， 硕士

【摘要】 自由空间光（FSO,Free Space Optical）通信是一种前沿通信技术,其利用大气作为传输介质,突破了传统有线通信的物理限制。与光纤通信和移动通信技术相比,FSO具有通信容量大、频谱资源不受限、抗电磁干扰能力强、部署便捷以及灵活性高等显著优势。基于FSO系统的高带宽特性,其在大数据时代背景下能够有效满足海量数据传输需求,展现出突出的技术优越性。因此,FSO系统具有广阔的发展前景和巨大的应用潜力。近年来,FSO的研究与开发取得了显著进展,其中轨道角动量（OAM,Orbital Angular Momentum）作为一种独特的光学特性发挥了关键作用。OAM模式具有正交性,通过调制OAM光束携带信息,能够提升FSO系统的信道容量。针对利用单个整数OAM阶实现信息传递的有限性,分数阶OAM叠加态可以有效拓展通信容量。并且在设备误差和大气湍流的影响下,现有的光学OAM模式识别技术存在明显局限性,导致OAM光束检测精度下降,从而制约了FSO系统的整体性能优化。针对上述问题,本论文研究基于深度学习的分数阶OAM模式叠加态识别。该方案通过神经网络的自适应特征提取能力,实现了OAM光束的高精度识别,有效拓展了通信容量,克服了传统方法在设备误差和大气湍流影响下的局限性。论文主要研究内容如下:基于小数据集的神经网络实现非对称分数阶OAM叠加态的识别。实验上利用空间光调制器制备了非对称分数阶OAM叠加态光束,并在CCD上捕获该光束的强度图像,用于制作模型所需数据集。通过设计一个7层神经网络,实现了非对称分数阶OAM叠加态的19分类识别,识别准确率达到99.34%。这一成果为非对称分数阶OAM叠加态的识别提供了一种高准确率且数据量需求小的全新方法,有望在基于OAM的光信息处理领域发挥重要作用,展现出较好的应用潜力。基于深度学习的大气湍流中OAM模式识别方法的研究。基于Matlab数值仿真平台,系统模拟了不同传输距离（1000米、1500米、2000米）和不同湍流强度(C_n²=10^-15m^-2/3、10^-14m^-2/3、10^-13m^-2/3)条件下OAM光束的传输演化过程,构建了包含多种传输距离、多种湍流常数的光强分布图像数据集。基于此,本研究对神经网络模型进行了优化,通过引入深度可分离卷积和残差连接,使模型能够有效提取湍流畸变图像中的关键特征,在测试集上实现了高达100%的识别准确率,显著提升了模型在湍流环境下的适应性。为进一步简化湍流环境下的研究流程,本研究自主开发了一套智能化大气湍流模拟系统。该系统可实时生成不同湍流条件下的光强分布图,为湍流环境下OAM光束的特性研究和算法验证提供了高效可靠的研究工具。更多还原

【Abstract】 Free-space optical communication（FSO）represents a cutting-edge communication technology that utilizes the atmosphere as its transmission medium,overcoming the physical limitations inherent in traditional wired communication systems.Compared to conventional fiber-optic and mobile communication technologies,FSO exhibits several distinctive advantages:large communication capacity,unrestricted spectrum resources,strong resistance to electromagnetic interference,ease of deployment,and high flexibility.Leveraging its high-bandwidth characteristics,FSO systems can effectively meet the demands of massive data transmission in the era of big data,demonstrating remarkable technical superiority.Consequently,FSO technology holds broad development prospects and tremendous application potential in modern communication systems.In recent years,significant progress has been made in the research and development of FSO communication,where Orbital Angular Momentum（OAM）has emerged as a pivotal optical property.The orthogonality of OAM modes enables enhanced channel capacity in FSO systems through OAM-based information encoding.While conventional integer OAM modes present limited information-carrying capacity,fractional OAM superposition states offer a promising solution for capacity expansion.However,current optical OAM recognition techniques exhibit notable limitations when subjected to equipment imperfections and atmospheric turbulence effects,leading to degraded detection accuracy that ultimately constrains the overall performance optimization of FSO systems.To address these challenges,this study investigates deep learning-based recognition of fractional OAM mode superpositions.The proposed approach leverages neural networks’adaptive feature extraction capability to achieve high-precision OAM beam identification,effectively expanding communication capacity while overcoming the limitations of conventional methods under equipment imperfections and atmospheric turbulence.The main research work and achievements of this thesis are as follows:Identifying asymmetric susperposition of fractional orbital-angular-momentum modes via netural network with small dataset.In experiments,asymmetric fractional OAM superposition beams were prepared using a spatial light modulator（SLM）.Intensity images of these beams were captured by a charge-coupled device（CCD）to construct the dataset required for model training.A seven-layer neural network was designed to achieve 19-class classification recognition of asymmetric fractional-OAM superposition states,with an identification accuracy of 99.34%.This work provides a novel,high-precision,and data-efficient approach for recognizing asymmetric fractional-OAM superposition states,demonstrating significant potential for applications in OAM-based optical information processing.Research on OAM mode recognition methods in atmospheric turbulence based on deep learning.Based on the Matlab numerical simulation platform,this study systematically simulated the propagation and evolution of OAM beams under different transmission distances（1000 m,1500 m,2000 m）and varying turbulence intensities(C_n²=10^-15m^-2/3,10^-14m^-2/3,10^-13m^-2/3),constructing a dataset containing light intensity distribution images across multiple transmission distances.Building upon this,the neural network model was optimized by incorporating depthwise separable convolution and residual connections,enabling the model to effectively extract key features from turbulence-distorted images.The optimized model achieved a recognition accuracy of up to 100%on the test set,significantly enhancing its adaptability in turbulent environments.To further streamline the research process under turbulence conditions,this study independently developed an intelligent atmospheric turbulence simulation system.This system can generate light intensity distribution patterns under different turbulence conditions in real time,providing an efficient and reliable research tool for studying OAM beam characteristics and validating algorithms in turbulent environments.更多还原