Deep Learning Based Information Extraction And Communication Of Spatial Structured Light

Spatial structured light field refers to a light field with special distribution of phase,polarization,etc.,which is typically vortex beam,cylindrical vector beam and so on.Structured light field presents unique dynamic characteristics,Orbital Angular Momentum(OAM)characteristics and topological structure,which opens new channels for the applications such as information transmission based on light wave or photon,energy transmission,and angular momentum transfer to matter.The spatial structured beams are widely applied to optical micro-manipulation,communication,particle trapping and so on.Especially in the field of communication,researches and explorations of spatial structured optical communication technology are expected to solve the new demands on the communication capacity,modulation method and security,which from the development of applications such as big data,artificial intelligence,"all-optical interconnection" and the popularization of intelligent terminals.In the study of spatial structured optical communication,it can be divided into two major categories: mode modulation and mode multiplexing.The key to the application of mode modulation and mode multiplexing lies in the effective identification of modes.The detection of modes in the existing research work is mainly based on the interference method and the diffraction method.After the spatial structured beam passes through the specific optical components,it is based on the obvious features of the diffraction or interference pattern to distinguish spatially structured beams of different modes.With the increase of topological load,the mode recognition range and the accuracy rate are gradually reduced,because of the interference of atmospheric turbulence,the limitation of devices and lack of recognition ability.Only the limited spatial structured optical mode can be detected.At the same time,the spatial structured beam may emerge distortion and mode crosstalk caused by the influence of atmospheric turbulence in the transmission,which also severely limits the practical application of spatial structured optical communication technology.Therefore,the core difficulty in using spatial structured beam communication is mode identification and mode distortion compensation.It is of great physical significance to study mode recognition and distortion compensation.The excellent performance of deep learning in image recognition and task learning is providing us with potential solutions for mode recognition and mode compensation.As the core technology of artificial intelligence,deep learning has shown great advantages in image processing and natural language processing.In the aspect of regular mode recognition,the mode is identified by interference and diffraction patterns.The information itself needs to have strong regularity and is within the identifiable range of the human eye.While deep learning can extract and recognize more hidden information,which will further broaden our research ideas on OAM mode recognition.In addition,deep learning can fully utilize the learning ability in specific tasks to acquire the distortion phase in the beam directly from the light intensity distribution information after learning.Its fast,powerful computing ability will also make the fast and real-time mode compensation is available.Its powerful,fast computing power will also provide possibility for fast,real-time mode compensation.In this paper,we combine deep learning with spatial structured optical communication technology,and study spatial structured light field control technology,realize flexible structure light field generation and regulation,further study the field distribution and transmission characteristics of spatial structured light field.We reveal the laws of transmission distortion and mode crosstalk of the spatial structured beam and develop the spatial structured beam mode identification technology based on deep learning,explore the feasibility of real-time compensation for spatial structured beam transmission distortion based on deep learning.The research results will not only provide new ideas and methods for efficient demodulation/demultiplexing of spatial structured beam mode information,but also help to promote the actual integration of artificial intelligence and spatial structured optical communication,then provide key technical support for new spatial structured optical communication.The main achievements of this paper are:1)The study utilizes holographic phase diagram design to achieve the generation and flexible regulation of radial high-order perfect vortex.Conventional LG light has limitations in many applications due to the dependence between its topological charge and beam size.The perfect vortex solves the problem of LG light because the beam size is not related to the topological load.However,the regulation of the existing perfect vortex light is limited to the regulation of a single perfect vortex light,and the effective research has not been conducted in the radial order and the independent regulation of the superimposed perfect beam.In this paper,we experimentally realized the generation of arbitrary controllable high-order perfect vortex light and regulation of angular order,radial order,ellipticity and multi-beam superposition through flexible design of holographic phase diagram.2)A vector light field mode coded modulation communication based on cross polarization modulation is proposed.Vector light is generated by coherently combining with left-handed and right-handed circularly polarized vortex light.And the modulation of the vector light is realized by manipulating the OAM states of the two circularly vortex lights,so that the vector optical communication has a dual channel capacity.And the key information can be separately encoded into two circularly polarized vortex beams,which can effectively improve the confidentiality and security of the communication link.3)It is proposed to use the deep learning to flexibly identify the vortex beam mode and apply it to shift key communication.The existing OAM mode detection research of vortex beam mainly uses interference,diffraction and other methods to identify and detect OAM modes through interference or diffraction patterns,but the range of mode recognition that these methods can achieve is limited.It is not possible to achieve fast,real-time recognition of mode.At the same time,atmospheric turbulence poses a greater challenge to human eye recognition using traditional detection methods.The OAM mode identification technology based on a feedforward neural network algorithm and convolutional neural network algorithm proposed in this paper can effectively utilize the advantages of deep learning in big data processing for efficient and real-time OAM modes detection of vortex beams.Not only the OAM modes of vortex beams can be widely recognized,but it can also effectively resist the effects of atmospheric turbulence.4)It is proposed to use deep learning to extract and compensate the distortion information of the structured beam after transmission in turbulen atmosphere,and apply it to communication performance improvement research.At present,in the aspect of structural beam distortion compensation,the methods used mainly include adaptive optics and related iterative algorithms.The adaptive optics method has high complexity and cannot achieve single-time and real-time compensation.Other iterative algorithms cannot get accurate distortion phase.The mode distortion compensation technology based on convolutional neural network proposed in this thesis provides a new solution to the above problems.Using the powerful learning function of the neural network,the phase distortion information is extracted and compensated according to the collected light intensity distribution.Not only does it have fast extraction and compensation capabilities for the phase distortion information,but it is also expected to be suitable for different transmission environments.The information extraction technique of phase distortion in atmospheric turbulence based on convolutional neural network is developed and applied to study about the spatial structured optical communication,including suppression of phase noise and inter-channel mode crosstalk caused by atmospheric turbulence and improvement the anti-interference ability of current structured optical multiplex communication systems.