Theoretical And Experimental Study On Neural Network-based Wavefront Correction For FSO Systems

In recent years,Optical Wireless Communication(OWC)has been widely concerned by academia and industry because of its ultra wide spectrum and good security.The outdoor application scenario of OWC mainly refers to Free Space Optical communication(FSO).The FSO can be used in many outdoor scenarios,including an effective solution to the "Last Kilometer" problem,emergency communication solution,base station backhaul and satellite communication.However,in practice,the transmission performance in FSO systems is mainly affected by the atmospheric turbulence,which causes wavefront aberration of the received light beam.Adaptive optics(AO)is the most commonly used technique to overcome the turbulence-induced wavefront aberration.However,there are still many problems to be solved in the AO technique.For example,the Wavefront Sensor(WFS)cannot work in the strong turbulence,and the WFS-less AO system based on blind optimization algorithm needs many iterations.In order to solve these problems,this dissertation focuses on design of a wavefront aberration correction system based on Convolution Neural Network(CNN),which can realize the fast correction of wavefront aberration in FSO communication systems.Firstly,the proposed CNN wavefront correction scheme is modeled.We simulate and analyze the performance of CNN based wavefront correction system,the feasibility of CNN method is verified by the correction results under strong turbulence.In order to furter banlance CNN wavefront correction performance and system complexity,the key factors of CNN based wavefront correction method are analyzed based on the theoretical analysis and simulation performance,including the number of Zernike polynomials,quantization bits and different CNN network structures.Then the CNN method is compared with the stochastic parallel gradient descent(SPGD)and simulated annealing(SA)from two aspects of correction performance and time complexity.Simulation results show that as the turbulence strength increases,the performance of SPGD and SA algorithms decrease rapidly,while the performance with the CNN method is stable and the power penalty with the CNN is less than SPGD and SA algorithms.Especially in strong turbulence,CNN performance is better than other algorithms.In addition,by analyzing the time required for different correction methods to compensate the wavefront aberration,it is found that the time required by CNN method is far less than that of SPGD and SA algorithms.Therefore,the CNN based wavefront correction method can improve the performance of aberration correction and greatly reduce the time required for correction,so as to meet the needs of rapid channel change.The CNN-based AO system for FSO application is experimentally investigated in the indoor lab and the performance of the CNN based wavefront correction method is studied.Experimental results show that the coupling power loss can be well reduced after CNN correction under both weak and strong turbulence.The average power loss reduces to 1.8(0.8)dB from 12.4(4.4)dB in the strong(weak)turbulence.Therefore,the feasibility of CNN based wavefront correction method and its applicability to different turbulence situations are verified by experiments.Finally,in order to further analyze the performance of CNN wavefront correction,the coupling efficiency distribution of the correction system based on CNN method before and after turbulence correction is theoretically analyzed.The research results show that the coupling efficiency before and after CNN correction obeys Rice distribution,which is verified in experiment.