Research On Frequency Offset Estimation And Deep Learning Algorithm For Coherent Optical FBMC System

In recent years,with the widespread use of mobile Internet,big data and other new services,people’s demand for communication capacity and information transmission rate is increasing.Various new technologies and services are emerging to make fiber optic communication networks face unprecedented challenges.To cope with the increasing demand for network bandwidth,Filter Bank Multicarrier with Offset Quadrature Amplitude Modulation(FBMCOQAM)technology has been widely used in fiber optic communication.Compared with Orthogonal Frequency Division Multiplexing(OFDM),FBMC-OQAM technology has the advantages of low out-of-band spectrum leakage,excellent time-frequency focusing characteristics and high spectral efficiency.However,as a typical multi-carrier modulation technique,FBMC-OQAM is also susceptible to carrier frequency offset,which can lead to significant degradation of the system transmission performance.At the same time,dispersion,attenuation and nonlinear effects within the fiber channel can also cause optical signal to be distorted.In summary,we need frequency offset compensation and channel equalization for the received signal at the FBMC-OQAM receiver.Focusing on the signal impairment problem of coherent optical FBMC-OQAM system(CO-FBMC-OQAM),this paper investigates the carrier frequency offset estimation method and the channel equalization method with deep learning,respectively.In terms of frequency offset estimation,the range of estimation of normalized fractional frequency offset by the existing algorithms is generally in the range of(-0.5,+0.5).When the signal is affected by large frequency deviation,it leads to inaccurate frequency offset estimation.To address this problem,a frequency offset compensation scheme based on a cascaded crosscorrelation function is designed in this paper.The solution first redesigns the FBMC-OQAM frame structure by inserting an auxiliary pilot into the transmitted data,and the expression for the frequency offset estimation is obtained at the receiver side by calculating the cascaded crosscorrelation function of the real-valued pilot.It can extend the estimation range of normalized fractional frequency offset to(-1,+1),and then continue to track the residual frequency offset to further improve the estimation accuracy of the frequency offset.In terms of channel equalization,traditional algorithms need to model complex fiber channels and then use complicated numerical models and approximations to equalize the channel impairments,resulting in high system complexity and unsatisfactory performance of channel equalization.To address this problem,this paper proposes a deep learning-based channel equalization scheme for CO-FBMC-OQAM systems using artificial neural networks instead of traditional channel equalization methods.In this scheme,the channel equalization performance of deep neural network,convolutional neural network,recurrent neural network and long short-term memory networks are tested separately and compared with the traditional channel equalization methods.The simulation results illustrate that the proposed scheme can solve the channel equalization problem for CO-FBMC-OQAM systems and has a stronger BER advantage.