Optimization Research Of Digital Back-propagation Algorithm Based On Deep Learning In Coherent Optical Communication Systems

With the booming development of mobile Internet and digital multimedia applications,the continuous high growth of business traffic transmission demand has brought huge capacity expansion pressure on fiber optic transmission systems.The coherent fiber optic communication system,as a reliable transmission solution with large channel capacity,has been widely used in the field of long-distance fiber optic communication now.For long-haul coherent fiber optic communication systems,the linear impairment of signal can be equalized by sophisticated digital signal processing algorithms,while the nonlinear impairment becomes the ultimate obstacle limiting the further upgrade of high-capacity long-range fiber optic communication systems.Among several schemes to equalize the nonlinear impairment in optical fiber,Digital Back-Propagation(DBP)has been proven to be an effective method to equalize the impact of nonlinear effects.The DBP algorithm equalization of fiber nonlinearity by solving the numerical solution of the nonlinear S chrodinger equation using split step Fourier method.However,the DBP algorithm requires a high computational complexity to ensure its performance,which makes it difficult to apply in practical systems.The development of artificial intelligence technology in recent years has brought a new solution idea for the equalization of fiber nonlinear damage.The deep learning-based Learned Digital Back-Propagation(LDBP)algorithm optimizes the parameters in DBP by mapping the iterative steps of DBP algorithm to the hidden layers of Deep Neural Networks(DNN)and using the gradient descent algorithm of DNN to optimize the parameters in DBP,which greatly improves the nonlinear equilibrium performance.However,as a new nonlinear equalization algorithm,LDBP still has several internal optimization spaces in terms of complexity and performance.Therefore,this paper will focus on the optimization scheme of LDBP-based nonlinear equilibrium algorithm in coherent fiber optic communication systems to conduct an in-depth study.The main research contents and innovation points are as follows:(1)To address the problem of high complexity of dispersion equalization algorithm in LDBP,a low-complexity dispersion equalization scheme with a mixture of time and frequency domains is proposed by optimizing both time domain filter taps and frequency domain dispersion coefficients using DNN.The performance of the proposed scheme is verified in a coherent fiber optic communication simulation system with a transmission distance of 32×100 km with 20 GBaud dual polarization 16-QAM.The results show that,compared with the dispersion equalization scheme in traditional LDBP algorithm,the lowcomplexity dispersion equalization scheme based on the mixture of time and frequency domains proposed in this paper reduces the computational complexity by 23%under the condition that the optimal performance is basically the same.(2)For the problem of high complexity of phase noise equalization algorithm in LDBP algorithm,a low-complexity phase noise equalization scheme in LDBP algorithm is proposed by using the training sequence to assist in estimating the signal phase noise during the LDBP training process.The phase noise equalization scheme proposed in this paper is investigated in a coherent fiber optic communication simulation system with 20 GBaud dual-polarization 16QAM over a transmission distance of 32×100 km.The results show that the proposed scheme reduces the computational complexity by about 43.3%compared with the phase noise equalization scheme used in the conventional LDBP algorithm,and does not significantly degrade the performance of LDBP algorithm.(3)Based on the characteristics of LDBP algorithm and perturbation technique,LDBP algorithm and perturbation technique are combined.By using DNN to optimally adjust the perturbation coefficients and the position of nonlinear equalization,an optimized perturbation-based LDBP algorithm is proposed.The performance of the proposed algorithm is verified by building a coherent fiber optic communication simulation system with 90 GBaud dual-polarization 16QAM over a transmission distance of 10×100 km.The results show that the final optimized LDBP algorithm has a 1.56 dB gain in the optimal Effective SNR and a 2 dB improvement in the optimal launch fiber power compared with the conventional LDBP algorithm.In summary,this paper proposes innovative improvement methods for the dispersion equalization scheme and the phase noise equalization scheme in the current LDBP algorithm,which effectively reduces the complexity of LDBP algorithm;meanwhile,the structure of the LDBP algorithm is optimized by combining perturbation technique,which effectively improves the nonlinear equalization performance of LDBP algorithm.Verified by simulation,the optimization scheme proposed in this paper is feasible and effective for LDBP algorithm in terms of complexity and performance,which provides an innovative idea for the application research of LDBP algorithm in long-haul coherent optical communication.