Deep Neural Network Based Chromatic Dispersion Estimation With Ultra-Low Sampling Rate for Optical Fiber Communication Systems

Fast and low-cost chromatic dispersion (CD) estimation is very important for fiber link that is dynamically reconfigurable in the next-generation optical networks. In this paper, a novel CD estimation method based on the deep neural network (DNN) with ultra-low sampling rate is proposed for the optical fiber transmission systems. The method can estimate CD for under-sampled signals by the trained DNN with a weighted average. To demonstrate the feasibility of the method, the simulation of a 28-GBaud five-channel optical fiber transmission system with 100-2000-km SSMF transmission is carried, where the sampling rate of the analog-to-digital converter (ADC) at the receiver is only 500 MHz. The results show that the maximum mean absolute error (MAE) of the estimated CD is less than 75 ps/nm with the reference cumulative CD from 1600 to 32000 ps/nm for QPSK and 16QAM signals. Meanwhile, the robustness of the proposed method is verified for amplifier spontaneous emission (ASE) and nonlinear (NL) noise. Furthermore, we experimentally demonstrated that the maximum MAE is less than 75 ps/nm when the transmission distance varies from 100 to 800 km in 20-GBaud QPSK optical fiber transmission system. In conclusion, the proposed DNN-based CD estimation method shows great potential for the cost-effective under-sampled systems.