CNN based OSNR estimation method for long haul optical fiber communication systems

Abstract

The optical signal-to-noise ratio (OSNR) is an critical factor in evaluating the performance of high speed optical fiber communication systems. In this work, we propose a convolutional neural network (CNN) and constellation diagrams based method to simultaneously estimate OSNR. In the training step, CNN extracts the essential features of the input constellation diagrams. Then, with the built model in the training step, the CNN outputs the OSNR of the signal under test. The simulation by VPI software is carried on a 9-channel long haul optical transmission system with the launched optical power of −3.0~+3.0dBm per channel and transmission distance up to 1000 km. The accuracy of OSNR estimation is almost 100%. The results show that the maximum test error of OSNR is less than 1.0 dB with the reference OSNR varied in the range of 15~30 dB (or 15.5~29.5 dB) for QPSK, 8PSK, 16QAM, and 20~35 dB (or 20.5~34.5) for 64QAM signal.