Vol. 53, Issue 2, pp. 281-289

Vol. 53 Issue 2 pp. 281-289

Identification of advanced optical modulation format and estimation of signal-to-noise-ratio based on parallel-twin convolutional neural network

Xiaowei Dong, Zhihui Yu


deep learning, PT-CNN, constellation diagram, modulation format identification, SNR estimation


In this paper, we design a parallel-twin convolutional neural network (PT-CNN) deep learning model and use the signal constellation diagram to realize the identification of six advanced optical modulation formats (QPSK, 4QAM, 8PSK, 8QAM, 16PSK, 16QAM) and signal-to-noise-ratio (SNR) estimation. The influence of PT-CNN with different layers and kernel sizes is investigated and the optimal network model is chosen. Simulation results demonstrate that the proposed method has the advantages of not requiring manual feature extraction, having the ability to clearly distinguish the six modulation formats with 100% accuracy when SNR of the received signal sequences is higher than 12 dB. In addition, the high-accurate SNR estimation is realized simultaneously without increasing additional system complexity.

Vol. 53
Issue 2
pp. 281-289

0.4 MB
OPTICA APPLICATA - a quarterly of the Wrocław University of Science and Technology, Faculty of Fundamental Problems of Technology