A CNN-based cost-effective modulation format identification scheme by low-bandwidth direct detecting and low rate sampling for elastic optical networks

Abstract Modulation format identification (MFI) is an indispensable technique in future optical performance monitoring (OPM) for elastic optical networks. The existing MFI scheme based on coherent detection combined with digital signal processing technology has good chromatic dispersion (CD) tolerance, but the high cost prevents its practical application at numerous intermediate nodes. By contrast, the existing direct detection-based MFI has the advantage of low cost but the CD tolerance is poor. In this paper a low-cost direct detection-based MFI scheme is proposed. The scheme not only reduces the cost of the MFI module by low-bandwidth direct detection and low rate sampling, but also ensures high MFI accuracy by extracting and learning features of asynchronous amplitude histograms (AAH) through convolutional neural network (CNN) as it has better accuracy and stability of identification compared with artificial neural network (ANN) and deep neural network (DNN) with the same computational complexity. A maximum correlation coefficient of AAH sequences of different modulation formats is proposed as a metric of discrimination for MFI, and its relationship with the bandwidth of MFI module is investigated in the scenario that the MFI module relative bandwidth changes from 0.00025 to 0.75 of service signal baud rate (SBR) and the CD vary from 0 ∼ 16000ps/nm. The simulation and experimental results show that the MFI module with 0.002 ∼ 0.02 SBR bandwidth could realize 100% MFI accuracy of three common QAM signals with the CD of 0 ∼ 16000ps/nm.