Multi-channel multi-task optical performance monitoring based multi-input multi-output deep learning and transfer learning for SDM

Abstract We propose a transfer learning (TL) simplified multi-input multi-output multi-task convolutional neural network (MIMO MT-CNN) to realize the optical performance monitoring (OPM) for space division multiplexing (SDM) fiber transmission systems. TL simplified MIMO MT-CNN is used for NRZ, RZ and PAM4 signal joint monitoring including modulation format identification (MFI) and optical signal-to-noise ratio (OSNR) estimation. Thanks to the TL technologies, the parameters used in single mode fiber transmission is transferred to SDM to reduce the required training data and epochs. The experiment has been demonstrated with MIMO MT-CNN for seven cores fiber systems. The results show that the accuracy of MFI can reach 100% for all signals. For NRZ, RZ, and PAM4, the root-mean-square errors (RMSE) of OSNR estimation are less than 0.6 dB, respectively. In addition, by implementing TL from simulation to experiment, the required training samples and epochs are reduced by 48% and 50%, respectively. Due to its low cost and being easy to implement, the proposed OPM solution has the potential for next-generation SDM fiber transmission systems. The model can realize short training time and multi-task multi-channel OPM for high-capacity SDM elastic optical networks, simultaneously.