Long Short-Term Memory Neural Network(LSTM-NN) Enabled Optical Performance Monitoring

In order to satisfy the ever-growing communication need for modern society,the next-generation fiber optical communication systems are drastically evolved towards high speed,large capacity,dynamic and flexible operation.In recent years,digital coherent communication technology has become the primary choice for next-generation fiber optical communication systems.Real-time parameter monitoring which can reflect the quality of optical signals is indispensable for reliable fiber optical transmission.Artificial neural networks(ANNs)are expected to improve the artificial intelligence of optical performance monitoring.However,the potential of ANNs for optical performance monitoring has not been fully exploited in terms of monitoring accuracy and intelligent operation.Therefore,this thesis has been focused on the research of long short-term memory neural network(LSTM-NN)based optical performance monitoring for digital coherent fiber optical transmission systems.The LSTM-NN enabled optical performance monitoring technique has the capability of high accuracy and smart response.The main research outcomes of this thesis are summarized as follows:(1)In order to monitor the transmission impairments arising in optical signals,LSTM-NN has been employed to automatically extract the features from the time-varied data.The operation principle has been explained and the performance comparisons with the existing optical performance monitoring techniques have been carried out.(2)The LSTM-NN enabled optical signal-to-noise ratio(OSNR)monitoring scheme has been proposed in this thesis.LSTM-NN is able to recognize the inherent OSNR feature from the received XI,XQ,YI,YQ data after the optical coherent detection.The transparent OSNR monitoring with respect to both modulation format and baud-rate has been numerically verified,together with its robust operation to chromatic dispersion(CD)and polarization mode dispersion(PMD).Finally,experimental verification of 34.94 GBaud PDM-16 QAM signal under the scenario of back-to-back transmission has been reported.The mean absolute error(MAE)of OSNR monitoring is 0.05 d B,and the response time is 0.5 ms.In comparison with traditional OSNR monitoring schemes,the proposed LSTM-NN based OSNR monitoring scheme is more accurate and intelligent.(3)An LSTM-NN enabled joint OSNR and CD monitoring scheme has been proposed for realizing simultaneous multi-parameter optical performance monitoring.Experimental verification of 5/10 GBaud PDM-16QAM/64 QAM signals transmission over 80?120 km standard single-mode fiber(SSMF)has been reported.The MAEs of simultaneous OSNR and CD monitoring are below 0.12 d B and 1.09 ps/nm,respectively,and the response time is0.64 ms.The experimental results not only verify that the proposed scheme is insensitive to both modulation format and baud-rate,but also proof that the proposed joint monitoring scheme can operate independently without the cross-sensitivity.