Feedforward Neural Network-Based EVM Estimation: Impairment Tolerance in Coherent Optical Systems

Error vector magnitude (EVM) is commonly used for evaluating the quality of m-ary quadrature amplitude modulation (mQAM) signals. Recently proposed deep learning techniques EVM estimation extend functionality conventional optical performance monitoring (OPM). In this article, we evaluate tolerance our developed scheme against various impairments in coherent systems. particular, analyze signal capabilities presence residual in-phase/quadrature (IQ) imbalance, fiber nonlinearity, and laser phase noise. We use feedforward neural networks (FFNNs) to extract information from histograms 100 symbols per IQ cluster sequence captured before carrier recovery. perform simulations considered impairments, along with an experimental investigation impact To investigate each impairment type, compare accuracy three training methods: 1) without impairment, 2) one model all 3) independent impairment. Results indicate a good generalization scheme, thus providing valuable reference developing next-generation intelligent OPM