We propose a fast and accurate signal quality monitoring scheme that uses convolutional neural networks for error vector magnitude (EVM) estimation in coherent optical communications. We build a regression model to extract EVM information from complex signal constellation diagrams using a small number of received symbols. For the additive-white-Gaussian-noise-impaired channel, the proposed EVM estimation scheme shows a normalized mean absolute estimation error of 3.7% for quadrature phase-shift keying, 2.2% for 16-ary quadrature amplitude modulation (16QAM), and 1.1% for 64QAM signals, requiring only 100 symbols per constellation cluster in each observation period. Therefore, it can be used as a low-complexity alternative to conventional biterror-rate estimation, enabling solutions for intelligent optical performance monitoring.

We exploit deep supervised learning and amplitude histograms of coherent optical signals captured before carrier phase recovery (CPR) to perform time-sensitive and accurate error vector magnitude (EVM) estimation for 32 Gbaud mQAM signal monitoring purposes. 

Error vector magnitude (EVM) has proven to be one of the optical performance monitoring metrics providing the quantitative estimation of error statistics. However, the EVM estimation efficiency has not been fully exploited in terms of complexity and energy consumption. Therefore, in this paper, we explore two deep-learning-based EVM estimation schemes. The first scheme exploits convolutional neural networks (CNNs) to extract EVM information from images of the constellation diagram in the in-phase/quadrature (IQ) complex plane or amplitude histograms (AHs). The second scheme relies on feedforward neural networks (FFNNs) extracting features from a vectorized representation of AHs. In both cases, we use short sequences of 32 Gbaud m-ary quadrature amplitude modulation (mQAM) signals captured before or after a carrier phase recovery. The impacts of the sequence length, neural network structure, and data set representation on the EVM estimation accuracy as well as the model training time are thoroughly studied. Furthermore, we validate the performance of the proposed schemes using the experimental implementation of 28 Gbaud 64QAM signals. We achieve a mean absolute estimation error below 0.15%, with short signals consisting of only 100 symbols per IQ cluster. Considering the estimation accuracy, the implementation complexity, and the potential energy savings, the proposed CNN- and FFNN-based schemes can be used to perform time-sensitive and accurate EVM estimation for mQAM signal quality monitoring purposes.

Robustness against the large linewidth semiconductor laser-induced impairments in coherent systems is experimentally demonstrated for a feedforward neural network-enabled EVM estimation scheme. A mean error of 0.4% is achieved for 28 Gbaud square and circular QAM signals and linewidths up to 12.3 MHz. 

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