Search

In recent years, many new modulation and multiplexing technologies and dynamic network concepts have been proposed to adapt to the ever-increasing network bandwidth and quality requirements. Network control platform has a systematic and intelli-gent development trend, which requires network managers to constantly monitor the parameters of the network and optimize the network state. However, it is not feasible to arrange additional monitoring equipment in a large range to obtain parameter informa-tion from the perspective of cost control. It is better to use known data and special algorithms to monitor and optimize network per-formance. Machine learning methods are increasingly adopted by the academic community because they are accurate and efficient enough to accomplish these tasks. This paper first reviews the different application scenarios of machine learning algorithms in op-tical network monitoring and optimization tasks. Then it reviews the research achievements in this field, and puts forward the ex-isting problems of machine learning-based optical network monitoring and optimization methods as well as the possible direction of future research. The optical performance monitoring based on machine learning includes failure identification, quality of transmis-sion estimation and channel power prediction. The network configuration optimization method based on machine learning includes reinforcement learning to optimize channel power. For future research direction, we believe that it is possible for researchers to use real data from network operators, newly collected data to dynamically train the model, and transfer learning and data enhancement techniques to ensure the robustness and generalization ability of the algorithm.

Recently, extensive research has been conducted to explore the utilization of machine learning (ML) algorithms in various direct-detected and (self)-coherent short-reach communication applications. These applications encompass a wide range of tasks, including bandwidth request prediction, signal quality monitoring, fault detection, traffic prediction, and digital signal processing (DSP)-based equalization. As a versatile approach, ML demonstrates the ability to address stochastic phenomena in optical systems networks where deterministic methods may fall short. However, when it comes to DSP equalization algorithms such as feed-forward/decision-feedback equalizers (FFEs/DFEs) and Volterra-based nonlinear equalizers, their performance improvements are often marginal, and their complexity is prohibitively high, especially in cost-sensitive short-reach communications scenarios such as passive optical networks (PONs). Time-series ML models offer distinct advantages over frequency-domain models in specific contexts. They excel in capturing temporal dependencies, handling irregular or nonlinear patterns effectively, and accommodating variable time intervals. Within this survey, we outline the application of ML techniques in short-reach communications, specifically emphasizing their utilization in high-bandwidth demanding PONs. We introduce a novel taxonomy for time-series methods employed in ML signal processing, providing a structured classification framework. Our taxonomy categorizes current time-series methods into four distinct groups: traditional methods, Fourier convolution-based methods, transformer-based models, and time-series convolutional networks. Finally, we highlight prospective research directions within this rapidly evolving field and outline specific solutions to mitigate the complexity associated with hardware implementations. We aim to pave the way for more practical and efficient deployment of ML approaches in short-reach optical communication systems by addressing complexity concerns. [ABSTRACT FROM AUTHOR]

The article presents the results of a method based on asynchronous delay-tap sampling (ADTS) and convolutional neural network (CNN) for determining simultaneously occurring disturbances described using the chromatic dispersion (CD), crosstalk and optical signal-tonoise ratio (OSNR) parameters. The ADTS method was used to generate training and test data for the convolutional network, which in turn was used to learn to recognize interference from said data. The tests were carried out for a transmission speed of 10 Gbit/s and for onoff keying (OOK) and differential phase shift keying (DPSK) modulation. Very good results were obtained in recognizing simultaneously occurring phenomena. Accuracy of over 99% was achieved for CD and crosstalk for DPSK modulation and over 98% for OOK modulation. In the case of amplified spontaneous emission (ASE) noise, slightly weaker results were obtained, above 95-96% for both modulations. Based on the conducted research, it was determined that the use of ADTS and CNN methods enables monitoring of simultaneously occurring CD, crosstalk, and ASE noise in the physical layer of the optical network, while maintaining the requirements for modern monitoring systems. [ABSTRACT FROM AUTHOR]

Copyright of Study on Optical Communications / Guangtongxin Yanjiu is the property of Study on Optical Communications Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

In this paper, a technique for optical performance monitoring (OPM) using deep learning-based artificial neural network (ANN) is implemented. ANN is trained with parameters derived from eye-diagram for the identification of optical signal to noise ratio (OSNR), chromatic dispersion (CD) and polarisation mode dispersion (PMD) simultaneously and independently in a 10 Gb/s system with non-return-to-zero (NRZ) on-off keying (OOK) data signal. ANN-based OPM confirms that the proposed approach can provide reliable estimated results. The mean squared errors for OSNR, CD and differential group delay (DGD) are found to be 4.6071 dB, 0.0417 ps/nm/km and 0.0016 ps/km, respectively. The proposed technique may be utilized in analyzing the signals of future heterogeneous optical communication networks intelligently. [ABSTRACT FROM AUTHOR]

Recently, extensive research has been conducted to explore the utilization of machine learning (ML) algorithms in various direct-detected and (self)-coherent short-reach communication applications. These applications encompass a wide range of tasks, including bandwidth request prediction, signal quality monitoring, fault detection, traffic prediction, and digital signal processing (DSP)-based equalization. As a versatile approach, ML demonstrates the ability to address stochastic phenomena in optical systems networks where deterministic methods may fall short. However, when it comes to DSP equalization algorithms such as feed-forward/decision-feedback equalizers (FFEs/DFEs) and Volterra-based nonlinear equalizers, their performance improvements are often marginal, and their complexity is prohibitively high, especially in cost-sensitive short-reach communications scenarios such as passive optical networks (PONs). Time-series ML models offer distinct advantages over frequency-domain models in specific contexts. They excel in capturing temporal dependencies, handling irregular or nonlinear patterns effectively, and accommodating variable time intervals. Within this survey, we outline the application of ML techniques in short-reach communications, specifically emphasizing their utilization in high-bandwidth demanding PONs. We introduce a novel taxonomy for time-series methods employed in ML signal processing, providing a structured classification framework. Our taxonomy categorizes current time-series methods into four distinct groups: traditional methods, Fourier convolution-based methods, transformer-based models, and time-series convolutional networks. Finally, we highlight prospective research directions within this rapidly evolving field and outline specific solutions to mitigate the complexity associated with hardware implementations. We aim to pave the way for more practical and efficient deployment of ML approaches in short-reach optical communication systems by addressing complexity concerns.

Direct Detection (DD) optical performance monitoring (OPM), Modulation Format Identification (MFI), and Baud Rate Identification (BRI) are envisioned as crucial components of future-generation optical networks. They bring to optical nodes and receivers a form of adaptability and intelligent control that are not available in legacy networks. Both are critical to managing the increasing data demands and data diversity in modern and future communication networks (e.g., 5G and 6G), for which optical networks are the backbone. Machine learning (ML) has been playing a growing role in enabling the sought-after adaptability and intelligent control, and thus, many OPM, MFI, and BRI solutions are being developed with ML algorithms at their core. This paper presents a comprehensive survey of the available ML-based solutions for OPM, MFI, and BFI in non-coherent optical networks. The survey is conducted from a machine learning perspective with an eye on the following aspects: (i) what machine learning paradigms have been followed; (ii) what learning algorithms are used to develop DD solutions; and (iii) what types of DD monitoring tasks have been commonly defined and addressed. The paper surveys the most widely used features and ML-based solutions that have been considered in DD optical communication systems. This results in a few observations, insights, and lessons. It highlights some issues regarding the ML development procedure, the dataset construction and training process, and the solution benchmarking dataset. Based on those observations, the paper shares a few insights and lessons that could help guide future research. [ABSTRACT FROM AUTHOR]

In advanced high-speed fiber optic communication systems, due to the introduction of dense wavelength division multiplexing technology, the signal spectral interval is getting narrower and narrower, and the traditional out-of-band Optical Signal-to-Noise Ratio (OSNR) monitoring technology is no longer accurate. Therefore, further study is required in the low-cost in-band OSNR monitoring scheme. A Deep Neural Network (DNN) link OSNR monitoring scheme for Intensity-Modulation and Direct Detection (IMDD) system is proposed. We used a 5-layer DNN trained from 550 000 datasets to successfully estimate the OSNR of the 2 GBaud On-Off Key (OOK) signal in the range of 5 to 15 dB, and the Mean Absolute Error (MAE) is less than 0.8 dB.

The exponentially growing demand for high-speed data necessitates more complex and versatile networks. Optimization and reliability assurance of such high-complexity networks is getting increasingly important. In this article, we experimentally validate our a machine learning-based framework that combines quality of transmission (QoT) estimation with soft-failure detection, identification, and localization based on the same latent space of a variational autoencoder running on optical spectra obtained by optical spectrum analyzers at high priority nodes in the network. We further investigate the advantages of a variational autoencoder-based soft-failure detection mechanism over a QoT metric-based approach. We use data acquired from optical transmission experiments involving different modulation formats and channel configurations. The results demonstrate that the proposed framework achieves reliable QoT estimation in real world scenarios. Additionally, it effectively detects soft-failures, identifies specific failure types and accurately localizes the occurrence of failures.

In this paper, a joint OSNR and nonlinear noise power estimation scheme based on multi-task deep neural network (MT-DNN) is proposed with the advantages of dispersion-insensitive, modulation-format-transparent for high-speed, long-haul, multi-channel fiber-optic communication systems. Amplitude histograms (AHs) are generated by processing the spectrums collected with different OSNR, launch power and transmission distance by an offline spectrum preprocessing flow. The MT-DNN can automatically learn the features of the AHs to achieve OSNR and nonlinear noise power estimation, simultaneously. For 4-quadrature amplitude modulation (4QAM), 16QAM and 64QAM signals under different transmission conditions, the average MAE and RMSE are calculated for the OSNR estimate and the nonlinear noise power estimate, which are both less than 1 dB. Moreover, the resistance of OSNR estimation to amplified spontaneous emission (ASE) noise and nonlinearity, and the tolerance of nonlinear noise estimation to launch power and transmission distance are investigated, respectively. The results demonstrate that the joint OSNR and nonlinear noise power estimation scheme is insensitive to dispersion, transparent to modulation format, and has high accuracy and high tolerance. This research provides a research reference value for future optical performance monitoring of coherent optical communication systems.

Fiber performance monitoring using modern online technologies in the next generation of intelligent optical networks allows for identifying the source of the degeneration and putting in protective steps to increase remote optical network stability & reliability. In this paper, the performance of the fiber performance monitoring tool (FPMT) technique was improved by integrating it with optical amplifier boards. In this regard, the improved technique detects optical layer events and all fiber soft and hard failures at the online remote rather than disrupting the data flow with a measurement accuracy for defect location of up to ~99.9%, small tolerance of up to ~1 m, the longest distance to detecting optical line defects of up to ~300km, and enhanced power budget for the system with optimum insertion-loss of up to ~0.0 dB. The proposed integration method provides better results with an excellent and efficient solution at fault location measurement & detection in real-time with good financial implications of the technique. The competitiveness of the improved technique over the actual optical networks has been successfully confirmed through application to Huawei labs infrastructure nodes and displayed experimental simulation results. [ABSTRACT FROM AUTHOR]

• In our proposed few-shot multi-task metric learning(MML) algorithm for optical performance monitoring, modulation format identification(MFI) is treated as a classification task, while optical signal-to-noise ratio (OSNR) estimation is handled as a regression task. • Using a 16-way-6-shot dataset, the algorithm achieved 100% accuracy in identifying six modulation formats after 200 epochs, with an OSNR mean squared error (MSE) of only 0.431 dB. • After ablation experiments, as fiber transmission distance increased, the OSNR MSE remained below 0.8 across distances ranging from 200 km to 800 km. This suggests that the integrated metric function provides quicker adaptability to the model and demonstrates exceptional robustness. • In the context of enhancing intelligence and adaptability, this capability is crucial for optical performance monitoring (OPM) equipment that needs to adapt to new scenarios and tasks. In our experiments, applying few shot metric learning for optical performance monitoring (OPM), we set the dataset as 16-way-6-shot. Modulation format identification (MFI) was utilized as a classification task, and optical signal-to-noise ratio (OSNR) estimation was used as a regression task for joint analysis. Multi-task metric learning (MML) used the adaptive weights to balance the weights of the three metric functions, six modulation formats (QPSK, 8QAM, 16QAM, 32QAM, 64QAM, 128QAM) are classified correctly with 100 % accuracy after 200 epochs. Furthermore, the lowest mean square error (MSE) of OSNR is 0.431 dB. Then, Ablation experiments compute the corresponding similarity (SIM) for each metric function show that the MSE of MML, SIM Local+Cosine , SIM Cosine+Point , SIM Local+Point , single-task metric learning (SML) and adaptive multi-task learning (AMTL) is 0.431 dB, 0.572 dB, 0.569 dB, 0.567 dB, 0.637 dB, 1.319 dB, respectively. The proposed model achieves the highest accuracy in MFI and the lowest MSE in OSNR estimation. Finally, when comparing the various metric functions while altering the transmission distance of the optical fiber, it was observed that MML stayed within an acceptable range between 200 km and 800 km. This shows that our algorithm requires only a small number of training samples to create a reasonably good model, offering a new approach to solving problems that arise in optical performance monitoring. [ABSTRACT FROM AUTHOR]

In the optical communication systems, the optical spectrum (OS) provides useful informations for optical performance monitoring and optical link diagnosis. In this paper, we investigate the neural-network based OS analysis (OSA) techniques able to simultaneously recognize the OS and estimate the optical signal to noise ratio (OSNR), cascaded filtering distortion (CFD) and carrier wavelength drift (CWD). We demonstrate that, compared with the multi-task artificial neural network (MT-ANN) and convolutional neural network (MT-CNN), the proposed multi-task cascaded ANNs (CANN) and cascaded CNNs (CCNN) can greatly improve the OSA performance and accelerate the training process by exploiting specific features and loss functions for different tasks. The CCNN able to deal with the high-resolution OS (HOS) can further improve the OSA performance compared with the CANN. The averaged OS recognition accuracy and OSNR, CFD and CWD estimation errors obtained with the CCNN (CANN) for the 15 kinds of optical signals are 99.32% (97.07%), 0.36 (0.55) dB, 0.24 (0.32) and 0.04 (0.06) GHz, respectively, even when the various OS distortions are present. The proposed CANN and CCNN with a better versatility, higher OSA accuracy and faster convergence speed are promising enabling techniques for the future intelligent OSA systems.

Error vector magnitude (EVM) is a metric for assessing the quality of m-ary quadrature amplitude modulation (mQAM) signals. Recently proposed deep learning techniques, e.g., feedforward neural networks (FFNNs) -based EVM estimation scheme leverage fast signal quality monitoring in coherent optical communication systems. Such a scheme estimates EVM from amplitude histograms (AHs) of short signal sequences captured before carrier phase recovery (CPR). In this work, we explore further complexity reduction by proposing a simple linear regression (LR) -based EVM monitoring method. We systematically compare the performance of the proposed method with the FFNN-based scheme and demonstrate its capability to infer EVM from an AH when the modulation format information is known in advance. We perform both simulation and experiment to show that the LR-based EVM estimation method achieves a comparable accuracy as the FFNN-based scheme. The technique can be embedded with modulation format identification modules to provide comprehensive signal information. Therefore, this work paves the way to design a fast-learning scheme with parsimony as a future intelligent OPM enabler.

We propose a simultaneous optical signal recognition (OSR) and optical signal-to-noise ratio (OSNR) monitoring method by using multi-task learning convolutional neural network (MTL-CNN) in conjunction with optical spectrums, which enables optical performance monitoring (OPM) in the optical transmission network. In order to achieve a trade-off between monitoring loss and time consumption of the MTL-CNN constructed for seven commonly used signals with a spectrum resolution of 10 pm, the number of feature map channels in four convolutional layers is delicately designed as 32, 32, 64, and 64 with task weights set to 0.4 and 0.2, corresponding to OSNR monitoring and OSR, respectively. Simulation results manifest that this method can realize the recognition of the received optical signals with an overall accuracy of 100% and also enable OSNR monitoring with the mean absolute error (MAE) of 0.262 dB. The proposed method shows strong robustness to various distortions and exhibits good performance in terms of time consumption. The effectiveness is further verified by proof-of-concept experiments in three signals. These illustrate that our method is a promising solution for multi-parameter monitoring with high accuracy and efficiency.

Weakly-coupled mode-division multiplexing (MDM) technique based on linearly-polarized (LP) modes in ultralow-modal-crosstalk few-mode fiber (FMF) has been considered as a promising candidate for capacity enhancement of next-generation optical transmission systems. Similar to polarization mode dispersion in single-mode fibers (SMFs), signals in weakly-coupled FMFs suffer from the impacts of intra-LP-mode dispersion (ILMD) for both degenerate and non-degenerate LP modes.In this paper, we propose an effective method for ILMD monitoring in weakly-coupled MDM systems. The ILMD evaluation method for degenerate LP modes is designed by the analysis of channel transfer matrix in a digital coherent receiver. Then the ILMD monitoring scheme is experimentally validated for all the 6 LP mods in a weakly-coupled FMF. The influences of chromatic dispersion, optical signal to noise ratio and inter-LP-mode crosstalk to the method are also investigated.

A new technique for fiber faults events detection and monitoring in optical communication network systems is proposed. The fiber performance monitoring tool is a new proposed technique designed to detect, locate, and estimate the fiber faults without interrupting the data flow with efficient costs and to improve the availability and reliability of optical networks as it detects fiber faults remotely in real time. Instead of the traditional old method, the new proposed FPMT uses an optical time domain reflectometer to detect multiple types of fiber failures, e.g., fiber breaks, fiber end face contamination, fiber end face burning, large insertion losses on the connector and interconnection, or mismatches between two different types of fiber cables. The proposed technique methodology to detect the fiber failures depends on analyzing the feedback of the reflected signal and the pattern shape of the reflected signal over network fiber lines, supports a higher range of distance testing and performance monitoring, and can be performed inside an optical network in real time and remotely by integrating with an OSC board. The proposed technique detects fiber faults with an average accuracy of measurement up to 99.8%, the maximum distance to detect fiber line faults is up to 150 km, and it can improve the system power budget with a minimal insertion loss of 0.4 dB. The superiority of the suggested technique over real networks was verified with success by the Huawei labs' infrastructure nodes in the simulation experiment results. [ABSTRACT FROM AUTHOR]

Optical performance monitoring (OPM), particularly the optical power and optical signal-to-noise ratio (OSNR) of each wavelength channel, are of great importance and significance and need to be implemented to ensure stable and efficient operation/maintenance of wavelength division multiplexing (WDM) networks. However, the critical monitoring module of existing solutions generally are too expensive, operationally inconvenient and/or functionally limited to apply over WDM systems with numerous nodes. In this paper, a low-cost and high-efficiency OPM scheme based on differential phase shift keying (DPSK)-modulated digital optical labels is proposed and demonstrated. Each pilot tone is modulated by digital surveillance information and treated as an identity indicator and performance predictor that ties up to each wavelength channel and thereby can monitor the performance of all wavelength channels simultaneously by only one low-bandwidth photoelectric detector (PD) and by designed digital signal processing (DSP) algorithms. Simulation results showed that the maximum errors of channel power monitoring and OSNR estimation were both less than 1 dB after 20-span WDM transmission. In addition, offline experiments were also carried out and further verified the feasibility of our OPM scheme. This confirms that the optical label based OPM has lower cost and higher efficiency and thereby is of great potential for mass deployment in practical WDM systems. [ABSTRACT FROM AUTHOR]

Direct Detection (DD) optical performance monitoring (OPM), Modulation Format Identification (MFI), and Baud Rate Identification (BRI) are envisioned as crucial components of future-generation optical networks. They bring to optical nodes and receivers a form of adaptability and intelligent control that are not available in legacy networks. Both are critical to managing the increasing data demands and data diversity in modern and future communication networks (e.g., 5G and 6G), for which optical networks are the backbone. Machine learning (ML) has been playing a growing role in enabling the sought-after adaptability and intelligent control, and thus, many OPM, MFI, and BRI solutions are being developed with ML algorithms at their core. This paper presents a comprehensive survey of the available ML-based solutions for OPM, MFI, and BFI in non-coherent optical networks. The survey is conducted from a machine learning perspective with an eye on the following aspects: (i) what machine learning paradigms have been followed; (ii) what learning algorithms are used to develop DD solutions; and (iii) what types of DD monitoring tasks have been commonly defined and addressed. The paper surveys the most widely used features and ML-based solutions that have been considered in DD optical communication systems. This results in a few observations, insights, and lessons. It highlights some issues regarding the ML development procedure, the dataset construction and training process, and the solution benchmarking dataset. Based on those observations, the paper shares a few insights and lessons that could help guide future research.

Recently, IP over elastic optical network (IP-over-EON) has become a promising architecture for metro and core networks. This work studies how to visualize both layers of an IP-over-EON in real time, at different granularities (e.g., at flow-level, lightpath-level, and link-level), and with self-adaptivity. Specifically, we consider the multilayer application of in-band telemetry (INT) and propose entropy-driven adaptive INT (namely, EntropyINT). We introduce stateful processing to programmable data plane (PDP) switches for EntropyINT, such that they can make local decisions to determine whether and what type of telemetry data about the IP and EON layers should be encoded in each packet. The local decisions are designed to be based on the amount of information that can be conveyed by telemetry data to the network automation system. Meanwhile, we make EntropyINT cooperate with out-of-band monitoring, to detect and locate exceptions in the EON layer. Our proposal is implemented in a real-world testbed of IP-over-EON, to evaluate its assistance to network automation. Experimental results verify the effectiveness of our proposal, and indicate that the telemetry data collected by EntropyINT and out-of-band monitoring can better assist the machine learning in network automation, for status prediction and anomaly detection. [ABSTRACT FROM AUTHOR]

An intelligent optical performance monitoring scheme for simultaneous modulation format identification (MFI) and optical signal-to-noise ratio (OSNR) monitoring is proposed and experimentally demonstrated in digital coherent receivers. The proposed scheme introduces convolutional neural network (CNN) to automatically extract and monitor modulation format and OSNR dependent features (the empirical distribution of intensity and differential-phase) which can be obtained after constant modulus algorithm (CMA) equalization. The experiment results show that 100% identification accuracies for all modulation formats (e.g. 28-GBaud PDM-QPSK/−8PSK/ −8QAM/−16QAM/−32QAM/−64QAM) are achieved at OSNR values are lower than the corresponding theoretical 20% FEC limit (BER = 2.4×10−2). Furthermore, under the chromatic dispersion (CD) variation from 0-ps/nm to 1200-ps/nm, the root-mean- square error (RMSE) and mean absolute error (MAE) of OSNR monitoring for all modulation formats are less than 0.0896-dB and 0.0657-dB, respectively. Subsequently, the influence of frequency offset and fiber nonlinearity effect on the intelligent optical performance monitoring scheme is also analyzed. We believe that our proposed multi-parameter monitoring scheme has the potential to be applied in the next-generation intelligent elastic optical networks. [ABSTRACT FROM AUTHOR]

A successful migration from current C-band based optical networks to a multiband scenario primarily depends on the development of solutions that can reliably measure physical properties of optical links over broad spectral transmission windows. Additionally, these solutions must be capable of delivering wavelength-dependent and spatially-resolved indicators that can empower network operators to identify faults before they lead to severe service disruptions. Recently, the exploitation of receiver based digital signal processing as a tool for optical performance monitoring has gained tremendous popularity. One successful example is the so-called in-situ power profile estimator, which can reconstruct the per-channel longitudinal power profile along the optical fiber link solely processing the received signal samples. In this work, we propose a novel application for the in-situ power profile estimator by harnessing it on multiple wavelengths to accurately estimate the spectral gain profile of C+L-band in-line Erbium-doped fiber amplifiers deployed in a 280-km single mode fiber link. Furthermore, we show how this scheme can be efficiently used to detect amplification-related anomalies, such as gain tilt and narrowband gain compression. In our measurements, we achieved a sub-dB estimation accuracy by comparing the proposed gain extraction approach with the back-to-back characterization obtained from an optical spectrum analyzer. [ABSTRACT FROM AUTHOR]

Pilot tone (PT) is a low frequency, small intensity modulation applied to high speed optical channel. Traditionally PT is used for channel identification and channel power monitoring. A pilot tone scheme suitable for optical performance monitoring (OPM) in coherent optical communication systems is described. Two effects, Stimulated Raman scattering (SRS) and dispersion fading, are discussed. An important enhancement, multiband pilot tone, is introduced, which provides sub-channel monitoring capability. With the advanced PT technologies, signal spectrum, and relative frequency offset between signal and optical filter can be monitored with sub-GHz resolution. PT also enables direct OSNR and fiber nonlinear noise monitoring with high accuracy and sensitivity. [ABSTRACT FROM AUTHOR]

In this paper, and for the first time in literature, optical performance monitoring (OPM) of super-channel optical networks is considered. In particular, we propose a novel machine learning OPM technique based on the use of transformed in-phase quadrature histogram (IQH) features and support vector regressor (SVR) to estimate different optical parameters such as optical signal-to-noise ratio (OSNR) and chromatic dispersion (CD). Two transformation methods, the two-dimensional (2D) discrete Fourier transform (DFT) and 2D discrete cosine transform (DCT), are applied to the IQH to extract features with a considerably reduced dimensionality. For the purpose of simulation, the OPM of a 7 × 20 Gbaud dual-polarization–quadrature phase shift keying (DP-QPSK) is considered. Simulations reveal that it can accurately estimate the various optical parameters (i.e., OSNR and CD) with a coefficient of determination value greater than 0.98. In addition, the effectiveness of proposed OPM scheme is examined under different values of polarization mode dispersion and frequency offset, as well as the utilization of different higher order modulation formats. Moreover, proof-of-concept experiments are performed for validation. The results show an excellent matching between the simulation and experimental findings. [ABSTRACT FROM AUTHOR]

We propose a cascaded neural network (NN) to simultaneously identify the modulation formats and monitor the optical-signal-to-noise ratio (OSNR). In the second-level network, it is a single deep NN (DNN) rather than multiple sub-networks, which makes the architecture more compact and can save the resource for real implementation. However, since the data set is constituted from all modulation formats, the universality can be guaranteed but not for the accuracy and the complexity. To accelerate the estimation process and improve the accuracy, we introduce the transfer learning (TL) and reconstruct the data set with a part from all of the modulation formats for universality and another part from a specific modulation format for TL to pursue higher accuracy. In the experiment, we compare the proposed cascaded single neural network (CSNN) with or without TL, cascaded multiple neural networks (CMNN) and adaptive multi-task learning (MTL) for MFI and OSNR monitoring. In the first-level NN, all of the three schemes can achieve the accuracy of MFI as 100%. In the second-level NN, the CSNN with TL (TL-CSNN) can significantly improve the training speed and decline the RMSE of 0.19 dB compared with CSNN without TL. The TL-CSNN also has faster convergence speed and is more stable compared with CMNN and adaptive MTL.

Monitoring the state of the optical network is a key enabler for programmability of network functions, protocols and efficient use of the spectrum. A particular challenge is to provide the SDN-EON controller with a panoramic view of the complete state of the optical spectrum. This paper describes an architecture for compact on-chip spectrometry targeting high resolution across the entire C-band to reliably and accurately measure the spectral profile of WDM signals in fixed and flex-grid architectures. An industry standard software tool is used to validate the performance of the spectrometer. The fabrication of the proposed design is found to be practical.

Compared with traditional wavelength division optical network, elastic optical network (EON) divides the network spectrum into smaller spectrum slots to improve the spectrum utilization, but the high-quality spectrum division also complicates the routing and spectrum allocation (RSA) problem. Various strategies are proposed for reducing the RSA complexity and improving system traffic bearing capacity. However, previous RSA strategies do not consider the changing physical layer impairments that will also impact signal quality and even lead to violation of quality of transmission (QoT), the data cannot be transmitted correctly if the link state is degraded. Therefore, cross-layer optimization is desired, which means that different layer information is taken into account in the RSA strategy. In this paper, we propose a new link state-aware (LSA) RSA strategy to guarantee the QoT requirements under different link states. At first, the link state is evaluated based on chromatic dispersion (CD) and optical signal-to-noise ratio (OSNR), and a LightGBM model is exploited for CD and OSNR estimation. In LSA-RSA strategy, the link state is considered as a metric for qualified routing paths finding, and the link capacity is calculated based on the link state and used in spectrum allocation. Simulation results show that the average CD and OSNR estimation errors of the LightGBM model are 0.28ps/nm and 0.68dB, respectively. Under different link states and traffic loads, the LSA-RSA strategy can reduce traffic failure probability by more than 20%, and traffic load can increase 40Erlang when the bandwidth blocking probability equals 10%.

Future communication systems are faced with increased demand for high capacity, dynamic bandwidth, reliability and heterogeneous traffic. To meet these requirements, networks have become more complex and thus require new design methods and monitoring techniques, as they evolve towards becoming autonomous. Machine learning has come to the forefront in recent years as a promising technology to aid in this evolution. Optical fiber communications can already provide the high capacity required for most applications, however, there is a need for increased scalability and adaptability to changing user demands and link conditions. Accurate performance monitoring is an integral part of this transformation. In this paper, we review optical performance monitoring techniques where machine learning algorithms have been applied. Moreover, since many performance monitoring approaches in the optical domain depend on knowledge of the signal type, we also review work for modulation format recognition and bitrate identification. We additionally briefly introduce a neuromorphic approach as an emerging technique that has only recently been applied to this domain.

في هذا البحث تم اقتراح تقنية جديدة لمراقبة تشتت نمط الاستقطاب polarization mode dispersion (PMD) في شبكات الألياف الضوئية. تعتمد التقنية على تحويل مركبتي الاستقطاب المتعامدتين للإشارة الضوئية المراقبة إلى نفس حالة الاستقطاب ومن ثم يتم جمعهما معاً بفرق طور (180 درجة). يتم قياس مطال النبضات الناتجة والتي تتناسب مع تأخير المجموعة التفاضلية differential group delay (DGD) الذي يصف تشتت (PMD) لوصلة الليف الضوئي. تتميز التقنية المقترحة مقارنة بتقنيات المراقبة الأخرى، بمجال مراقبة جيد، وحساسية عالية، وعدم الحاجة لتعديل جهاز الإرسال، وبنية بسيطة، وعدم تأثرها بالتشتت اللوني (CD، وزمن استجابة سريع والقدرة على العمل بمعدلات بت مختلفة. تم استخدام برنامج المحاكاة الضوئي VPIphotonics لإثبات صحة عمل وحدة المراقبة المقترحة، وتحليل أدائها.

The aim of the research was to explore the possibilities of using the Asynchronous Delay Tap Sampling (ADTS) and Convolutional Neural Network (CNN) methods to monitor the simultaneously occurring phenomena in the physical layer of the optical network. The ADTS method was used to create a data sets showing the combination of Chromatic Dispersion (CD), Crosstalk and Optical to Signal Noise Ratio (OSNR) as optical disturbances in graphic form. Data were generated for 10 GB/s, Non-return-to-zero On–off keying (NRZ-OOK) and Differential Phase Shift Keying (DPSK) modulation and bit delays: 1 bit, 0.5 bit and 0.25 bit. A total of 6 data sets of 62,000 images each were obtained. The learning process was carried out for the number of epochs 50 and 1000. From the obtained learning results of the network, models with the best R 2 matching factor were selected. The learned models were further used to study the recognition of three phenomena simultaneously. The tests were carried out on sets of 2500 images in a combination of interference in the following ranges: 400–1600 ps/nm for CD and 10–30 dB for Crosstalk and OSNR. Very good results were obtained for recognizing simultaneously occurring phenomena using models learned up to 1000 epoch. Accuracy of over 99% was obtained for CD and Crosstalk for both modulations. In the case of the OSNR phenomenon, slightly weaker results were obtained above 96% in most cases. For models taught up to 50 epoch, very good results were obtained for the CD phenomenon (over 99%). For Crosstalk weaker results for OOK modulation were obtained. Poor results were obtained for the OSNR phenomenon, where recognition accuracy ranged from 50 to 80%, depending on the type of modulation and bit delay. Based on the conducted research, it was established that the use of ADTS and CNN methods enables monitoring of simultaneously occurring CD, Crosstalk and OSNR interference in the physical layer of the optical network, while maintaining the requirements for Optical Performance Monitoring systems. These requirements are met for network models learned up to 1000 epoch. [ABSTRACT FROM AUTHOR]

A correlation optical time-domain reflectometry (C-OTDR) method is presented, which measures the propagation delay with an accuracy of a few picoseconds. This accuracy is achieved using a test signal data rate of 10 Gbit/s and employing cross-correlation and pulse fitting techniques. In this paper we introduce and evaluate the basic signal processing steps, investigate the measurement accuracy, and discuss applications for monitoring link delay and chromatic dispersion of long fiber spans as well as temperature sensing applications. [ABSTRACT FROM AUTHOR]

Optical performance monitoring (OPM), particularly the optical power and optical signal-to-noise ratio (OSNR) of each wavelength channel, are of great importance and significance and need to be implemented to ensure stable and efficient operation/maintenance of wavelength division multiplexing (WDM) networks. However, the critical monitoring module of existing solutions generally are too expensive, operationally inconvenient and/or functionally limited to apply over WDM systems with numerous nodes. In this paper, a low-cost and high-efficiency OPM scheme based on differential phase shift keying (DPSK)-modulated digital optical labels is proposed and demonstrated. Each pilot tone is modulated by digital surveillance information and treated as an identity indicator and performance predictor that ties up to each wavelength channel and thereby can monitor the performance of all wavelength channels simultaneously by only one low-bandwidth photoelectric detector (PD) and by designed digital signal processing (DSP) algorithms. Simulation results showed that the maximum errors of channel power monitoring and OSNR estimation were both less than 1 dB after 20-span WDM transmission. In addition, offline experiments were also carried out and further verified the feasibility of our OPM scheme. This confirms that the optical label based OPM has lower cost and higher efficiency and thereby is of great potential for mass deployment in practical WDM systems.

A machine learning (ML)-based multifunctional optical spectrum analysis technique is proposed to perform not only the conventional analysis functions but also the extended analysis functions, including center wavelength detection, optical signal-to-noise (OSNR) calculation, bandwidth recognition, as well as spectral distortion diagnosis. We have investigated four widely used ML algorithms, including support vector machine (SVM), artificial neural network, k-nearest neighbors, and decision tree. First, the wavelengths, OSNRs, and bandwidths of optical signals are processed by four ML methods based on the spectral data. The good performance and fast processing speed are obtained, especially for SVM, achieving the optimal accuracy (100%) and the least test time. For the need of the practical application, we also investigate the more complicated case, where wavelength, OSNR, and bandwidth are variable simultaneously so that the ML should analyze these three parameters comprehensively. Even in this case, the overall accuracy is still larger than 99.1%. In addition, the extended analysis functions are also studied to diagnose the spectral distortion caused by the cascaded filtering effect and off-center filtering effect. The number of cascaded filters and the offsets of filter shift and laser drift can be effectively estimated by the SVM with high average accuracy and low standard deviation, which are useful for failure detection and distortion recovery. This technique has the potential to be applied in the optical spectrum analyzer to implement the multifunctional spectrum analysis or in the optical performance monitor to execute the spectral distortion diagnosis.

We propose an efficient and modulation-format-transparent method to jointly estimate optical signal-to-noise ratio (OSNR) and frequency offset (FO) by using signal spectrum correlations for coherent optical fiber communication systems. Based on the signal spectrum correlation analysis, a coarse FO estimation (FOE) and the corresponding compensation can be conducted after chromatic dispersion compensation (CDC). Then, OSNR monitoring can be accurately operated without deterioration caused by FO. Meanwhile, to realize fine FOE, down-sampling process of signal can be used to reduce the complexity of fast Fourier transform-based FOE (FFT-FOE) without losing FOE resolution. Simulation results show that when the OSNR is in the range of 10 to 30 dB, the proposed scheme presents an absolute OSNR estimation error lower than 0.18 dB. Finally, we experimentally demonstrate our scheme in an optical back-to-back (B2B) transmission link using 28 Gbaud dual-polarization (DP)-4/16/32-quadrature amplitude modulation (QAM) formats, and an absolute OSNR estimation error lower than 0.66 dB is achieved for OSNR ranging from 15 to 30 dB. [ABSTRACT FROM AUTHOR]

A cost-effective optical performance monitoring (OPM) scheme is proposed to realize modulation format identification (MFI), baud rate identification (BRI), chromatic dispersion identification (CDI), and optical signal-to-noise ratio (OSNR) estimation of optical signals simultaneously. This technique is based on multi-task learning (MTL) neural network model with adaptive asynchronous delay tap plot (AADTP) and asynchronous amplitude histogram (AAH) by direct detection in the intermediate nodes of optical networks. The generation of AADTP depends on the sampling rate but not the symbol rate, which makes the scheme transparent to the baud rate. The combined inputs of AADTP with AAH improve accuracies of the neural network, compared with a single input. This scheme is verified experimentally where signals with two formats, quadrature phase shift keying (QPSK) and 16 quadrature amplitude modulation (16QAM), two baud rates, 14 GBaud and 28 GBaud, and three CD situations, 0 ps/nm, 858.5 ps/nm, and 1507.9 ps/nm, are adopted. The best accuracies of MFI, BRI, CDI are 100%, 99.81%, and 99.83%, respectively. Meanwhile, the lowest average mean absolute error (MAE) of OSNR estimation is 0.2867 dB over the range of 10–24 dB (QPSK) and 15–29 dB (16QAM). It is cost-effective and practical for the proposed OPM technique to be applied in the intermediate nodes to construct smart optical networks since it uses only one photodetector assisted with an advanced deep learning algorithm. [ABSTRACT FROM AUTHOR]

In this paper, and for the first time in literature, optical performance monitoring (OPM) of super-channel optical networks is considered. In particular, we propose a novel machine learning OPM technique based on the use of transformed in-phase quadrature histogram (IQH) features and support vector regressor (SVR) to estimate different optical parameters such as optical signal-to-noise ratio (OSNR) and chromatic dispersion (CD). Two transformation methods, the two-dimensional (2D) discrete Fourier transform (DFT) and 2D discrete cosine transform (DCT), are applied to the IQH to extract features with a considerably reduced dimensionality. For the purpose of simulation, the OPM of a 7 × 20 Gbaud dual-polarization–quadrature phase shift keying (DP-QPSK) is considered. Simulations reveal that it can accurately estimate the various optical parameters (i.e., OSNR and CD) with a coefficient of determination value greater than 0.98. In addition, the effectiveness of proposed OPM scheme is examined under different values of polarization mode dispersion and frequency offset, as well as the utilization of different higher order modulation formats. Moreover, proof-of-concept experiments are performed for validation. The results show an excellent matching between the simulation and experimental findings.

We propose a cascaded neural network (NN) to simultaneously identify the modulation formats and monitor the optical-signal-to-noise ratio (OSNR). In the second-level network, it is a single deep NN (DNN) rather than multiple sub-networks, which makes the architecture more compact and can save the resource for real implementation. However, since the data set is constituted from all modulation formats, the universality can be guaranteed but not for the accuracy and the complexity. To accelerate the estimation process and improve the accuracy, we introduce the transfer learning (TL) and reconstruct the data set with a part from all of the modulation formats for universality and another part from a specific modulation format for TL to pursue higher accuracy. In the experiment, we compare the proposed cascaded single neural network (CSNN) with or without TL, cascaded multiple neural networks (CMNN) and adaptive multi-task learning (MTL) for MFI and OSNR monitoring. In the first-level NN, all of the three schemes can achieve the accuracy of MFI as 100%. In the second-level NN, the CSNN with TL (TL-CSNN) can significantly improve the training speed and decline the RMSE of 0.19 dB compared with CSNN without TL. The TL-CSNN also has faster convergence speed and is more stable compared with CMNN and adaptive MTL. [ABSTRACT FROM AUTHOR]

This article considers, for the first time, optical performance monitoring (OPM) in few mode fiber (FMF)-based optical networks. 1-D features vector, extracted by projecting a 2-D asynchronous in-phase quadrature histogram (IQH), and the 2D IQH are proposed to achieve OPM in FMF-based network. Three machine learning algorithms are employed for OPM and their performances are compared. These include support vector machine, random forest algorithm, and convolutional neural network. Extensive simulations are conducted to monitor optical to signal ratio (OSNR), chromatic dispersion (CD), and mode coupling (MC) for dual polarization-quadrature phase shift keying (DP-QPSK) at 10, 12, 16, 20, and 28 Gbaud transmission speeds. Besides, M-ary quadrature amplitude modulation (M = 8 and 16) is considered. Also, the OPM accuracy is investigated under different FMF channel conditions including phase noise and polarization mode dispersion. Simulation results show that the proposed 1D projection features vector provides better OPM results than those of the widely used asynchronous amplitude histogram (AAH) features. Furthermore, it has been found that the 2D IQH features outperform the 1D projection features but require larger number of features samples. Additionally, the effect of fiber nonlinearity on the OPM accuracy is investigated. Finally, OPM using the 2D IQH features has been verified experimentally for 10 Gbaud DP-QPSK signal. The obtained results show a good agreement between both simulation and experimental findings. [ABSTRACT FROM AUTHOR]

With the rapid development of optical communication systems, more advanced techniques conventionally used in long-haul transmissions have gradually entered systems covering shorter distances below 100 km, where higher-speed connections are required in various applications, such as the optical access networks, inter- and intra-data center interconnects, mobile fronthaul, and in-building and indoor communications. One of the techniques that has attracted intensive interests in short-reach optical communications is machine learning (ML). Due to its robust problem-solving, decision-making, and pattern recognition capabilities, ML techniques have become an essential solution for many challenging aspects. In particular, taking advantage of their high accuracy, adaptability, and implementation efficiency, ML has been widely studied in short-reach optical communications for optical performance monitoring (OPM), modulation format identification (MFI), signal processing and in-building/indoor optical wireless communications. Compared with long-reach communications, the ML techniques used in short-reach communications have more stringent complexity and cost requirements, and also need to be more sensitive. In this paper, a comprehensive review of various ML methods and their applications in short-reach optical communications are presented and discussed, focusing on existing and potential advantages, limitations and prospective trends.

Real-time optical performance monitoring (OPM) is of the utmost importance in adaptive optical networks to enable awareness of channel conditions and to achieve high quality of service. In single-mode fiber (SMF)-based networks, optical signal-to-noise ratio (OSNR) and chromatic dispersion (CD) monitoring have been extensively studied in the literature. In this work, we consider OPM in few-mode fiber (FMF) networks employing non-coherent detection. OPM in such networks is a challenging task, as FMF has an additional performance-limiting impairment over SMF, namely mode coupling (MC). Here, we propose an OPM scheme to estimate three FMF channel parameters: OSNR within the range of 8 to 20 dB, CD within the range of 160 to 1120 ps/nm, and different levels of MC. The proposed scheme uses a stacked auto-encoder (AE) to extract features with reduced dimensionality compared to the original data. These features are used to train an artificial neural network (ANN) regressor. Simulation results show that the proposed OPM scheme can accurately estimate the OSNR, CD, and MC with root mean square error (RMSE) values of 0.0015 dB, 0.28 ps/nm, and 7.88 × 10−6, respectively. The performance of proposed OPM scheme is also evaluated against different types of features commonly used in literature.

This paper considers Godard's error as signal quality metric to monitor chromatic dispersion (CD), nonlinear parameter and modulation format in the DSP module of the coherent receivers. We first review a CD monitoring based on Godard's error that can be able to accurately monitor arbitrarily large dispersion values in uncompensated transmission links in combination with frequency domain equalizer, then extend the previous nonlinear parameter monitoring method based on Godard's error by blindly obtaining the optimized value γξp to significantly improve the adaptive capability, and present a simple and effective modulation format monitoring based on Godard's error. Meanwhile, the effectiveness has been experimentally verified in 128-Gb/s PDM-QPSK, 192-Gb/s PDM-8QAM, and 256-Gb/s PDM-16QAM systems.

Monitoring the state of the optical network is a key enabler for programmability of network functions, protocols and efficient use of the spectrum. A particular challenge is to provide the SDN-EON controller with a panoramic view of the complete state of the optical spectrum. This paper describes an architecture for compact on-chip spectrometry targeting high resolution across the entire C-band to reliably and accurately measure the spectral profile of WDM signals in fixed and flex-grid architectures. An industry standard software tool is used to validate the performance of the spectrometer. The fabrication of the proposed design is found to be practical. [ABSTRACT FROM AUTHOR]

Optical performance monitoring (OPM) and the corresponding telemetry systems play an important role in modern optical transport networks based on software-defined networking (SDN). There have been extensive studies and standardization activities to build high-speed and high-accuracy OPM/telemetry systems that can ensure sufficient monitoring data for effective network control and management. However, current solutions for OPM/telemetry assume that control and management planes (C/M-plane) always provide sufficient bandwidth (BW) to deliver telemetry data. Unfortunately, in the event of several concurrent network failures (e.g., following a large-scale disaster), C/M-plane networks can become heavily degraded and/or unstable, and even experience isolation of some of their parts. Under such circumstances, the existing OPM systems would hardly function. To enhance resiliency and to ensure the quick recovery of OPM/telemetry in case of disaster, we propose an approach for quick recreation of OPM and for achieving robust telemetry based on OpenConfig YANG. Our proposal addresses three key problems: (1) how to quickly recreate the lost OPM capability, (2) how to address the mismatch between the high data rate of OPM and the low BW in the C/M-plane network, and (3) how to flexibly reconfigure the telemetry system to be adaptive to sudden BW changes in the C/M-plane network. We implement a testbed and experimentally demonstrate that our proposal can tolerate low post-disaster bandwidth and can adapt the telemetry system following the changing conditions of the C/M-plane network. [ABSTRACT FROM AUTHOR]

Supported by advanced digital signal processing algorithms and application specific integrated circuits, coherent receivers in elastic optical networks will be capable of measuring link impairments in real time. Specifically, coherent receivers can work as soft optical performance monitors. Optical spectra usually contain rich information about optical links and have been exploited to assist soft failure detection and identification. However, acquiring optical spectra needs the deployment of numerous optical spectrum analyzers. Instead, the digital spectra of received signals in coherent receivers are easy to obtain without the penalty of additional hardware. In this paper, we explore the feasibility of the digital spectra in assisting soft-failure detection (SFD) and soft failure identification (SFI). A digital spectrum based SFD and SFI framework is proposed. A dual-stage SFD structure is employed to reduce the monitoring and processing overhead in optical nodes. At the first-stage SFD, only bit error rate and received optical power are collected. When an anomalous sample is detected, extra digital spectrum features are extracted and collected for the second-stage SFD. Extensive numerical results are presented to analyze the digital spectrum characteristics and feature distributions of four common soft failures. Finally, we experimentally evaluate the detection and identification performance of the proposed method. With reasonable complexity, a false positive rate of 0.42% and a false negative rate of 1.47% can be achieved for SFD, and an identification accuracy of 99.55% can be obtained for SFI. [ABSTRACT FROM AUTHOR]

A dual-stage algorithm structure is proposed to improve estimation accuracy and reliability for low-margin elastic optical network. At the first-stage, a multitask learning-based artificial neural network (MTL-ANN) is proposed to estimate multiple parameters simultaneously. At the second-stage, a threshold-based decision module is deployed to divide the estimation results into reliable results and doubtful results. As to the doubtful results, we investigate the deviation range and underestimate the results to allocate adequate system margin. The algorithm structure is experimentally demonstrated for optical signal-to-noise ratio (OSNR) monitoring and modulation format identification (MFI) in a polarization division multiplexing (PDM) coherent optical system. Signals’ amplitude histograms (AHs) of circular constellation diagrams are selected as the input features. The results show that the MFI accuracy of nine M-QAM formats under consideration is 100%. With 93.6% OSNR estimation accuracy at first-stage, OSNR estimation with accuracy higher than 99% is achieved for the reliable results. In addition, the confidence level of doubtful results within 3 dB deviation is 0.96. [ABSTRACT FROM AUTHOR]

With the development of 5G technology, high definition video and internet of things, the capacity demand for optical networks has been increasing dramatically. To fulfill the capacity demand, low-margin optical network is attracting attentions. Therefore, planning tools with higher accuracy are needed and accurate models for quality of transmission (QoT) and impairments are the key elements to achieve this. Moreover, since the margin is low, maintaining the reliability of the optical network is also essential and optical performance monitoring (OPM) is desired. With OPM, controllers can adapt the configuration of the physical layer and detect anomalies. However, considering the heterogeneity of the modern optical network, it is difficult to build such accurate modeling and monitoring tools using traditional analytical methods. Fortunately, data-driven artificial intelligence (AI) provides a promising path. In this paper, we firstly discuss the requirements for adopting AI approaches in optical networks. Then, we review various recent progress of AI-based QoT/impairments modeling and monitoring schemes. We categorize these proposed methods by their functions and summarize advantages and challenges of adopting AI methods for these tasks. We discuss the problems remained for deploying AI-based methods to a practical system and present some possible directions for future investigation. [ABSTRACT FROM AUTHOR]

In this work, deep learning is used to monitor coherent channel performance with eye diagram measurement. Experiments show that the proposed technique can determine the modulation format, optical signal to noise ratio (OSNR), roll-off factor (ROF), and timing skew of a quadrature amplitude modulation (QAM) transmitter with high accuracy. Trained vanilla convolutional neural network (CNN) and MobileNet can be utilized to jointly monitor above four parameters with >98% prediction accuracy for 32 GBd coherent channels with quadrature phase shift keying (QPSK), 8-QAM or 16-QAM formats, under OSNR from 15 to 40 dB, IQ skew from −15 to 15 ps, ROF from 0.05 to 1. Our proposed deep learning approach outperforms many traditional machine learning methods, such as decision tree, k-nearest neighbor algorithm (KNN), and histogram of oriented gradient (HOG) based support vector machine (SVM). Unlike other optical performance monitoring approaches, the use of eye diagram measurement combined with deep learning could enable joint monitoring of multiple system performance parameters with reduced hardware implementation complexity. Comparing with vanilla CNN, MobileNet has relatively simplified iteration algorithm, thus reduces the requirement on the computational power, while still maintaining high accuracy for classification issues. [ABSTRACT FROM AUTHOR]