Today’s telecommunication networks have become sources of enormous amounts of widely heterogeneous data. This information can be retrieved from network traffic traces, network alarms, signal quality indicators, users’ behavioral data, etc. Advanced mathematical tools are required to extract meaningful information from these data and take decisions pertaining to the proper functioning of the networks from the network-generated data. Among these mathematical tools, machine learning (ML) is regarded as one of the most promising methodological approaches to perform network-data analysis and enable automated network self-configuration and fault management. The adoption of ML techniques in the field of optical communication networks is motivated by the unprecedented growth of network complexity faced by optical networks in the last few years. Such complexity increase is due to the introduction of a huge number of adjustable and interdependent system parameters (e.g., routing configurations, modulation format, symbol rate, coding schemes, etc.) that are enabled by the usage of coherent transmission/reception technologies, advanced digital signal processing, and compensation of nonlinear effects in optical fiber propagation. In this paper we provide an overview of the application of ML to optical communications and networking. We classify and survey relevant literature dealing with the topic, and we also provide an introductory tutorial on ML for researchers and practitioners interested in this field. Although a good number of research papers have recently appeared, the application of ML to optical networks is still in its infancy: to stimulate further work in this area, we conclude this paper proposing new possible research directions.

/pdf/an-overview-on-application-of-machine-learning-techniques-in-3eel2cbu3o.pdf

An Overview on Application of Machine Learning Techniques in Optical Networks

Client-side optics are facing an ever-increasing upgrading pace, driven by upcoming 5G related services and datacenter applications. The demand for a single lane data rate is soon approaching 200 Gbps. To meet such high-speed requirement, all segments of traditional intensity modulation direct detection (IM/DD) technologies are being challenged. The characteristics of electrical and optoelectronic components and the performance of modulation, coding, and digital signal processing (DSP) techniques are being stretched to their limits. In this context, we witnessed technological breakthroughs in several aspects, including development of broadband devices, novel modulation formats and coding, and high-performance DSP algorithms for the past few years. A great momentum has been accumulated to overcome the aforementioned challenges. In this article, we focus on IM/DD transmissions, and provide an overview of recent research and development efforts on key enabling technologies for 200 Gbps per lane and beyond. Our recent demonstrations of 200 Gbps short-reach transmissions with 4-level pulse amplitude modulation (PAM) and discrete multitone signals are also presented as examples to show the system requirements in terms of device characteristics and DSP performance. Apart from digital coherent technologies and advanced direct detection systems, such as Stokes–vector and Kramers–Kronig schemes, we expect high-speed IM/DD systems will remain advantageous in terms of system cost, power consumption, and footprint for short reach applications in the short- to mid- term perspective.

/pdf/200-gbps-lane-im-dd-technologies-for-short-reach-optical-2nuta7bqsd.pdf

200 Gbps/Lane IM/DD Technologies for Short Reach Optical Interconnects

We present the results of the comparative performance-versus-complexity analysis for the several types of artificial neural networks (NNs) used for nonlinear channel equalization in coherent optical communication systems. The comparison is carried out using an experimental set-up with the transmission dominated by the Kerr nonlinearity and component imperfections. For the first time, we investigate the application to the channel equalization of the convolution layer (CNN) in combination with a bidirectional long short-term memory (biLSTM) layer and the design combining CNN with a multi-layer perceptron. Their performance is compared with the one delivered by the previously proposed NN-based equalizers: one biLSTM layer, three-dense-layer perceptron, and the echo state network. Importantly, all architectures have been initially optimized by a Bayesian optimizer. First, we present the general expressions for the computational complexity associated with each NN type; these are given in terms of real multiplications per symbol. We demonstrate that in the experimental system considered, the convolutional layer coupled with the biLSTM (CNN+biLSTM) provides the largest Q-factor improvement compared to the reference linear chromatic dispersion compensation (2.9 dB improvement). Then, we examine the trade-off between the computational complexity and performance of all equalizers and demonstrate that the CNN+biLSTM is the best option when the computational complexity is not constrained, while when we restrict the complexity to some lower levels, the three-layer perceptron provides the best performance.

/pdf/performance-versus-complexity-study-of-neural-network-1zxty94i7v.pdf

Performance Versus Complexity Study of Neural Network Equalizers in Coherent Optical Systems

Responding to the growing bandwidth demand by emerging applications, such as fixed-mobile convergence for fifth generation (5G) and beyond 5G, 100 Gb/s/ λ access network becomes the next research focus of passive optical network (PON) roadmap. Intensity modulation and direct detection (IMDD) technology is still considered as a promising candidate for 100 Gb/s/ λ PON attributed to its low cost, low power consumption, and small footprint. In this paper, we achieve 100 Gb/s/ λ IMDD PON by using 20G-class optical and electrical devices due to its commercial availability. To mitigate the system linear and nonlinear distortions, neural network (NN) based equalizer is used and the performance is compared with feedforward equalizer (FFE) and Volterra nonlinear equalizer (VNE). We introduce the rules to train and test the data while using NN-based equalizer to guarantee a fair comparison with FFE and VNE. Random data have to be used for training, but for test, both random data and pseudorandom bit sequence are applicable. We found that the NN-based equalizer has the same performance with FFE and VNE in the case of linear distortion, but outperforms them in a strong nonlinearity case. In the experiment, to improve the loss budget, we increase the launch power to 18 dBm, achieving a 30-dB loss budget for 33 GBd/s PAM8 signal at the system frequency response of 16.2 GHz, attributed to the strong nonlinear equalization capability of NN.

Machine Learning for 100 Gb/s/ λ Passive Optical Network

We introduce for the first time the utilization of Long short-term memory (LSTM) neural network architectures for the compensation of fiber nonlinearities in digital coherent systems. We conduct numerical simulations considering either C-band or O-band transmission systems for single channel and multi-channel 16-QAM modulation format with polarization multiplexing. A detailed analysis regarding the effect of the number of hidden units and the length of the word of symbols that trains the LSTM algorithm and corresponds to the considered channel memory is conducted in order to reveal the limits of LSTM based receiver with respect to performance and complexity. The numerical results show that LSTM Neural Networks can be very efficient as post processors of optical receivers which classify data that have undergone non-linear impairments in fiber and provide superior performance compared to digital back propagation, especially in the multi-channel transmission scenario. The complexity analysis shows that LSTM becomes more complex as the number of hidden units and the channel memory increase, however LSTM can be less complex than Digital Back Propagation in long distances (>1000 km).

Compensation of Fiber Nonlinearities in Digital Coherent Systems Leveraging Long Short-Term Memory Neural Networks

We demonstrate 56Gbps PAM4 PON transmission over 25km SSMF using 10G-class DML and APD with 6 GHz 3-dB bandwidth. 29 dB loss budget is achieved by a novel equalization technique based on convolutional neural network.

56 Gbps IM/DD PON based on 10G-Class Optical Devices with 29 dB Loss Budget Enabled by Machine Learning

Increasing the serial bitrate to 25-Gb/s per channel as well as coexisting with the same optical distribution network of the 10G-ethernet passive optical network (EPON) is challenging. The bandwidth limitation and the reduced dispersion tolerance at high date rate make it much more difficult to achieve higher loss budget. In this paper, we have first shown a demonstration of a symmetric 4 × 25-Gb/s time-wavelength-division-multiplexed passive optical network system based on nonreturn-to-zero on-off-keying format supporting 0–20 km reach in O-band. 10G-class directly modulated lasers (DMLs) are employed to serve as both upstream and downstream transmitters owing to the low cost and the frequency chirp induced by DML can be used for bandwidth equalization. Instead of using digital signal processing, the chirp management and frequency equalization are simultaneously realized in optical domain through our proposed dispersion supported equalization technique. By employing a spool of dispersion-shifted fiber in optical line terminal (OLT), a single device can be used to simultaneously equalize the frequency response of multiple downstream and upstream channels. Then, we further use a semiconductor optical amplifier as a booster and preamplifier at the OLT for downstream and upstream, respectively. A power budget of 26 and 32 dB with 0–20 km reach of standard single mode fiber for downstream and upstream transmission at 1310 nm is obtained. The experimental results shown in this paper reveal that our proposed cost-effective scheme would be a promising candidate for the next generation 100G-EPON.

Symmetric 100-Gb/s TWDM-PON in O-Band Based on 10G-Class Optical Devices Enabled by Dispersion-Supported Equalization

Directly-modulated laser (DML) is widely employed in intensity modulation and direct detection (IMDD) system due to its low cost and high output power. However, the corresponding frequency chirp is regarded as one of the main disadvantages for its application in passive optical networks (PONs). In this paper, we theoretically analyze the frequency response evolution of DML based system under different chirp and dispersion conditions, proving that the system bandwidth can be improved by interactions between negative dispersion and DML chirp. Based on this concept, we experimentally demonstrated downstream 50 Gb/s PAM4 signal transmission over 20 km single-mode fiber (SMF) access based on the 10 Gb/s DML operating at 1310 nm and avalanche photodiode (APD). A dispersion-shifted fiber (DSF) providing −150 ps/nm dispersion at 1310 nm in the optical line terminal (OLT) is used to pre-equalize the frequency response of bandwidth-limited directly modulated signals in the optical domain. Thanks to our proposed dispersion-supported equalization (DSE) technique, the system bandwidth can be improved by 5 GHz. Feed-forward equalization (FFE), decision feedback equalization (DFE) and Volterra filter are employed to evaluate the signal performance improvement, respectively. By evaluating the receiver sensitivity, the DSE combined with FFE scheme shows 2 dB improvement than the complex Volterra algorithm, indicating its potential to reduce the complexity of digital signal processing (DSP) and therefore a lower cost and power consumption in PON.

Lei Xue

Papers

200 Gbps/Lane IM/DD Technologies for Short Reach Optical Interconnects

Machine Learning for 100 Gb/s/ λ Passive Optical Network

56 Gbps IM/DD PON based on 10G-Class Optical Devices with 29 dB Loss Budget Enabled by Machine Learning

Symmetric 100-Gb/s TWDM-PON in O-Band Based on 10G-Class Optical Devices Enabled by Dispersion-Supported Equalization

Optics-Simplified DSP for 50 Gb/s PON Downstream Transmission using 10 Gb/s Optical Devices