The development of digital twins to represent the optical transport network might enable multiple applications for network operation, including automation and fault management. In this work, we propose a deep-learning-based digital twin for the optical time domain, named OCATA. OCATA is based on the concatenation of deep neural network (DNN) modeling of optical links and nodes, which facilitates representing lightpaths. The DNNs model linear and nonlinear noise, as well as optical filtering. Additional DNN-based models are proposed to extract useful lightpath metrics, such as lightpath length, number of optical links, and nonlinear fiber parameters. OCATA exhibits low complexity, thus making it ideal for real-time applications. Illustrative results for the application of OCATA to disaggregated and mixed disaggregated-proprietary optical network scenarios reveal remarkable accuracy.

Optical network automation requires accurate physical layer models, not only for provisioning but also for real-time analysis. In particular, in-phase (I) and quadrature (Q) constellation analysis enables deep understanding of the characteristics of optical connections (lightpaths), e.g., their length. In this paper, we present methods for modeling lightpaths based on deep learning. Specifically, we propose using autoencoders (AEs) and deep neural networks. Models are trained and composed in a sandbox domain with the information received from the network controller and sent to the node agent that uses them to compare the features extracted from the received signal and the expected features returned by the models. We investigate two different use cases for lightpath analysis focused on lightpath length and optical signal power. The results show a remarkable accuracy for the lightpath modeling and length prediction and a noticeable performance of the AEs for unsupervised IQ constellation feature extraction and relevance analysis.

Digital Twins (DT) are needed in elastic optical networks to enable further advance and pave the way towards autonomous networks. Dynamical models self-adjusting with their physical counterpart have to maintain low computational complexity and guarantee good accuracy, regardless of the network topology and the technology supporting the optical lightpath. In this work, we propose data-driven models based on cascaded learning for single carrier signal as well as digital subcarrier multiplexing signals. Firstly, we explore the potentials of such a DT by proposing an algorithm for lightpath provisioning. In addition, the performances of the proposed models are compared with the ones of analytical models. Secondly, we experimentally evaluate its accuracy to predict the impact of optical filtering penalties. The simulations show an overall good accuracy for different signal formats and demonstrate their viability for lightpath provisioning. Similarly, the experimental results demonstrate the feasibility of implementing such models to estimate the quality of transmission for digital subcarrier multiplexed signals.

Network automation promises to reduce costs while guaranteeing the required performance; this is paramount when dealing with the forecasted highly dynamic traffic that will be generated by new 5G/6G applications. In optical networks, autonomous lightpath operation entails that the optical receiver can identify the configuration of a received optical signal without necessarily being configured from the network controller. This provides relief for the network controller from real-time operation, and it can simplify the operation in multi-domain scenarios, where an optical connection spans across more than one domain. Consequently, in this work, we propose a blind and low complex modulation format (MF) and symbol rate (SR) identification algorithm. The algorithm is based on studying the effects of decoding an optical signal with different MFs and SRs. Extensive MATLAB-based simulations have been carried out which consider a coherent wavelength division multiplexed system based on 32 and 64 quadrature amplitude modulated signals at up to 96 GBd, thus enabling bit rates of up to 800 Gb/s/channel. The results show remarkable identification accuracy in the presence of linear and nonlinear noise for a wide range of feasible configurations.