Source-linked AI summary
Artificial Intelligence (AI) Methods in Optical Networks: A Comprehensive Survey
Javier Mata, Ignacio de Miguel, Ramó n J. Durá n, Noemí Merayo, Sandeep Kumar Singh, Admela Jukan, Mohit Chamania
TL;DR
Optical communication networks face increasing complexity, dynamism, noise, nonlinear distortions, and operational challenges. This paper comprehensively surveys AI applications across optical transmission and network design, control, and management, finding that AI can provide efficient, adaptable solutions in complex scenarios. The survey also identifies opportunities and challenges for future optical networking.
Problem
Optical networks must address complex, dynamic conditions, including physical impairments and heterogeneous devices, while classical analytical approaches can be limited by complexity, adaptability, and scalability.
Method
The paper conducts a comprehensive survey of AI techniques for optical transmission, network design and control, and network management.
Results
The reviewed literature covers AI applications from component characterization and performance monitoring to nonlinearities mitigation, QoT estimation, planning, connection establishment, reconfiguration, and software-defined networking.
Takeaways & Limitations
AI techniques are promising for finding optimal or near-optimal solutions in highly complex optical-network scenarios and adapting to changing or unexpected conditions.
Takeaways & Limitations
Expectation maximization depends on transmission-link parameters and is therefore not applicable to dynamic optical networks.
Abstract
from arXiv · showhide
Artificial intelligence (AI) is an extensive scientific discipline which enables computer systems to solve problems by emulating complex biological processes such as learning, reasoning and self-correction. This paper presents a comprehensive review of the application of AI techniques for improving performance of optical communication systems and networks. The use of AI-based techniques is first studied in applications related to optical transmission, ranging from the characterization and operation of network components to performance monitoring, mitigation of nonlinearities, and quality of transmission estimation. Then, applications related to optical network control and management are also reviewed, including topics like optical network planning and operation in both transport and access networks. Finally, the paper also presents a summary of opportunities and challenges in optical networking where AI is expected to play a key role in the near future.
1. Introduction
The paper surveys how AI techniques can improve optical communication systems and networks, covering applications from transmission components to network control and management. It also frames current research and the opportunities and challenges associated with AI-based optical networking.
- AI techniques are increasingly being applied to optical communication networks and systems, spanning photonic devices, transmission, switching, and network management.
- The paper reviews current approaches that use AI mechanisms to increase optical-network performance.
- The paper concludes its review by identifying further opportunities and challenges for AI in optical networking.
- The survey covers AI applications in optical transmission and networking, including component operation, performance monitoring, network design, control, and management.
2. An Overview of AI and Related Techniques
The paper presents optical networking as an AI setting involving uncertainty, sequential decision-making, learning, and multiple intelligent-agent techniques. It surveys AI subfields and examples ranging from optimization and game theory to knowledge-based methods and machine learning.
- The survey classifies AI techniques used or expected to matter in optical networking and organizes the reviewed literature by these subfields.Figure 1 groups the relevant references within the AI categories discussed in this section.
- Search methods and optimization theory: Search algorithms and optimization theory support optical network design and control, while metaheuristics complement them when assumptions are relaxed or network size becomes prohibitive.Examples include breadth-first search, linear programming, mixed-integer linear programming, simulated annealing, genetic algorithms, and swarm optimization.
- Game theory: Game theory is relevant when multiple intelligent agents interact and one agent’s actions affect others in the network.The survey cites applications in hybrid radiofrequency/free-space-optics and elastic optical networks.
- Statistical models: Optical networks require robust handling of uncertainty because events can be nondeterministic and environmental information may be incomplete.Bayesian networks are identified as tools for constructing robust models under uncertainty.
- Decision-making algorithms: Decision-making in optical networking can be modeled as sequential decision problems in uncertain environments, using utility-based methods and Markov decision processes.MDPs use transition and reward models to derive policies associating decisions with reachable states.
- Learning methods: Learning enables intelligent agents to improve future-task performance, adapt to environmental changes, and address unforeseen scenarios.The section describes Bayesian, maximum-a-posteriori, maximum-likelihood, supervised, and reinforcement-learning methods, including Q-learning for path and wavelength selection.
3. Applications of AI in Optical Transmission
AI techniques are applied across optical transmission to configure components, monitor impairments, mitigate nonlinearities, and estimate quality of transmission. The survey reports improved modeling, detection, monitoring, and prediction, while identifying constraints in training requirements and dynamic-link applicability.
- Component characterization and operation: AI improves optical-transmission device operation by modeling component behavior and optimizing transmitter, laser, and EDFA settings.Applications include simulated annealing for optical-comb flatness, self-tuning mode-locked lasers, and regression-based recommendations for EDFA channel add/drop strategies.
- Performance monitoring: AI-based monitoring estimates optical impairments including OSNR, chromatic dispersion, polarization-mode dispersion, and accumulated nonlinearities.Neural networks and principal-component-analysis pattern recognition support simultaneous monitoring, including approaches independent of bitrate and modulation format when signals belong to a known set.
- Performance monitoring: 400,000 samples and at least 5 layers are required by one DNN approach to achieve accurate OSNR estimation, resulting in long training time.The survey presents this as a scalability and training-cost constraint of the reported DNN-based estimator.
- Receivers and nonlinearities: Machine learning mitigates nonlinear impairments through format recognition, physics-informed signal processing, nonlinear classification, and clustering-based demodulation.Reported applications include autonomous identification of QPSK, 8PSK, and 16QAM; nonlinear SVM decision boundaries; k-nearest-neighbor multi-class detection; and mitigation of IQ imbalance, bias drift, and phase noise.
- QoT estimation: Up to four orders of magnitude faster than the Q-Tool, case-based reasoning estimates whether lightpaths satisfy QoT requirements while maintaining high successful-classification rates.The approach reuses or adapts similar cases from a knowledge base and was demonstrated in a WDM 80 Gb/s PDM-QPSK testbed with a very small knowledge base.
4. Applications of AI in Optical Networking
AI techniques support optical-network planning, resource allocation, connection establishment, traffic-aware operation, and virtual-topology reconfiguration. The reviewed methods use optimization, metaheuristics, clustering, neural networks, and historical knowledge to address complex or changing network conditions.
- Optical network planning: Optical-network planning uses genetic algorithms, particle swarm optimization, and related metaheuristics for topology, survivability, dimensioning, routing, wavelength assignment, and regenerator placement.These methods address cost, quality-of-transmission, fault-tolerance, QoS, and energy-efficiency constraints across transport and hybrid WDM/OCDM networks.
- Optical network planning: K-means clustering schedules WDM-star-network messages and avoids consecutive messages to the same destination, protecting channel utilization.The algorithm addresses both message sequencing and channel assignment by using the produced clusters.
- Connection establishment: 25% lower computational time is reported for case-based reasoning in dynamic WRON connection establishment while maintaining or improving performance.The method exploits similar past cases stored in a knowledge base; other reviewed approaches include ACO for energy-efficient routing and demand selection.
- Connection establishment: More than 7500 times faster than a discrete-event simulator, an artificial-neural-network mechanism estimates lightpath blocking probability from topology and physical-layer characteristics.Principal component analysis is applied before the neural-network estimation.
- Connection establishment: Metaheuristics and neural networks address RSA and RMLSA in elastic optical networks, including routing, modulation, spectrum, core assignment, multicast, protection, and traffic prediction.The survey reports coevolutionary methods outperforming tabu-search and simulated-annealing proposals, while historical-information neural networks outperform RSA methods without such information.
- Network reconfiguration: virtual topologies: Virtual-topology reconfiguration uses GA, ACO, and cognitive mechanisms to obtain survivable mappings and reduce energy consumption, congestion, or other operational objectives.Feasible solutions are reported for large topologies where integer-linear-programming methods cannot obtain them, and remembered successful solutions improve performance.
- Access and dynamic network control: AI also supports traffic-aware access-network control, with SDN-informed LTE uplink-downlink configuration improving packet latency and jitter.The mechanism calculates an optimal configuration from traffic dynamics across an XG-PON and LTE-radio-access fronthaul.
5. New Opportunities and Challenges for the Use of AI in Optical Networks
The paper identifies opportunities for AI in optical transmission, attack detection, network-management automation, joint network–computing operation, and on-chip networking, while noting unresolved complexity and emerging applications.
- 5.1. Optical Transmission Systems, and Attack and Intrusion Detection: AI can support emerging transmission technologies, quality-of-transmission estimation, performance monitoring, and attack or intrusion detection in optical networks.Attack detection and localization are identified as a significant opportunity that had not yet emerged in the optical arena to the authors’ knowledge.
- 5.1. Optical Transmission Systems, and Attack and Intrusion Detection: Statistical attack-identification methods face non-trivial computational complexity in large-scale networks, motivating AI-based detection and localization.The proposed statistical approaches use optical measurements to identify and localize attacks.
- 5.2. Automating Network Management Operations: AI-based techniques can help automate optical network management by analyzing telemetry across heterogeneous, multi-vendor devices, although open issues remain.Streaming telemetry introduced by SDN can support efficient information collection throughout the network.
- 5.2. Automating Network Management Operations: AI can optimize in-operation routing and resource-allocation problems and may enable preemptive service relocation away from predicted network failures.The cited applications include RWA, RMLSA, and RMCSA optimization and network reconfiguration.
- 5.3. Efficient Joint Operation of Networks and Computing Resources: AI is expected to facilitate joint operation of network and computing devices for VNF distribution, task allocation, predictive caching, and human-action prediction in IoT and tactile-Internet applications.The paper describes these applications as supporting the performance requirements of emerging networked services.
- 5.4. Applications in On-Chip Networks: AI-enabled optical networks-on-chip are presented as an alternative to electronic networks-on-chip that could reduce power consumption and computation time.The discussion connects this opportunity to the computational demands of AI algorithms and emerging AI-oriented chip technologies.
6. Summary
The paper surveys AI techniques and their applications across optical transmission and network design and control. It concludes that AI is well suited to complex, dynamic optical networks and highlights future roles in advanced transmission, security, automation, computing-resource coordination, and on-chip networking.
- 6. Summary: The survey classifies optical-networking AI literature into search and optimization, game theory, knowledge-based reasoning and planning, statistical models, decision-making, and learning methods.Figure 1 organizes the reviewed references according to these AI subfields and techniques.
- 6. Summary: The paper reviews AI applications for optical transmission and for the design and control of optical networks.Transmission topics include component characterization and operation, monitoring, nonlinearities, and QoT estimation; network topics include planning and connection establishment.
- 6. Summary: AI techniques have often been more efficient than classical approaches whose deterministic or semi-analytical models are limited by complexity, adaptability, or scalability.The comparison is made in the context of impairment-aware optical network operation and related control and design issues.
- 6. Summary: AI can find optimal or near-optimal solutions in highly complex, data-intensive scenarios without exhaustive analytical or semi-analytical models.The paper connects this capability to the increased complexity and dynamism of current optical communication networks.
- 6. Summary: The paper identifies future opportunities in advanced transmission, attack and intrusion detection, heterogeneous-network management automation, joint network–computing operation, and on-chip networking.Attack and intrusion detection is described as not yet significantly explored in optical networks to the authors’ knowledge.