چكيده به لاتين
One way to increase network capacity is to more accurately estimate current network performance and exploit it with lower margins. To achieve the benefits of reducing margins, the following elements must be considered: better physical models, per-light-path power management, optical performance monitoring, online network re-optimization and an adapted control plane. In this regard, to reduce margins and adapt more quickly to current network performance in dynamic applications, it is necessary to improve the accuracy and reduce runtime in the implementation of tools. In general, all of the mentioned elements are NP-complete problems and therefore the use of machine learning methods appears rational and efficient.
Accurate QoT estimation is necessary to minimize margins and optimize the network. Indeed, to quickly provision a new lightpath or to reroute an existing one in reaction to a failure, accurate and fast QoT evaluation is required. To this aim, tools are provided utilizing RBF, GRNN and PNN neural networks for estimating the QoT of unestablished lightpaths, which results in 99.6% accuracy. On the other hand, modular neural networks were utilized to take into account the launch power and modulation format of neighboring channels. Therefore, the network can operate closer to the real conditions and consequently its costs are significantly reduced. To fully leverage system margins stemming from network loading (i.e., from nonlinear impairments), careful power allocation is required. Therefore, the CEGA algorithm is proposed to allocate the optimal power to optical channels, which shows a significant reduction in the runtime of the algorithm compared to the convex approach (by several orders of magnitude). This method is suitable for both static as well as time-critical dynamic network planning with fast convergence requirement. In general, the RWPA (routing, wavelength, power allocation) problem in WDM networks is an NP-complete problem, however suboptimal solutions could be derived by decomposing this problem into sub problems of routing, wavelength assignment and power optimization. However, this method of solution, which is used to reduce computational complexity, can lead to inefficient resource allocation. To jointly optimize the network resources, the CEDE algorithm is proposed. Besides improving the network achievable rate by 5%, this algorithm is able to significantly reduce the runtime (by several orders of magnitude). In addition to the addressed elements, the information obtained from the optical performance monitoring equipment can be used to further reduce the design margins. Moreover, this information can be used in the field of lightpath management, such as failure localization. Therefore, a solution is proposed using a genetic algorithm for localizing the multi-failures. This method can localize the failures within only a few tens of milliseconds and with an accuracy of almost 100%. The proposed QoT estimation tools are very useful in reducing the required time for localizing failures.
Designing low-margin optical networks requires real-time solutions to estimate network performance. The results show that the proposed tools are able to be an efficient solution to reduce network margins by accurately estimating, optimizing and subsequently improving network performance. Moreover, with a significant reduction in computational complexity, the proposed tools can be utilized in the design of large-scale dynamic optical networks.