Optical networks are typically designed with several types of margins, each of which impacts the trade-off between network cost and network availability (e.g., robustness to unforeseen events). This paper presents a system vendor’s approach to this trade-off, with the goal of maximizing the useful system bandwidth while still meeting the required quality of service. Tracking, predicting, and reacting to the quality of transmission in a synergistic bottom-up and top-down approach is the central theme. This includes physical-layer considerations such as intelligent use of line-system monitors, modeling the time variation of various noise components, and dynamic, rate-flexible modems with advanced constellation selection. The control layer relies on machine learning techniques, including reward-based reinforcement learning, to extract useful information from the vast amount of collected data and trigger appropriate measures in response. To better adapt to the increased vulnerability resulting from margin reduction, multilayer control is advocated. Overall, the authors believe that near-zero margin networking is possible, and in fact is necessary, to meet the ongoing pressure to reduce network cost.
You must log in
to add comments.