CIAN 2016 Annual Report CIAN 2016 Annual Report | Page 15

Machine Learning to mitigate optical physical layer impairments Next generation of optical networks will be more dynamic in WDM bandwidth assignment. Rapid addition and deletion of channels presents a challenge to the power stability of the network. In a lightpath with cascaded Erbium Doped Fiber Amplifiers (EDFA), power excursions of each amplifier during channel configurations may accumulate. We introduced a machine learning technique for optimized assignment of channels that reduces the impact of power divergence in EDFAs. Power divergence during rapid channel provisioning is characterized from historical operation records of the network. Using machine intelligence, we can identify the channel dependence of power divergence. Industrial Collaborations Lightwave Research Laboratory has a wide-range of industrial partners, ranging from equipment vendors to network operators. Valuing the importance of applied research, we join their R&D activities for their current and next-generation products. Calient Technologies and Polatis (optical switch vendors) and AT&T are examples of our industrial partners in the optical network research activities. Dr. Keren Bergman Research Professor Dr. Payman Samadi Research Scientist Yishen Huang Ph.D. Candidate Yiwen Shen Ph.D. Candidate www.cian-erc.org 15