Improving data center optical networks with cross-layer machine learning

With the fast growth of 5G services and Internet of Things (IoT) applications, there is a critical need for creative solutions to handle the increasing backbone network traffic. This research presents a new method for machine-learning-based wavelength-routing networks with interactive transmission to improve optical connection smoothness in future data centers. Our method features a versatile transmitter capable of broadcasting at multiple coding rates (N), dynamically adjusting to connections based on expected Bit Error Rate (BER) levels. By employing a neural network (NN)-based classifier to pre-process data and classify BER, we demonstrate significant improvements in classification accuracy across various values of N and switching connections. This approach facilitates the development of adaptive and efficient optical data center networking, ultimately aiming to boost network performance and reliability.