Efficient Laser Frequency Allocation in Packet-Optical Nodes with Coherent Transceivers
Constantine A. Kyriakopoulos
TL;DR
The paper addresses efficient laser frequency allocation for coherent transceivers in edge packet-optical whiteboxes within ROADM metros. It implements a telemetry-driven reinforcement learning approach, using a Q-Learning-based module within the PacketCTL to select frequency slots based on CMIS feedback, with Netconf/YANG-enabled agents and RedisDB coordination. The results show a 20-25% reduction in the average laser configuration time and demonstrate convergence with ~3k training episodes and scalability up to 16 pluggables. This work provides a practical, vendor-independent path to faster, more efficient control planes for next-generation transport networks, with relevance to 6G fronthaul/backhaul needs.
Abstract
The introduction of silicon chipsets with the capability of processing incoming optical packet traffic, creates a new generation of packet-optical nodes, the whiteboxes. Their inherent functionality of carrying pluggable coherent transceiver modules, extends their scope in the field of transport optical networks. Also, their hybrid nature improves the overall efficiency since higher layer functionality is now performed at wire speed. This is feasible by embedding in the computational logic, apart from the typical packet inspection routines and security features, Machine Learning-based traffic engineering as well. In this work, a topology based on multiple packet-optical nodes residing at the edges (ingress and egress) of a ROADM network, is evaluated according to the average laser frequency configuration time. This is achieved by exploiting telemetry analytics which are collected by the CMIS driver of the pluggable transceivers, for allocating efficient frequency slots to the source and destination of connectivity requests traversing through the ROADM network. This pair of nodes is supervised by the packet SDN controller which is part of the control plane of the optical transport network. This controller receives telemetry feedback from the whiteboxes and uses it to execute efficient ML techniques locally, for finding efficient frequency slots for the incoming transport requests. Next, it applies them to request's edge nodes. The decrease of the average laser configuration time is achieved in the evaluated topology, improving the overall efficiency of the control plane.