Machine Learning aided design and planning of optical networks

This research activity investigates the benefits brought by the introduction of distance adaptive transceivers supporting multiple modulation formats and baud rates in flexible grid networks in terms of spectrum occupation and transceiver utilization. Methodological approaches for traffic deployment over optical ring metro networks and mesh backbone networks relying on Integer Linear Programming, as well as heuristics and analytical bounds for near-to-the-optimum solutions, are being investigated.

Moreover, the impact of usage of spatial division multiplexing by exploiting multimode/multicore optical fibers is under evaluation, also in the context of ultra-wide-band optical transmission.

Furthermore, machine learning approaches for quality of transmission estimation are being researched, with a specific focus on the adoption of explainable artificial intelligence techniques. Other domains of application of machine learning algorithms in the context of automated optical networks management are traffic prediction and fault detection.

ERC Sector:

  • PE7_8 Networks (communication networks, networks of sensors, robots...)


  • Optical fiber networks
  • WDM networks routing and spectrum assignment
  • Automated optical network management