Time-Sensitive Networking (TSN) is expected to become a key layer 2 technology for 5G and Beyond (5GB) transport networks (TN) as it allows for services with stringent and deterministic quality-of-service constraints and their coexistence with non-performance-sensitive traffic. Autonomous solutions for configuring TSN-based TNs are essential to ensure the deterministic QoS requisites of the 5GB streams while facilitating the zero-touch management of the network and reducing the operational costs. However, due to the configuration flexibility offered by TSN networks, using exact optimization methods to develop such solutions usually results in algorithms with high computational complexity. In this work, we propose and evaluate an initial design of a Reinforcement Learning (RL)-based solution for the long-term configuration of asynchronous TSN-based 5GB TNs. We successfully validated the proper operation of the proposal for an industrial private 5G scenario.

This paper has been presented in XXXVII Symposium of the International Union of Radio Science (URSI) 2022 celebrated in Malaga (Spain) from 5th to 7th September. This work includes a preliminary design and proof-of-concept to apply Reinforcement Learning for computing long-term configurations in asynchronous Time-Sensitive Networking (TSN)-based 5G and beyond transport networks. Most of this work has been carried out within the European 5G-CLARITY project (https://www.5gclarity.com/).

H2020 research and innovation project 5G-CLARITY (Grant No. 871428)

6G-CHRONOS (TSI-063000-2021-28)

TRUE5G (PID2019- 108713RB-C53)