One of the main features of the 5G system architecture is network slicing, which allows the creation of multiple end-to-end logical networks on top of the same physical infrastructure, so that each slice can be optimised to the requirements of specific service and application domains. The management of slices in the Radio Access Network (RAN) requires capacity sharing mechanisms to dynamically modify the amount of radio resources allocated to each RAN slice in each cell ensuring both the fulfilment of the RAN slice requirements and an efficient use of the radio resources.
This is the problem addressed by the Proof Of concept of a Radio access neTwoRk slicing solution based on Artificial InTelligence (PORTRAIT) Project, whose main objective is to carry out the proof of concept of a deep learning solution for RAN slicing in an industrially relevant environment, paving the way for market exploitation and technology transfer. In this direction, PORTRAIT takes as starting point a Multi-Agent Reinforcement Learning (MARL) solution based on the Deep Q-Network (DQN) technique for RAN slicing management developed in the context of previous project SONAR-5G, where the solution was formulated, extensively assessed by means of simulations and where the functional framework and the information models for its implementation in the 3GPP/O-RAN ecosystem were delineated. Then, PORTRAIT intends to perform the proof of concept (PoC) of this solution in the context of the on-going 5GCAT pilot for testing the 5G technology in different scenarios and specifically in a use case on network capacity management for touristic beach environments. This use case is deploying a neutral host network for sharing resources among different operators and thus provides a perfect fit for the PoC demonstration of the DQN-MARL RAN slicing solution. PORTRAIT has designed an implementation plan that includes different stages: (1) Definition of the PoC architecture that enables the demonstration of the solution in the operational environment of the 5GCAT pilot. (2) Implementation of the solution in the PoC architecture, including the different modules, i.e. trainer and RAN slice manager, the interfaces for internal and external communication and the dashboard with the graphical user interface to configure and monitor the solution. (3) Training of the solution with datasets that will be built by means of real network measurements extracted from the components of the 5GCAT pilot as well as from a real local 5G deployment. (4) PoC demonstrations in the operational environment. (5) Exploitation, technology transfer and entrepreneurship activities, in which the project will define and execute a plan for the industrial exploitation of the solution and will undertake activities to foster entrepreneurship. In this respect, PORTRAIT has conducted a preliminary analysis of go-to-market strategies and value chains and has established contacts with relevant stakeholders. Besides, PORTRAIT momentum is very well sustained as long as network slicing is seen as a 5G business accelerator and artificial intelligence ecosystem is boosting.