MANAGEMENT OF TELECOMMUNICATION NETWORKS USING AI/ML TECHNOLOGIES

Abstract

The paper presents the results of research into the current state and new trends in the use of AI/ML in the management of networks and services. The basic technologies that are fundamental for the end-to-end creation of resources and their management are considered. With the help of such technologies, 5G networks can automate resource processing and isolation between logical and physical resources. In this sense, the latest approaches in the RAN domain with an open radio access network (O-RAN) are considered. It also focuses on the relevant controllers based on the European Telecommunications Standards Institute Management and Orchestration (ETSI MANO) stack, and in particular on how the computing resources are handled. This is followed by a review of fundamental ideas and recent trends in data plane programming at the packet level and at the optical switch level, focusing on the transport of network resource management. It also describes the creation of a data pipeline to provide a management stack for net-work monitoring and analytics.

The main approaches to the management of various architectural components of the 5G network have been studied. The review notes that the diversity of problems at each architectural level naturally requires different approaches. RAN resource management by detecting user behavior patterns and RAN resource allocation using reinforcement learning (RL) and Q-learning techniques are also considered. Traditional optimization methods and approaches based on artificial intelligence for solving network orchestration problems containing any computing resources are analyzed. SoA methods for achieving slicing of the data plane through isolation of coexisting traffic are given. The extremely important issue of pooling resources and AI methods used to facilitate the implementation of pooling procedures were also considered.

The peculiarities of the application of machine learning methods in the network platform, as well as the management of 5G software networks based on ML and QoE evaluation, are considered. This is followed by case studies covering QoE management, VNF deployment, and slice management. The main challenges are identified, how they can be solved and instructions for operators are suggested. ML algorithms used for power management, user scheduling, user association, and spectrum allocation are reviewed.