Page 123 - ITU Journal, Future and evolving technologies - Volume 1 (2020), Issue 1, Inaugural issue
P. 123
ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1
MACHINE LEARNING-ASSISTED CROSS-SLICE RADIO RESOURCE OPTIMIZATION:
IMPLEMENTATION FRAMEWORK AND ALGORITHMIC SOLUTION
Ramon Ferrús, Jordi Pérez-Romero, Oriol Sallent, Irene Vilà, Ramon Agustí
Dept. of Signal Theory and Communications, Universitat Politècnica de Catalunya (UPC), c/ Jordi Girona, 1-3, Barcelona, Spain
NOTE: Corresponding author: Ramon Ferrús (ferrus@tsc.upc.edu)
Abstract – Network slicing is a central feature in 5G and beyond systems to allow operators to customize
their networks for different applications and customers. With network slicing, different logical networks, i.e.
network slices, with specific functional and performance requirements can be created over the same
physical network. A key challenge associated with the exploitation of the network slicing feature is how to
efficiently allocate underlying network resources, especially radio resources, to cope with the spatio-
temporal traffic variability while ensuring that network slices can be provisioned and assured within the
boundaries of Service Level Agreements / Service Level Specifications (SLAs/SLSs) with customers. In this
field, the use of artificial intelligence, and, specifically, Machine Learning (ML) techniques, has arisen as a
promising approach to cater for the complexity of resource allocation optimization among network slices.
This paper tackles the description of a feasible implementation framework for deploying ML-assisted
solutions for cross-slice radio resource optimization that builds upon the work conducted by 3GPP and O-
RAN Alliance. On this basis, the paper also describes and evaluates an ML-assisted solution that uses a Multi-
Agent Reinforcement Learning (MARL) approach based on the Deep Q-Network (DQN) technique and fits
within the presented implementation framework.
Keywords – 5G, cross-slice resource optimization, deep learning, machine learning, network slicing
resources efficiently [7]-[14]. Remarkably, the
1. INTRODUCTION
automation of the life-cycle management of
Network slicing allows operators to customize their network slices in the RAN requires two main
networks for different applications and customers functionalities: slice admission control and cross-
[1], [2]. Slices can differ in functionality (e.g. air slice resource optimization.
interface capabilities, mobility tracking features), in Slice admission control is needed to decide on the
performance requirements (e.g. latency, availability, acceptance or rejection of a new RAN slice creation
reliability and data rates), or they can serve only request with specific coverage, functional (i.e.
specific users (e.g. public safety users, corporate features) and performance (e.g. service quality,
customers, or industrial users). A network slice can capacity) requirements. Under Network as a Service
provide the functionality of a complete network, (NaaS) business models such as neutral host
including radio access network and core network services, the slice requirements will be determined
functions. Support for network slicing has been by the Service Level Agreement (SLA) / Service
introduced by the 3rd Generation Partnership Level Specifications (SLS) established between the
Project (3GPP) as part of the first release of the Fifth service provider (e.g. the operator of a RAN
Generation (5G) system specifications (Release 15), infrastructure installed in a venue) and the
with multiple enhancements still to follow in future customer (e.g. a Mobile Network Operator - MNO).
releases, as reflected by different study items in The fulfillment of the RAN slice requirements may
progress, such as [3]-[6].
result in the need to guarantee the availability of a
The creation and management of network slices is certain amount of radio resources to the new slice,
especially challenging in the Radio Access Network defined in terms of, e.g. number of Resource Blocks
(RAN), where multiple slices can be delivered over (RBs) per cell, percentage of cell capacity, etc.
the same radio channel and the system shall Therefore, the slice admission control shall estimate
guarantee that the allocation and distribution of the the amount of radio resources required by the new
radio resources within the radio channel is done so slice and decide whether this can be enforced given the
that specific requirements per slice can be fulfilled deployed network capacity and the amount of
(e.g. guaranteed capacity) while using radio resources consumed by the already admitted slices.
© International Telecommunication Union, 2020 103