Artificial Intelligence for 5G Networks

Thursday, 20 June 2019, 14:00-17:30, room R6
Organisers:
  • David M. Gutierrez Estevez (Samsung R&D Institute UK)
  • Yue Wang (Samsung Electronics, Uk)
  • Anastasius Gavras (Eurescom GmbH, Germany)
  • Jose M. Alcaraz Calero (University of the West of Scotland, UK)

 

Motivation and Background

The use of AI for network operation and management is known to have great potential to enhance the network performance and efficiency, therefore has received significant interest in both research and industry standardization groups. Prominent industry standard groups, such as ETSI ISG ENI and ZSM, and 3GPP SA2 and SA5, have made great efforts to build industry consensus on network architecture and interfaces to enable the use of AI in the network. However, the standardization on AI in network management and orchestration is just at its beginning. Open questions on AI for network management need joint efforts from different standardization groups, as well as between the research and industry standardization groups.

Likewise, the topic is having a big impact in the research communities, and 5G-PPP projects are also part of this trend. In this workshop, 5G-MoNArch will showcase recent results on the employment of AI to achieve resource elasticity, i.e., an efficient and autonomous utilization of computational resources in the network, by enhancing the design of VNFs and their scaling mechanisms. In SliceNet, AI is investigated to achieve cognitive network management to improve both operation experience for network operators and quality of experience for vertical users, especially in the context of network slicing. However, traditionally it has been difficult for research and innovation projects to make significant impact on standards in a timely manner.

The workshop aims to bridge the gap between research and standards on AI for network management and orchestration. It will bring together researchers and industry experts, and stimulate debate and discussions on what are the significant problems to resolve to fully explore the potential of AI in networks, and how the relevant results provided by 5GPPP projects can be leveraged, therefore maximize their impacts to standards.

 

Structure

SESSION I

Chair: Anastasius Gavras, EURESCOM
14:00 – 14:30 Keynote Speaker  Hans Schotten, German Research Center for Artificial Intelligence DFKI & University of Kaiserslautern, Germany Title: “AI as Key Enabler for 5G Campus Networks”

14:30 – 15:45 5G-PPP Research Activities (5G-MoNArch & SliceNet)  5G-MoNArch#1: Marco Gramaglia, Universidad Carlos III de Madrid, Spain Title: “Bringing Elasticity into 5G: From VNF Operation to AI-based Orchestration”

 5G-MoNArch#2: Asterios Mpatziakas, Centre for Research and Technology Hellas, Greece Title: “Slice-aware Resource Orchestration of an Elastic 5G Network via Evolutionary Algorithms”

 SliceNet#1: Jose Alcaraz-Calero, University of the West Scotland, UK
Title: “Self-Restoring Video User Experience in 5G Networks Based on a Cognitive Network Management Framework”

 SliceNet#2: Salvatore Spadaro, Universitat Politecnica de Catalunya, Spain Title: “A QoE-oriented Cognition-based Management System for 5G Slices: The SliceNet Approach”

SESSION II

Chair: David Gutierrez, Samsung R&D Institute UK
16:00 – 17:30 Standardisation Activities  Ray Forbes, ETSI ENI, Huawei, UK Title: “AI automation in ENI and ZSM”

 Diego Lopez, ETSI ZSM, Telefonica, Spain Title: “Standardising closed-loop network management: from AI integration to data-flow provenance”

 Alessandro Trogolo, 3GPP, Telecom Italia, Italy Title: “3GPP Analytic Functions for 5G Networks”

 Slawomir Stanczak, ITU-T FG-ML5G, Fraunhofer HHI, Germany Title: “Challenges and Research Directions in Machine Learning for 5G and Beyond”

Panel discussion with keynote and Session II speakers

TALKS ABSTRACTS

AI as Key Enabler for 5G Campus Networks 5G campus networks allow vertical industries to benefit from the advantages of 5G technology independent of public mobile networks. In particular the very challenging requriements of the automation industry on performance and capabilities of the mobile infrastructure can probably easier be addessed in these network option. Here, AI is seen as key enabler to address the complexity and versatility of tasks to be addressed. This talk will review AI related challenges and solutions.

Bringing Elasticity into 5G: From VNF Operation to AI-based Orchestration Operating a 5G (and beyond-5G) system will be substantially more complex than it used to be in previous generations of mobile networks. To full exploit the advanced features of dynamic reconfiguration introduced by network virtualization technologies in a cost efficient way, network operators shall employ elastic algorithms that can scale resources according to the real demand. In this extended abstract we first introduce the concept of network elasticity and then describe two algorithms for the elastic network operator: (i) CARES, a computational-aware radio scheduler and (ii) DeepCog, a deep learning algorithm for resource orchestration in a multi slice network.

Slice-aware Resource Orchestration of an Elastic 5G Network via Evolutionary Algorithms Fifth generation mobile networks (5G) will enable new use cases for industries and vertical markets via numerous innovative approaches that overcome limitations of existing systems. Two concepts essential for the realization of 5G, are network elasticity and slicing. The application of these concepts allows the simultaneous hosting of more services that use a common resource pool while reducing operation and capital expenses. The realization of these benefits demands an efficient method to scale
the resources allocated between different slices often with diverse demands. We apply a multiobjective approach, based on evolutionary algorithms to accomplish optimized resource orchestration between cloud-based slices in a 5G network deployed over a large European city. Numerical results are provided for the proposed approach and are compared to other allocation schemes.

Self-Restoring Video User Experience in 5G Networks Based on a Cognitive Network Management Framework Video applications such as streaming are expected to dominate the traffic of the incoming Fifth generation (5G) networks. It is essential for 5G service video providers and/or network operators to provide assurances for both the overall status of the network and the quality of their video transmissions in order to meet the final users’ expectations. In this contribution, we propose a video optimisation scheme which is implemented as a Virtualised Network Function (VNF), which in turn, facilitates its on-demand deployment in a flexible way in response to an intelligent analysis of the current network traffic conditions. We leverage a cognitive network management framework to analyse both network status metrics and video stream requirements to evaluate if any optimisation action is required. The testing and evaluation focus on the functional tests and scalability evaluation of the proposed scheme. Moreover, the bandwidth saving is assessed to demonstrate the significant benefit in traffic reduction for a 5G system that adopts the proposed approach.

A QoE-oriented Cognition-based Management System for 5G Slices: The SliceNet Approach Provisioning of network slices with appropriate Quality of Experience (QoE) guarantees is one of the key enablers for 5G networks. However, it poses several challenges in the slice management that need to be addressed to achieve an efficient end-to-end (E2E) services delivery. These challenges, among others, include the estimation of QoE Key Performance Indicators (KPIs) from monitored metrics and the corresponding reconfiguration operations (actuations) in order to support and maintain the desired quality levels. In this context, SliceNet provides a design and an implementation of a cognitive slice management framework that leverages machine learning (ML) techniques in order to proactively maintain network conditions in the required state that assures E2E QoE, as perceived by the vertical customers.

AI automation in ENI and ZSM The presentation will explore the latest work in ETSI ISGs on AI and Automation, Covering the Coordination strategy of Zero Weight. Self-Healing, optimization and towards Zero Touch. It will unpack the Architectures and management issues. And how and where AI may be used to benefit the systems.

Standardising closed-loop network management: from AI integration to data-flow provenance One of the key aspects to address the highly distributed, heterogeneous and diverse nature of network infrastructures is the availability of open standards to facilitate interoperability at all levels. The applicability of autonomous management methods based on evidence (AI, analytics, etc) requires a certain degree of standardization, with the additional challenge of achieving it up to the right detail, avoiding the ossification of architectures while guaranteeing openness and interoperability. This talk will present the most salient initiatives in this direction, including ETSI ZSM and ENI ISGs and the work of the NMRG within IETF/IRTF.

3GPP Analytic Functions for 5G Networks The new 3GPP 5G architecture for mobile networks foresees, according to a service based approach, a flexible way to build up a management system that can be tailored to the specific management needs. As a part of this flexible architecture based on Management Services, new features can be introduced to improve the quality of the automation and orchestration toward an autonomous management system (zero touch concept). Intent drive requirements, Network data analytics, Management data analytics and Self Organizing Network (SON) shall be used to enhance and optimize the management of the network and the slices.
Challenges and Research Directions in Machine Learning for 5G and Beyond Wireless communications poses some fundamental challenges to machine learning (ML). Due to high mobility, wireless links exhibit ephemeral and highly dynamic nature; moreover, the links are corrupted by noise and are in general exposed to interference, while wireless resources (spectrum and energy) are scarce. All this may greatly limit the capacity of wireless networks. Since data (including measurement data) is not available at a single point but distributed among different locations, there is a strong need for distributed ML algorithms that efficiently use the scarce wireless resources in 5G networks. Moreover, the new ML methods need to provide robust results based on small data sets and under latency constraints. In this talk, we will discuss why ML has the potential for substantially enhancing current radio access networks (RAN) solutions, and we will give an overview of promising techniques and algorithmic approaches for overcoming the above mentioned challenges in 5G networks. We argue in favor of hybrid-driven methods that combine purely data-driven ML approaches with classical model-based methods, while making use of the available expert knowledge. In particular, it is in general necessary to exploit structures in the wireless channel, transmission signals and various functions such as load maps. We will consider both kernel-based methods and deep neural networks. Meeting the latency requirements of 5G networks requires massive parallelization. We’ll discuss how to parallelize certain radio access network (RAN) functions to GPU architectures to achieve orders-of-magnitude acceleration.