Download - Deliverable D2.1 Description and definition of targeted PoCs

MARSAL — H2020-ICT-2020-2

Deliverable D2.1 Description and definition of targeted PoCs

Document Summary Information

Grant Agreement No 101017171 Acronym MARSAL

Full Title Machine Learning-based, Networking and Computing Infrastructure Resource Management of 5G and Beyond Intelligent Networks

Start Date 01/01/2021 Duration 36 months

Project URL https://www.marsalproject.eu/

Deliverable D2.1 – Description and definition of targeted PoCs

Work Package WP2

Contractual due date M8 Actual submission date 10/09/2021

Nature Report Dissemination Level Public

Lead Beneficiary OTE

Responsible Author Alexandros Kostopoulos (OTE)

Contributions from All partners

MARSAL — H2020-ICT-2020-2 Deliverable D2.1 – Description and definition of targeted PoCs [Public]

Revision history

Version Issue Date Changes Contributor(s)

V0.1 01/05/2021 Initial ToC Alexandros Kostopoulos (OTE)

V0.2 01/06/2021 Contributions in Section 2 John Vardakas (IQU)

V0.3 15/06/2021 Contributions in Section 3 Md Arifur Rahman (ISW), Evgenii Vinogradov (KUL), Sabrina De Capitani di Vimercati (UNIMI), Pierangela Samarati (UNIMI), Sergio Barrachina (CTTC), Roberto Gonzalez (NEC), Miquel Payaro (CTTC)

V0.4 07/07/2021 Contributions in Section 4 Simon Pryor (ACC), Charalambos Klitis (eBOS), Sergio Barrachina (CTTC), Md Arifur Rahman (ISW), Roberto Gonzalez (NEC), Giuseppe Siracusano (NEC), Emmanouel Varvarigos (ICCS), Panagiotis Kokkinos (ICCS), Polyzois Soumplis (ICCS), Philippe Chanclou (ORANGE), Alexandros Kostopoulos (OTE), Idan Yehosua Barnea (NVIDIA), Kostas Chartsias (ICOM), Dimitrios Kritharidis(ICOM), Evangelos Pikasis (ICOM)

V0.5 15/07/2021 Contributions in Section 5 Evgenii Vinogradov (KUL), Md Arifur Rahman (ISW), Philippe Chanclou (ORANGE), Sergio Barrachina (CTTC), Miquel Payaro (CTTC), Roberto Gonzalez (NEC), Charalambos Klitis (eBOS), Kostas Chartsias (ICOM),


Dimitrios Kritharidis(ICOM), Evangelos Pikasis (ICOM), Sabrina De Capitani di Vimercati (UNIMI), Pierangela Samarati (UNIMI)

V0.6 21/07/2021 Contributions in Section 6 and document integration

Alexandros Kostopoulos (OTE)

V0.7 03/08/2021 First review iteration Miquel Payaro (CTTC)

V0.8 10/08/2021 Addressing comments All partners

V0.9 17/08/2021 Second review iteration Christos Verikoukis (CTTC), John Vardakas (IQU)

V1.0 10/09/2021 QA and submission to EU Ioannis Chochliouros (OTE)

Disclaimer

The content of the publication herein is the sole responsibility of the publishers, and it does not necessarily represent the views expressed by the European Commission or its services.

While the information contained in the documents is believed to be accurate, the authors(s) or any other participant in the MARSAL consortium make no warranty of any kind with regard to this material including, but not limited to the implied warranties of merchantability and fitness for a particular purpose.

Neither the MARSAL Consortium nor any of its members, their officers, employees or agents shall be responsible or liable in negligence or otherwise howsoever in respect of any inaccuracy or omission herein.

Without derogating from the generality of the foregoing neither the MARSAL Consortium nor any of its members, their officers, employees or agents shall be liable for any direct or indirect or consequential loss or damage caused by or arising from any information advice or inaccuracy or omission herein.

Copyright message

© MARSAL Consortium, 2021-2023. This deliverable contains original unpublished work except where clearly indicated otherwise. Acknowledgement of previously published material and of the work of others has been made through appropriate citation, quotation or both. Reproduction is authorised provided the source is acknowledged.


CONTENTS

List of Acronyms and Abbreviations .............................................................................................................. 7

List of Figures ..............................................................................................................................................15

List of Tables ...............................................................................................................................................16

Executive summary .....................................................................................................................................17

1 Introduction .........................................................................................................................................19

2 MARSAL Overall Architectural Approach ...............................................................................................21

3 Technological Enablers .........................................................................................................................25

3.1 Cell-free MIMO networks .............................................................................................................25

3.2 Fronthauling technologies for 6G networks ..................................................................................26

3.3 Multi-objective optimization towards self-Driven, elastic Infrastructures .....................................26

3.4 Data security and privacy in multi-tenant infrastructures .............................................................27

3.5 Distributed ML for privacy and network security...........................................................................28

4 5G Use Cases and Verticals Overview ....................................................................................................29

4.1 3GPP use cases .............................................................................................................................29

4.1.1 3GPP Use Case Goals and Drivers ..............................................................................................29

4.1.2 3GPP UC1: Intra DU (Cluster) Cell-Free, Single UE .....................................................................29

4.1.3 3GPP UC2: Intra DU (Cluster) Cell-Free, Multiple UE ..................................................................29

4.1.4 3GPP UC3: Inter DU (Cluster) Cell-Free, Single UE .....................................................................30

4.1.5 3GPP UC4: Inter DU (Cluster) Cell-Free, Multiple UE, Mobility ...................................................30

4.1.6 3GPP UC5: Inter CU (Region) Cell-Free, Single UE ......................................................................30

4.1.7 3GPP UC6: Inter CU (Region) Cell-Free, Multiple UE, Mobility ...................................................31

4.1.8 3GPP UC7: INTRA/INTER NG-RAN mobility – cell free BACKWARD COMPATIBILITY ....................31

4.2 O-RAN use cases ...........................................................................................................................31

4.2.1 Use case 1: Context-based Dynamic Handover Optimization Management for V2X ...................32

4.2.2 Use case 2: Flight Path-based Dynamic UAV Radio Resource Allocation .....................................32

4.2.3 Use case 3: RRA for UAV Application .........................................................................................33

4.2.4 Use case 4: QoE Optimization ...................................................................................................33

4.2.5 Use case 5: Traffic Steering .......................................................................................................33

4.2.6 Use case 6: mMIMO Beamforming Optimization .......................................................................34

4.2.7 Use case 7: RAN Sharing ............................................................................................................34

4.2.8 Use case 8: QoS-based Resource Optimization ..........................................................................35

4.2.9 Use case 9: RAN slicing SLA assurance .......................................................................................35


4.2.10 Innovations Beyond O-RAN Uses Cases ..................................................................................35

4.3 ITU-R use case domains ................................................................................................................36

4.4 Use cases and verticals in 5G PPP projects ....................................................................................37

4.4.1 5G ERA ......................................................................................................................................37

4.4.2 5GMediaHub .............................................................................................................................37

4.4.3 VITAL-5G ...................................................................................................................................38

4.4.4 5G-PHOS ...................................................................................................................................38

4.4.5 5G-ROUTES ...............................................................................................................................38

4.4.6 MonB5G ....................................................................................................................................39

4.4.7 5G-SOLUTIONS ..........................................................................................................................39

4.4.8 5G-EPICENTRE ...........................................................................................................................40

4.4.9 6G BRAINS.................................................................................................................................40

4.4.10 AI@EDGE ...............................................................................................................................41

4.4.11 HEXA-X ..................................................................................................................................41

4.4.12 5G-TOURS .............................................................................................................................41

4.4.13 5G-MEDIA .............................................................................................................................42

4.4.14 5G-COMPLETE .......................................................................................................................43

4.5 Industrial initiatives ......................................................................................................................43

4.5.1 NGMN .......................................................................................................................................43

4.5.2 ATIS ..........................................................................................................................................45

4.5.3 5G-ACIA (for industrial domain only) .........................................................................................46

4.5.4 NEM (for media and content only) ............................................................................................47

5 MARSAL Use Cases Description .............................................................................................................48

5.1 MARSAL Testbeds .........................................................................................................................48

5.1.1 KUL cell-free MIMO testbed (Leuven, belgium) .........................................................................48

5.1.2 ORANGE experimental platform (lannion, france) .....................................................................49

5.1.3 CTTC experimental platform (barcelona, spain) .........................................................................50

5.2 PoC1: Cell-free networking in dense and ultra-dense hotspot areas ..............................................52

5.2.1 Experimentation Scenario 1.1: Dense User-Generated Content distribution with mmWave Fronthauling .........................................................................................................................................53

5.2.2 Experimentation Scenario 1.2: Ultra-dense video traffic delivery in a converged Fixed-Mobile network ................................................................................................................................................59

5.3 PoC2: Cognitive assistance and its security and privacy implications in 5G and Beyond.................64


5.3.1 Experimentation Scenario 2.1: Cognitive Assistance and Smart Connectivity for next-generation sightseeing ...........................................................................................................................................66

5.3.2 Experimentation Scenario 2.2: Data security and privacy in multi-tenant infrastructures ..........74

6 Conclusions ..........................................................................................................................................82

References ..................................................................................................................................................83


List of Acronyms and Abbreviations

Acronym Description

3GPP The Third Generation Partnership Project

4G The Fourth Generation of Mobile Communications

5G The Fifth Generation of Mobile Communications

5G ACIA 5G Alliance for Connected Industries and Automation

5G PPP 5G Public Private Partnership

5GVP 5G Vertical Enablement Platform

AI Artificial Intelligence

AP Access Point

API Application Programming Interface

APP Application

AR Augmented Reality

ATIS Alliance for Telecommunications Industry Solutions

B5G Beyond 5G

BBIC Baseband Integrated Circuit

BBU Baseband Unit

BFT Byzantine Fault Tolerance

BS Base Station

BVLOS Beyond Visual Line of Sight

BVNO Broadcast Virtual Network Operator

CAM Connected and Automated Mobility

CDN Content Delivery Network

CF Cell-Free


CITEL Inter-American Telecommunication Commission

CPU Central Processing Unit

C-RAN Cloud-RAN

CSI Channel State Information

CSP Communication Service Provider

CSMA Carrier Sense Multiple Access

CSMA-CA Carrier Sense Multiple Access protocol with Collision Avoidance

CU Central Unit

CUCP Central Unit-Control Plane

CU-UP Central Unit-User Plane

DAO Decentralized Autonomous Organization

DAS Distributed Antenna Systems

DL Download link

DC Data Center

DP Dynamic Programming

DRL Deep Reinforcement Learning

DU Distributed Unit

DSP Digital Signal Processing

DSRC Dedicated Short-Range Communications

E2E End-to-End

E-UTRAN Evolved Universal Terrestrial Radio Access Network

eMBB enhanced Mobile Broadband

EP Embedding Propagation


ETSI European Telecommunications Standards Institute

EU European Union

EV Electrical Vehicle

FiWi Fiber-Wireless

FMC Fixed Mobile Convergence

FoF Factories of the Future

FTTH Fiber to the Home

GDPR General Data Protection Regulation

gNB gNodeB

GNSS Global Navigation Satellite System

GPS Global Positioning System

GSM Global System for Mobile Communications

GUI Graphical User Interface

HTML Hypertext Markup Language

HW Hardware

ICT Information and Communication Technologies

IEEE Institute of Electrical and Electronic Engineers

IIoT Industrial Internet of Things

IMT International Mobile Communications

IoT Internet of Things

IoV Internet of Vehicles

IP Internet Protocol

ITS Intelligent Transport System

ITU International Telecommunication Union


ITU-R International Telecommunication Union – Radiocommunications Sector

ITU-T International Telecommunication Union – Telecommunication Standardization Sector

KPI Key Performance Indicator

LTE Long Term Evolution

M2M Machine-to-Machine

MAC Medium Access Control

MEC Mobile Edge Computing

MEO Multi-access Edge Orchestrator

MIMO Multi-Input Multi-Output

ML Machine-Learning

mMIMO Multi-Input Multi-Output

mMTC Massive Machine Type Communications

MMW Millimetre Wave

MNO Mobile Network Operator

MSP Media Service Provider

MVNO Mobile Virtual Network Operator

NEM New European Media

NFV Network Function Virtualisation

NFVO NFV Orchestrator

NGMN Next Generation Mobile Network

NGN Next Generation Network

NG-RAN Next Generation Radio Access Network

NO Network Operator


NP Network Performance

NPU Network Processing Unit

NR New Radio

NSA Non-Standalone

NSaaS Network Slicing As a Service

OAI Open Air Interface

OC Optical Carrier

ODL OpenDayLight

O-RAN Open Radio Access Network

OTT Over the Top

PII Personal Identifiable Information

PoC Proof of Concept

PtP Point-to-Point

PtMP Point-to-Multipoint

PHY Physical Layer

PPDR Public Protection and Disaster Relief

PPP Public-Private Partnership

PTP Precision Time Protocol

QoE Quality of Experience

QoS Quality of Service

RAN Radio Access Network

RE Radio Equipment

REC Radio Equipment Control

RFIC Radio-Frequency Integrated Circuit


RFID Radio-Frequency Identification

RF Radio Frequency

RRA Radio Interface and Resource Allocation

RRH Remote Radio Head

RSRP Reference Signal Received Power

RSRQ Reference Signal Received Quality

RU Radio Unit

SA Standalone

SC Small Cell

SDN Software Defined Network

SDTN Software Defined Transport Network

SINR Signal to Interference Noise Ratio

SLA Service Level Agreement

SMARTER Study on New Services and Markets Technology Enablers

SNR Signal-to-Noise Ratio

SoC System on Chip

SON Self-Organising Network

SoTA State of the Art

SP Service Performance

SPAN Services and Protocols for Advanced Networks

TAE Threat Analysis Engine

TDD Time Division Duplex

TDE Threat Detection Engine

TI Tele-Immersive


TMN Telecommunications Management Network

T&L Transport & Logistics

UAV Unmanned Aerial Vehicle

UC Use Case

UE User Equipment

UGC User Generated Content

UHD Ultra High Definition

UL Upload link

UMTS Universal Mobile Telecommunications System

URLLC Ultra-Reliable Low Latency Communications

USRP Universal Software Radio Peripherals

UTRAN Universal Terrestrial Radio Access Network

V2V Vehicle-to-Vehicle

V2X Vehicle-to-Everything

VIM Virtual Infrastructure Manager

VM Virtual Machine

VNF Virtual Network Function

VNI Virtual Network Instances

VR Virtual Reality

vRAN Virtualized RAN

WAVE Wireless Access in Vehicular Environments

WDM Wavelength-Division Multiplexing

Wi-Fi, WiFi Wireless Fidelity

WG Working Group


WP Work Package

XAI eXplainable Artificial Intelligence

XG-PON Gigabit Passive Optical Network


List of Figures

Figure 1: The MARSAL network architecture ................................................................................................21 Figure 2: O-RAN specified use cases for 5G networks [37] ...........................................................................32 Figure 3: KUL cell-free MIMO testbed ..........................................................................................................48 Figure 4: ORANGE experimental platform ....................................................................................................50 Figure 5: CTTC experimental platform .........................................................................................................51 Figure 6: Dense user-generated content distribution with mmWave fronthauling scenario .........................53 Figure 7: Overview of the PoC1 scenario mapped into the MARSAL architecture .........................................55 Figure 8: Mobile operation based on fixed operation facilities for large indoor CAMPUS (or stadium, mall, etc.) use-case ..............................................................................................................................................60 Figure 9: Experimentation set-up for the use case about content delivery in a converged Fixed-Mobile network using PtP & WDM for fronthaul and PtP & PON for mid/back-haul ................................................61 Figure 10: Mapping scenario with MARSAL architecture ..............................................................................62 Figure 11: Example of outdoors sightseeing [87] .........................................................................................67 Figure 12: Microsoft Hololen [88] ................................................................................................................68 Figure 13: Concept of the preliminary use case (experimentation scenario 2.1) ...........................................69 Figure 14: Overview of the experimentation scenario 2.1 mapped into the MARSAL architecture. ..............71 Figure 15: Mapping experimentation scenario 2.2 to MARSAL architecture .................................................76 Figure 16: Target scenario to measure the privacy vs. accuracy tradeoff .....................................................77 Figure 17: Example of malware detection from the sequence of hosts contacted ........................................81


List of Tables

Table 1: Preliminary KPI definition for the experimentation scenario 1.1 (Test scenario 1 & 2) ....................56 Table 2: Preliminary KPI definition for the experimentation scenario 1.1 (Test scenario 3) ..........................57 Table 3: Preliminary KPI definition for the experimentation scenario 1.1 (Test scenario 4) ..........................58 Table 4: Preliminary KPI definition for the experimentation scenario 1.1 (Test scenario 5) ..........................59 Table 5: Preliminary network requirements for the experimentation scenario 1.2. ......................................63 Table 6: Preliminary KPI definition for the experimentation scenario 1.2 (baseline scenario) .......................64 Table 7: Preliminary KPI definition for the experimentation scenario 2.1 (Test scenario 1) ..........................73 Table 8: Preliminary KPI definition for the experimentation scenario 2.1 (Test scenario 2) ..........................73 Table 9: Preliminary KPI definition for the experimentation scenario 2.1 (Test scenario 3) ..........................74 Table 10: Preliminary KPI definition for the experimentation scenario 2.2 (secure and private information sharing among tenants) ...............................................................................................................................78 Table 11: Preliminary KPI definition for the experimentation scenario 2.2 (blockchain-based smart-contracts platform for network slicing) .......................................................................................................................79 Table 12: Preliminary KPI definition for the experimentation scenario 2.2 (security and privacy for the data stored in the cloud) .....................................................................................................................................80 Table 13: Preliminary KPI definition for the experimentation scenario 2.2 (final users’ network security) ....81


Executive summary MARSAL proposes a new paradigm of elastic virtual infrastructures that integrate in a transparent manner a variety of novel radio access, networking, management and security technologies, which will be developed to deliver end-to-end transfer, processing and storage services in an efficient and secured way. It targets the development and evaluation of a complete framework for the management and orchestration of network resources in 5G and beyond, by utilizing a converged optical-wireless network infrastructure in the access and fronthaul/midhaul segments. At the network design domain, MARSAL targets the development of novel cell-free based solutions that allows the significant scaling up of the wireless APs in a cost-effective manner by exploiting the application of the distributed cell-free concept and of the serial fronthaul approach, while contributing innovative functionalities to the O-RAN project. In parallel, in the fronthaul/midhaul segments MARSAL aims to radically increase the flexibility of optical access architectures for Beyond-5G Cell Site connectivity via different levels of fixed-mobile convergence. At the network and service management domain, the design philosophy of MARSAL is to provide a comprehensive framework for the management of the entire set of communication and computational network resources by exploiting novel ML-based algorithms of both edge and midhaul DCs, by incorporating the Virtual Elastic Data Center/Infrastructure paradigm. Finally, at the network security domain, MARSAL aims to introduce mechanisms that provide privacy and security to application workload and data, targeting to allow applications and users to maintain control over their data when relying on the deployed shared infrastructures, while AI and Blockchain technologies will be developed in order to guarantee a secured multi-tenant slicing environment.

This deliverable presents the first outcomes from Task 2.1, focused on the description of targeted PoCs. First of all, we present a wide range of use cases and verticals proposed by the literature and we investigate the common aspects with the PoCs that are considered by MARSAL. We investigate two different types of use cases and verticals. First of all, we present a group of 3GPP and O-RAN use cases correspondingly, and we define which of them are considered by the MARSAL project. Then, we provide an overview of the use case domains proposed by ITU (i.e., eMBB, URLLC, mMTC). Specific use cases and verticals by 5G PPP projects are also considered (e.g., 5G ERA, 5GMedi-Hub, VITAL-5G, 5G-PHOS, 5G-ROUTES, MonB5G, 5G-SOLUTIONS,5GEPICENTRE, 6G BRAINS, AI@EDGE, HEXA-X, 5G-TOURS, 5G-MEDIA, 5G-COMPLETE), as well as industrial initiatives (NGMN, ATIS, 5G-ACIA, NEM), which are either related to the MARSAL PoCs, or focused on domains of our wider research interest.

In order to demonstrate a set of experimentation scenarios in MARSAL, we will use the KUL cell-free MIMO testbed, as well ORANGE and CTTC experimental platforms. The KU Leuven Cell-free massive MIMO testbed consists of National Instruments USRPs connected with a MIMO processor responsible for data processing or storage. The ORANGE experimental platform (Lannion) will be used to evaluate the transportation of a given mobile interface through different sorts of optical access networks and provide intelligent management of such networks to optimize the mobile KPIs. The CTTC 5G testbed is based on OpenAirInterface, Amarisoft 5G, and ETSI OSM MANO. It leverages a Cloud Radio Access Network (C-RAN) architecture with a 5G core and a fully virtualized 5G RAN.

In MARSAL, we focus on a wide range of experimentation scenarios. The first domain (PoC1) includes a set of scenarios focused on cell-free networking in dense and ultra-dense hotspot areas. The second domain (PoC2) includes scenarios related to cognitive assistance, as well as security and privacy implications in 5G and beyond.

In particular, PoC 1 is focused on cell-free networking in dense and ultra-dense hotspot areas. Experimentation Scenario 1.1 considers dense User-Generated Content (UGC) distribution with mmWave


fronthauling. The main objective of this scenario is to demonstrate and evaluate MARSAL distributed cell-free RAN in terms of increased capacity and spectral efficiency gains, and the adaptivity of dynamic clustering and Radio Resource Management (RRM) mechanisms in managing connectivity resources in a dynamic environment with varying hotspots areas.

Experimentation Scenario 1.2 investigates ultra-dense video traffic delivery in a converged fixed-mobile network. This scenario will showcase MARSAL’s solution towards Fixed-Mobile Convergence in an ultra-dense indoors context. Mobile clients served by a distributed Cell-Free RAN will be sharing the Optical Midhaul with third party FTTH clients.

PoC 2 is focused on cognitive assistance and its security and privacy implications in 5G and Beyond. Experimentation Scenario 2.1 is about cognitive assistance and smart connectivity for next-generation sightseeing. In this scenario, the deployment of two real-time and interactive cloud-native applications for outdoors sightseeing supporting human-centered interaction via 3D cameras is envisioned in the MARSAL’s multi-tenant elastic edge Infrastructure. These applications would be offered to users equipped with untethered AR glasses. Both applications would endure an enhanced strolling experience by showing overlaid information relevant to their surroundings and enabling virtual artifacts manipulation, while considering background traffic from other applications and services.

Experimentation Scenario 2.2 addresses data security and privacy technical challenges in multi-tenant infrastructures. We approach the security and privacy in 6G networks in a holistic way. We present a modular design to offer four different layers of security and privacy that could be applied in very different contexts. MARSAL technology will ensure the isolation of the different slices while offering the possibility of collaboration among the different tenants. To this end, we will demonstrate how the usage of smart contracts can be paired with the private representation of data, allowing the sharing of information among different tenants and the owner of the infrastructure that can be interested in the optimization of different ML models. Moreover, we will demonstrate how policies can be used to safely store data in the cloud, testing different allocation strategies that ensure the perpetual security of the data. Then, we move our focus to the network and we will demonstrate how the browsing patterns of users can be analysed in real time to alert final users against malicious behaviours they may have, before they get in trouble.

For each experimentation scenario, we provide a list of testing scenarios for the evaluation process, as well as a set of preliminary targeted KPIs. It should be noted here that these KPIs will be further enhanced and refined in the upcoming deliverables. Apart from the experimentation scenario description itself, we identify the involved stakeholders and we map each scenario with the MARSAL architecture to highlight which technical components will be the core ones.

The next WP2 deliverables will be focused on the network architecture specifications (D2.2), the requirements of management and security components (D2.3) and the final release of MARSAL architecture (D2.4). The detailed PoC definitions and requirements will act as blueprints for their implementation in WP6.


1 Introduction 5G networks are set to address the demands of a fully connected and mobile society, enabling a wide variety of applications over the same infrastructure, while carrying 45% of the total mobile traffic [1] and serving up to 65% of the world’s population [2]. These numbers are expected to increase due to the urbanization of global population and the increase of the size and volume of mega-cities, that is, cities with population greater than 10 million [3],[4].

5G changes the landscape of mobile networks in a profound way, with an evolved architecture supporting unprecedented capacity, spectral efficiency, and increased flexibility. Moreover, 5G adopts Edge computing as a key paradigm, evolving from centralized architectures (e.g., based on C-RAN) towards multiple tiers of Edge nodes and a virtualized RAN (vRAN). Open RAN initiatives, such as O-RAN, have a key role in this evolution, complementing the 3GPP 5G standards with a foundation of vRAN network elements. However, these technologies have been in large developed in isolation between them, making difficult to fully exploit their capabilities in an integrated, end-to-end and secure manner. Algorithms do not only run in the cloud, and optical and wireless links cannot be abstracted in the same way. On top of this, when going to cell-free networking concepts such as in MARSAL, more nodes and links will be interconnected, serving local and global secure applications, and thus, within the context of B5G/6G networks, it is essential to rethink the architecture and algorithms running elastically at the scale of a city or building level.

In general, application traffic flows from and towards end-users and end-devices, served by multiple levels of storage and computing entities from the edge to the cloud, while utilizing a diverse set of wireless and optical technologies in the fronthaul, midhaul and backhaul network segments. These infrastructure resources belong to different administrative domains, operate in parallel in the same network areas and are usually shared between competing flows, computations and data in static and/or statically multiplexed manner. Thus, it is clear that targeted innovation activities need to take place to fully exploit key technological developments, towards a disaggregated infrastructure model, where technological infrastructure blocks can be transparently and flexibly replaced by others, while offering similar networking and/or computing offerings and control and monitoring capabilities.

In particular, at MARSAL project level, we identify key advances towards B5G/6G networks which are required both in the network design and network/service orchestration levels. First of all, the network infrastructure should be able to support multiple distributed edge nodes and a huge number of Access Points, which are coordinated and orchestrated by entities in a low-cost and near-zero latency manner. Moreover, a unified and hierarchical infrastructure is essential in order to provide an intelligent management of communication, computation and storage resources, which can be further enhanced by incorporating efficient Machine-Learning (ML) algorithms. Last but not least, the support of multiple tenants should be followed by the application of mechanisms that are able to guarantee data and information security and integrity. This could potentially play a vital role in enabling various use-cases and industry verticals targeted in B5G/6G systems.

Within this context, MARSAL proposes a new paradigm of elastic virtual infrastructures that integrate in a transparent manner a variety of novel radio access, networking, management and security technologies, which will be developed in order to deliver end-to-end transfer, processing and storage services in an efficient and secured way. To this end, MARSAL focuses on three aspects to enable a new generation of ultra-dense, cost-efficient, flexible and secure networks: network design pillar, virtual elastic infrastructure pillar, and network security pillar.


For the network design pillar, MARSAL pushes cell-free networking towards the distributed processing cell-free concept, and enables wireless mmWave fronthaul solutions, which will be implemented and integrated with existing vRAN elements, while being in-line with the O-RAN Alliance.

In parallel, MARSAL’s second pillar is built based on the Elastic Edge Computing notion, targeting to optimize the functionality of the Mobile Edge Computing (MEC) and the network slicing management systems via a hierarchy of analytic and decision engines.

Finally, under its third pillar, MARSAL will develop novel ML-based mechanisms that guarantee privacy and security in multi-tenancy environments, targeting both end users and tenants.

This deliverable presents the first outcomes from Task 2.1, which is focused on the definition of the two PoC requirements, the specifications and the corresponding KPIs. The proposed KPIs are preliminary, and they will be further enhanced and refined in the upcoming deliverables of WP2.

This deliverable presents the core technological enablers considered by MARSAL. We investigate a wide range of use cases and verticals to identify the similarities and differences compared to the MARSAL experimentation scenarios. Then, we describe the two main PoCs and the corresponding four experimentation scenarios. In particular, we provide a detailed description, we identify the main technological challenges, we present the involved stakeholders and their roles, and we investigate the required testing and evaluation process. This analysis will be useful for the definition of the PoC requirements, including their components, applications and business KPIs, and addressing their target values. The PoCs in all facilities will be parameterized, determining the technical parameters that may affect their implementation and refining them to the level of detail required for conducting the project effectively. The detailed PoC definitions and requirements will act as blueprints for their implementation in WP6.

This document is organised as follows:

Section 2 provides a conceptual overview of the MARSAL architecture. A more detailed analysis of the architecture will be presented in Deliverable D2.2.

Section 3 investigates the core technological enablers we consider in this project. We briefly present the current State-of-the-Art and we focus on MARSAL innovations.

In Section 4, we present a wide range of use cases and verticals proposed by the literature and we investigate the common aspects with the PoCs that are considered by MARSAL. In particular, we present a group of 3GPP and O-RAN use cases, the three use case domains proposed by ITU, specific use cases and verticals which have been considered by 5G PPP projects, as well as industrial initiatives, which are either related to the MARSAL PoCs, or focused on domains of our wider research interest.

Section 5 describes the main PoCs, which will be evaluated during our project. In particular, we present the considered experimentation scenarios and the involved stakeholders. We also map the scenarios with the MARSAL architecture, and we consider the testing and evaluation process.

We conclude our remarks in Section 6.


2 MARSAL Overall Architectural Approach This section presents an overview of the MARSAL network architecture. Please note that a detailed presentation of the MARSAL architecture will be provided in Deliverable D2.2.

MARSAL aims to provide an evolved architecture towards B5G/6G, offering unprecedented degrees of flexibility and closed-loop autonomy at all tiers of the infrastructure, and significantly improved spectral efficiency via cell-free networking. The overall architecture and the structure of the envisioned B5G/6G MARSAL is depicted in the generic schematic of Figure 1, and includes all the main infrastructure elements that are deployed within MARSAL project. MARSAL adopts an evolved 3GPP NG-RAN which is extended with emerging cell-free technologies for network densification. Moreover, the MARSAL architecture considers innovations at the optical transport domain and significant evolutions of the MEC system towards fully elastic Edge Computing. MARSAL deploys a distributed Edge infrastructure with Data Centres (DCs) structured in 2 tiers, featuring Regional Edge and Radio Edge nodes. Radio Edge DCs will host the Network Functions of the (virtualized) RAN, which fully aligned with the O-RAN specifications.

Figure 1: The MARSAL network architecture

At the network level, the emphasis is on innovations at the RAN and fronthaul domains that will unlock the potential of cell-free networking in future B5G/6G networks. MARSAL envisions cell-free networking as a key component of B5G/6G RANs, that will offer unprecedented spectral efficiency (SE) and performance which is not constrained by inter-cell interference. The MARSAL network architecture is based on novel cell-free networking mechanisms that will allow the significant scaling up of AP deployment in a cost-effective manner, by exploiting the distributed processing cell-free concept. The novel mechanisms are based on the disaggregation of the traditional cell-free Central Processing Unit (CPU) in Distributed Units (DUs) and a Central Unit (CU) in line with the 3GPP NG-RAN architecture. Regarding the wireless fronthaul links, the MARSAL architecture is based on an innovative mmWave Hybrid MIMO solution, specifically targeting cell-free networks, with advanced beamforming and beamsharing capabilities. In this way, a new AP topology adaptation in cell-free networks, and advanced scenarios can be supported, with APs reassigned to different DUs on demand. MARSAL’s cell-free innovations will be implemented and integrated with existing vRAN elements for the first time and will be contributed back to the O-RAN project. Specifically, the MARSAL


network architecture will be aligned with the O-RAN Alliance architecture, which represents an evolution of Cloud RAN (C-RAN), further disaggregating the) and complementing the 3GPP 5G standards with a foundation of virtualized RAN (vRAN) network elements and packet-based interfaces. Specifically, O-RAN disaggregates the BBU in a DU with the real-time functions, and a CU with the non-real time functions. The latter is further disaggregated into the CU-User Plane (CU-UP) and the CU-Control Plane (CUCP). MARSAL’s CU User Plane function (i.e., CU-UP) and DU will be deployed at MARSAL’s Radio Edge, and the CU-CP Near-RT RIC at the Regional Edge.

At the network service and management level, the MARSAL architecture will consider a novel hierarchical control plane solution, federating the SDN controllers of the fixed and mobile segments of the network under a common orchestration subsystem. MARSAL proposes the disaggregation of the Non-RT into Near-RT SDN Control function that will be hosted by the Near-RT RIC at the Regional Edge nodes. Thus, near real-time reaction to workload variations will be supported, at sub-second timescales. Moreover, MARSAL proposes the deployment of Software-Defined Transport Network Controllers (i.e., SDTNs) at the Regional Edge to control the fixed segment. Both domains will be federated under MARSAL’s Core Tier NFVO, based on ETSI OSM, which will provide Network Slicing as a Service (NSaaS) functionality as per 3GPP TR 28.801 [5] specifications. Thus, end-to-end slicing with centralized orchestration is supported, while still allowing innovative closed-loop (or ML-driven) control of each individual domain. The MARSAL architecture also considers a novel Fixed Mobile Convergence (FMC) solution, to facilitate integrated connectivity of mobile and fixed (i.e., FTTH) services. MARSAL’s solution involves two transmission approaches seamlessly integrated at the Regional Edge node, including a standard Point-to-Point (PtP) connection with or without WDM and a very disruptive Point-to-Multipoint (PtMP) approach based on XGS-PON modules.

In addition, while previous approaches adopted a common VM-based technology stack for MEC and NFV, MARSAL approach is based on Cloud-Native technologies (i.e., Docker Containers, Kubernetes Virtual Infrastructure Managers (VIMs)) which are widely regarded as the future of vertical application development [6]. While support for Kubernetes VIMs is gradually emerging in MEC platforms (e.g., in StarlingX [7]), there is currently a gap in supporting disaggregated Cloud-Native apps. To fill this gap, MARSAL proposes extensions to the Multi-access Edge Orchestrator (MEO) to support the disaggregation of application functions, that will be defined as collections of helm charts, both horizontally (i.e., across Edge sites) and vertically (i.e., from the cell site towards the core cloud). MARSAL will consider the deployment of the aforementioned functions, either at the “bare metal” of the MEC hosts’ NFVIs, or within VNFs, as proposed in the NFV-IFA 029 [8], thus compatible with the NSaaS sub-system. Moreover, MARSAL will extend the Mobile Edge platform at the host level, to allow MEC apps to be accessed by any UE, irrespective of physical location. Dynamic Virtual Network Embedding algorithms will be explored, to determine the optimal disaggregation of application functions at any Edge DC, considering Compute, Networking, and Storage constraints, thus achieving increased resource utilization.

In parallel, the MARSAL architecture is built by considering a novel, distributed approach that involves Analytic Engines at all tiers of the Edge infrastructure, and Decision Engines at the two Core-Tier orchestration subsystems. Analytic Engines, the first pillar of automation, analyse and federate measurements to achieve Context Awareness, and Decision Engines, as the second pillar, Plan and React to Context changes, delegating data-driven local control decisions to the lower tiers of the hierarchy. For the first pillar of automation, MARSAL will design and implement an innovative, decentralized approach to achieve global Context Awareness, using for the first time Representation Learning such as Embedding Propagation (EP) or the GraphSage algorithms. State-of-the-art context representation methods for 1D and 2D datasets are not appropriate for MARSAL’s diverse network and application data, that can be best represented with a graph-


like abstraction. To this end, network slices and MEC applications will be represented as the nodes of a knowledge graph, along with their defining variables and parameters (e.g., SLA requirements, latency budgets, cost considerations, energy efficiency goals). MARSAL proposes to apply EP mechanisms to build the node representations (or embeddings) of the knowledge graph, iterating over the data and minimizing the differences among neighbouring embeddings in the graph. For the second pillar of automation, the resulting embeddings that represent the current state (or context) of the MARSAL infrastructure in a highly compressed form (i.e., encoded as multidimensional normalized arrays) can be transmitted to the Core Tier Decision Engines. The embeddings are fed to downstream ML algorithms implemented by the Decision Engines that jointly orchestrate Network Slices, Network Services and MEC applications continuously and automatically evaluating current context under required policy. Due to the high number of (potentially conflicting) parameters and policy requirements involved, MARSAL will consider multi-objective optimization techniques that achieve different trade-offs between optimality and complexity.

Finally, at the network security level, MARSAL introduces mechanisms that guarantee privacy and security in multi-tenancy environments, targeting both end users and tenants. MARSAL aims to deliver a decentralized, blockchain-based platform that supports network slicing transactions via smart contracts, targeting multi-tenant infrastructures for the first time. In this platform, the MNO, MVNOs, and OTT vertical application owners form a decentralized autonomous organization, which can dynamically negotiate network slice contracts, flexibly integrating large and small players without the need for a centralized entity. Smart Contracts facilitate direct contracts among entities that can be dynamically renegotiated based on real-time supply and demand. MARSAL’s smart contract platform will be implemented with a private, permissioned blockchain solution where tenants will co-own the validator nodes' network after approval and authentication. Going a step further, MARSAL will also incorporate new privacy-preserving context representations, which will allow MARSAL’s data-driven NSaaS subsystem to operate without exposing tenants’ business and operational data. Context awareness requires the exchange of local embeddings (via EP mechanisms) that represent the nodes of a knowledge graph, risking information leakage. These embeddings have inherent anonymization properties, as they represent nodes as compressed, high-dimensional arrays, while the application of EP algorithms iteratively minimizes the differences among neighbouring embeddings, further decreasing the risk of re-identification of the original node. MARSAL will also integrate innovative techniques to guarantee that embeddings can’t be reversed and can be shared among competing partners and the NSaaS sub-system, without any risk of disclosing confidential information.

The MARSAL architecture will also incorporate an innovative NFS Gateway, controlled by the OTT application provider, to serve as the foundation of trust. The DCS gateway will be the intermediary between the (trusted) OTT application, and (untrusted) DCS infrastructure. The NFS gateway will be extended to implement a novel data pipeline for the controlled sharing of data among different parties. The gateway will support for first time a policy language, extending SoTA solutions such as Open Digital Rights Language or JSON-LD, effectively enforceable in the data protection context of different stakeholders that will permit the specification of sharing and processing restrictions over data. Finally, the NFS gateway will implement a novel probabilistic scheme that protects the integrity of computations based on the randomized injection of pre-computed and replicated computational tasks. Thus, a unified solution for data obfuscation and integrity assurance will be implemented for the first time, which varies the probability of randomized injections based on the degree of protection or performance required.

To improve the performance of current signature-based solutions for dealing with zero-day or evolving attacks, MARSAL considers hardware accelerated solutions for a decentralized Threat Detection Engine and a centralized Threat Analysis Engine. ML-based threat detection, that has demonstrated an improved ability


to extract complex non-linear relationships in attack data, will be leveraged for the design of MARSAL’s Threat Detection Engine (TDE). Moreover, MARSAL will leverage the capabilities of a new generation of programmable SDN Switches, which allows MARSAL’s data-plane to behave as a distributed barrier against threats, securing the entire transport infrastructure and intercepting cyber-attacks at a very early stage. Detected cyber-attacks can be isolated by the data-plane at the level of individual traffic flows, going beyond traditional slice-centric approaches and towards micro-segmentation.

Moreover, MARSAL’s network security level also considers a centralized Threat Analysis Engine (TAE), that operates as an ML Fusion Centre, collecting and correlating metadata and features extracted from the P4 pipeline of the decentralized TDE. This allows complex attacks such as Advanced Persistent Threats that simultaneously target multiple network nodes, to be detected. Furthermore, it provides system-wide consistency and correlation for events occurring within all the involved P4 pipelines. MARSAL will exploit the flows’ destination IPs, and specifically their sequence, since this information is both unencrypted and readily available as part of the SDN Switches’ telemetry framework. Thus, the TAE will associate observed flows (with unknown status) with malicious ones based on the sequence of IPs accessed. MARSAL’s solution will involve feeding a Deep Neural Network with sequences of IPs to build a vector representation of network flows; intuitively, flows with a small distance from malicious flows should also be flagged as malicious.


3 Technological Enablers MARSAL targets the development of novel cell-free networking mechanisms that will allow the significant scaling up of AP deployment in a cost-effective manner, by exploiting the application of the distributed processing cell-free concept. MARSAL proposes to disaggregate the traditional cell-free Central Processing Unit (CPU) in Distributed Units (DUs) and a Central Unit (CU) in line with the 3GPP NG-RAN architecture. APs are interconnected with MARSAL’s cell-free DUs via fronthaul connections, either wirelessly or in a bus configuration (i.e., Serial Fronthauling). Regarding the wireless fronthaul links, MARSAL will design and prototype an innovative mmWave Hybrid MIMO solution. Many innovative algorithms for optimized AP clustering, local coordination, energy efficiency, topology adaptation via beam steering will be proposed in order to achieve maximum exploitation of the channel and traffic correlation patterns. MARSAL’s cell-free innovations will be implemented and integrated with existing vRAN elements (e.g., vO-DU, Near-RT RIC, etc.) for the first time, and will be contributed back to the O-RAN specifications. At the transport domain, standard Optical Ethernet technologies are leveraged for the midhaul interfaces, including Point-to-Point and Point-to-Multipoint PONs while a WDM-based optical ring is utilized to interconnect the Regional Edge nodes. Moreover, MARSAL’s network is fully aligned with the Fixed-Mobile Convergence (FMC) paradigm, allowing the midhaul optical fibres to be flexibly shared with Fixed services (i.e., FTTH).

3.1 Cell-free MIMO networks Network densification (i.e., increasing the number of base stations per unit of space) is one of the main techniques that resulted in improving spectral efficiency in in 4G and 5G cellular networks [9]. The drawback of this solution is high interference that negatively affects performance of cell-edge users [10]. Networks of next generations will have to deal with even higher density of infrastructure to provide the expected performance [11]. This requires re-thinking of the underlying architecture to eliminate the cell boundaries [12],[10].

In MARSAL, we design a Cell-Free (CF) massive MIMO network which refers to a network with densely deployed RUs cooperatively serving User Equipment units through coherent joint transmission and reception [12],[11] using the same time-frequency resources. Consequently, the concept of cells is eliminated, motivating the name.

Regarding the network architecture, MARSAL will disaggregate the traditional CPU in multiple DUs, in line with 3GPP’s 5G architecture [13] and will propose novel solutions based on fully distributed (aligned with O-RAN Alliance specifications [14]), data-driven processing and local coordination. The disaggregation is vital for creating scalable versions of cell-free architectures available in SoTA [15], which will unlock the potential of deploying cell-free networking in future 6G networks with massive RU deployments.

Next, we will address cluster formation (selection of serving RUs). In contrast with available simplistic distance-based solutions [16], we will dynamically allocate a sub-set (or cluster) of RUs to each UE based on i) the radio propagation environment; ii) quality of CSI estimates; iii) constraints introduced by computation requirements; iv) fronthaul links capacity, and v) user mobility. Going a step further, MARSAL will propose innovative ML algorithms for optimal cluster formation focusing on real-time operation by using historic data coming from the network. Additionally, we consider investigating ML-based solutions for advanced modulation schemes and/or channel estimation/equalization.

Finally, MARSAL will consider a cluster of RUs served by multiple DUs for the first time (in contrast to disjoint clusters in [15]), where we will provide an inter-DU coordination algorithm for decoding the actual signal.


Moreover, we will be exploring inter-DU coordination requirements and their effect in the spectral efficiency performance and propose a dynamic adaptability of the coordination levels jointly addressing RU-DU and DU-DU coordination for the first time.

3.2 Fronthauling technologies for 6G networks The currently available implementations of cell-free networks rely on optical fiber for interconnecting DU and RUs [17]. Unfortunately, this solution is unscalable. 4G and, especially, 5G networks faced this issue due to their high density. As we mentioned above, 6G networks are going to drastically increase the network density making the exclusive use of optical fronthaul economically unacceptable.

In MARSAL, we rely on i) the scalable cell-free architecture described above (allowing to find an optimal balance between RU functionality and fronthaul capacity) and ii) the SoTA fronthaul concepts formulated by ORAN alliance [14]. We propose two cheaper and sustainable alternatives to costly optical fiber deployments: i) wireless fronthaul and ii) serial fronthaul. On the one hand, wireless connection is cheaper; it allows for dynamic reconfiguration without hardware deployment when the network topology is changed. On the other hand, when several RUs are connected in a bus mode, the same wire can be reused for communication with a number of RUs instead of one.

For the wireless fronthaul, we aim at solving several critical problems: i) define ORAN functional splits (e.g., capacity/latency requirements); ii) define the array requirement and make a prototype); ii) define the RU deployment strategies (e.g., optimizing RU density and link reliability depending on the obstacles). Moreover, since several RUs serve one user in cell-free networks, the same information can be sent by DU in different directions. Hence, the designed wireless fronthaul will support multicasting based on a hybrid beamforming architecture [18]. Another problem of densely deployed networks is interference mitigation. We aim to solve it by array reconfiguration (e.g., by nulling towards the interferer).

Using serial fronthaul in cell-free provides us with an opportunity to share the channel estimates between the RUs without going through the central entity (DU/CU). In MARSAL, we will exploit this feature to propose a scalable low-latency interference cancellation technique (e.g., based on parallel of sequential approaches [19].

3.3 Multi-objective optimization towards self-Driven, elastic Infrastructures MARSAL will follow a hierarchical approach towards a Self-Driven Virtual elastic infrastructure (Section 2) that provides high quality end-to-end transfer, processing and storage services in an efficient way. To achieve this, global Context Awareness is required and will be provided by Representation Learning and Embedding Propagation [20] algorithms. To avoid bottlenecks inherent in centralized approaches, decentralized protocols will be explored to allow each Edge node to independently generate node representations (or embeddings) for their own part of the graph, exchanging EP messages with neighbouring nodes iteratively to minimize the differences between embeddings. Next, the optimization ML algorithms will leverage the provided current network state, described by knowledge graphs, to jointly orchestrate Network Slices, Network Services and MEC applications continuously and automatically. In this regard, MARSAL will extend the ME platform at the host level, to allow MEC apps to be accessed by any UE, irrespective of physical location, for inter-DC load balancing. Dynamic multi-objective Virtual Network Embedding algorithms will be explored and implemented, to determine the optimal disaggregation of application functions at any Edge Data-Centre, considering Compute, Networking, Storage, and latency constraints as defined by the MEC application manifest.


However, the optimization algorithms need to efficiently address simultaneously a high number of (potentially conflicting) parameters and policy requirements, which significantly increase the complexity. Hence, we will consider multi-objective optimization techniques [21] that achieve different trade-offs between optimality and complexity and improve application service delivery over the available resources. In particular, multi-agent Deep Reinforcement Learning [22] will be considered to simultaneously address the multiple objective functions related to policy requirements. In that case, multiple agents with diverse reward functions will be competing to solve the single-objective problems via Deep Reinforcement learning. Deep reinforcement-learning has been proposed as a powerful data-driven model for environments with huge system states and the need for a real-time reward function. In addition, techniques based on Dynamic Programming (DP) and Meta/Heuristics will be investigated to further improve the performance of the aforementioned mechanisms.

3.4 Data security and privacy in multi-tenant infrastructures The ability of supporting data-intensive applications that require the analysis of data under the control of different parties is a priority for 5G/B5G mobile networks [23]. The considered scenario is typically characterized by a diversity of data sources stored on different nodes, possibly under the control of different parties. Data analysis may then require data exchanges and cooperation between these different parties. In such a context, there are several issues that need to be investigated [24]. A first problem is related to the fact that data may need to be selectively accessed in a cooperative way for executing certain analysis (queries). This implies the need of exchanging data and of executing collaborative computations that, however, should be controlled to avoid information leakage. For instance, data stored at one node might be released selectively, in restricted form, only to other specific nodes and within specific domains. Several proposals have addressed this problem, but they do not consider the possibility of protecting data through, for example, on-the-fly encryption (e.g., [25]). MARSAL will instead provide a solution for expressing and enforcing data sharing constraints, considering the cost of operation execution and ensuring their enforcement even in non-trusted environments where data may be possibly encrypted (e.g., [26]). Such a solution will include a flexible model for representing, in an easy and effective way, the access privileges to portions of distributed data, supporting different levels of visibility over the data (e.g., plaintext visibility or encrypted visibility). The access privileges will regulate data sharing and flow among providers, also considering the trust assumptions on the parties involved in the data sharing and flow. MARSAL will develop a mediator and enforcer of such access privileges.

A concern related to the storage and collaborative processing of data is the lack of control over the computation and hence the uncertainty about the correctness of the result. This is a well-known problem and the research and industrial communities have devoted many efforts to the development of techniques to assess integrity of the result of computations outsourced to external parties (e.g., [27],[28]). However, the problem of how to use such techniques and of assessing their effectiveness in different application scenarios still need to be further investigated. MARSAL will focus on probabilistic techniques since they can be applied in contexts where computations are not fixed a priori. In this case, the detection of integrity violations will be based on the combined adoption of approaches such as data replications and markers [27]. The goal is to define a formal model for assessing the effectiveness and synergy of the probabilistic techniques adopted. The model will allow different parties to tune the amount of control to be enforced (and therefore the security guarantees to enjoy and the performance overhead to pay), considering different contexts or applications. Attention will be also devoted to the design of techniques for distributing different data chunks to different providers for providing better confidentiality guarantees (e.g., sensitive data could be stored


within a trusted provider and non-sensitive data and/or an obfuscated version of sensitive data at an untrusted provider).

3.5 Distributed ML for privacy and network security The usage of ML technologies is becoming pervasive with applications ranging from the diagnosis of cancer [29] to the selection of advertising online. The analysis of network data to improve the security or the performance [30] is not an exception with relevant advances in the past years. However, the usage of data is not exempt of issues, with the privacy of the final users (and the confidentiality of company data) on the focus when data should be shared among different partners.

In MARSAL we propose to research and develop representation learning techniques [31] to represent the operational data generated by different MVNOs and MNOs running slices in the MARSAL platform. The data generated with this method will allow the seamless sharing of data among network tenants and with the infrastructure owned without the sharing of any personal or confidential information.

Moreover, MARSAL aims to deliver a decentralized, blockchain-based platform that supports network slicing transactions via Smart Contracts [32], targeting multi-tenant infrastructures for the first time. In this platform, the MNO, MVNOs, and OTT vertical application owners form a Decentralized Autonomous Organization (DAO) which can dynamically negotiate Network Slice contracts, flexibly integrating large and small players without the need for a centralized, trusted entity. Smart Contracts facilitate (and automate) direct contracts among entities that can be dynamically renegotiated based on real-time supply and demand.

Finally, to improve the performance of current signature-based solutions for dealing with zero-day or evolving attacks, MARSAL provide hardware accelerated solutions for a decentralized Threat Detection Engine [33] and a centralized Threat Analysis Engine. ML-based threat detection, that has demonstrated an improved ability to extract complex non-linear relationships in attack data, will be leveraged for the design of MARSAL’s Threat Detection Engine (TDE).


4 5G Use Cases and Verticals Overview In this section, we present a wide range of use cases and verticals proposed by the literature and we investigate the common aspects with the PoCs that are considered by MARSAL. In particular, we investigate two different types of use cases and verticals. Sections 4.1 and 4.2 present a group of 3GPP and O-RAN use cases correspondingly, and we define which of them are investigated by the MARSAL project. Then, we provide an overview of the use case domains proposed by ITU (Section 4.3), specific use cases and verticals which have been considered by 5G PPP projects (Section 4.4), as well as industrial initiatives (Section 4.5), which are either related to the MARSAL PoCs, or focused on domains of our wider research interest.

4.1 3GPP use cases 4.1.1 3GPP USE CASE GOALS AND DRIVERS The goal of this section is to define use cases, highlight the differences, advantages, and increased complexity of the MARSAL cell-free architecture, compared to more traditional 3GPP 5G Standalone cellular networks. The following 3GPP use cases increase in complexity, with increasingly distributing higher layers of functionality and consequent need for co-ordination from DU to CU then 5GC. The aim is to allow selection of the lower complexity use cases for implementation and demonstration within MARSAL, however, the higher complexity use cases may be beyond MARSAL implementation scope and subjects for future study.

The 3GPP use cases allow demonstration of the benefits of user-centric cell-free over cellular networks, as described in [12], namely:

Higher SNR with smaller variations; Better ability to manage interference, compared to small-cell networks; Coherent transmission increases the SNR.

In the following subsections, we present a group of 3GPP use cases and we define which of them will be investigated by the MARSAL project.

4.1.2 3GPP UC1: INTRA DU (CLUSTER) CELL-FREE, SINGLE UE

Definition: Attachment and end-to-end connectivity of a single UE to a cell-free cluster of multiple RUs managed by a single DU without mobility.

Aim: This UC illustrates the simplest form of user-centric cell-free operation where a single DU manages multiple RUs in a cell-free network, to show a UE connecting to and operating e2e. This UC will be demonstrated by MARSAL.

Involved entities/resources: Single UE. Multiple RUs connected to a single DU with cell-free operation. A standard 3GPP Rel-15/16 CU and 5GC will provide the end-to-end IP connectivity via single PDU session.

As there is only a single DU, there will be no need for any inter-DU (or higher layer) interfaces or coordination.

4.1.3 3GPP UC2: INTRA DU (CLUSTER) CELL-FREE, MULTIPLE UE

Definition: Attachment and end-to-end connectivity of multiple UEs to a cell-free cluster of multiple RUs managed by a single DU without mobility (extension of UC1).

Aim: This UC illustrates basic user-centric cell-free operation where a single DU manages multiple RUs, in a cell-free network, to show UEs connecting to and operating e2e. This UC will be demonstrated by MARSAL.


Involved entities/resources: Multiple UEs. Multiple RUs connected to a single DU with cell-free operation. A standard 3GPP Rel-15/16 CU and 5GC will provide the end-to-end IP connectivity via single PDU session.

As there is only a single DU, there will be no need for any inter-DU (or higher layer) interfaces or coordination.

4.1.4 3GPP UC3: INTER DU (CLUSTER) CELL-FREE, SINGLE UE

Definition: Attachment and end-to-end connectivity of a single UE to multiple cell-free clusters of multiple RUs managed by a single DU (per cluster) without mobility, within 1 region (single CU).

Aim: This UC illustrates coordinated user-centric cell-free operation of multiple clusters of DUs managing multiple RUs, in a cell-free network, to show a cluster-edge UE connecting to and operating e2e. This UC will be demonstrated by MARSAL.

Involved entities/resources: Single UE (connecting to multiple clusters). Multiple clusters of RUs connected to DUs with cell-free operation. 3GPP aligned CU and 5GC will provide the end-to-end IP connectivity via single PDU session (The CU may use standardized NR-NR Dual Connectivity of 3GPP TS 37.340 [34] or implement cell-free specific extensions to support the ‘UE spread over multiple DUs’ functionality. This remains to be studied).

As there are multiple DUs, there will be a need for inter-DU interfaces and coordination. The UP/DL TDD timeslot allocations for the UE will need to be identical across the multiple DU MAC schedulers. The separate cluster fronthaul networks (and RUs/DUs) will require stringent time and frequency synchronization for cell-free operation.

4.1.5 3GPP UC4: INTER DU (CLUSTER) CELL-FREE, MULTIPLE UE, MOBILITY

Definition: Attachment and end-to-end connectivity of multiple UEs to multiple cell-free clusters of multiple RUs managed by a single DU (per cluster) with mobility (extension of UC3), within 1 region (single CU).

Aim: This UC illustrates coordinated user-centric cell-free operation of multiple clusters of DUs managing multiple RUs, in a cell-free network, to show a dynamic mix of local and cluster-edge UE connecting to and operating e2e. Inter-DU mobility will be considered. This UC may be demonstrated by MARSAL.

Involved entities/resources: Single UE (connecting to multiple clusters). Multiple clusters of RUs connected to DUs with cell-free operation. 3GPP aligned CU and 5GC will provide the end-to-end IP connectivity via single PDU sessions (CU constraints as per UC3) with support for mobility of UEs.

As there are multiple DUs, there will be a need for inter-DU interfaces and coordination. The UP/DL TDD timeslot allocations for the UE will need to be identical across the multiple DU MAC schedulers. The separate cluster fronthaul networks (and RUs/DUs) will require stringent time and frequency synchronization for cell-free operation.

4.1.6 3GPP UC5: INTER CU (REGION) CELL-FREE, SINGLE UE

Definition: Attachment and end-to-end connectivity of a single UE to multiple cell-free clusters of multiple RUs managed by a single DU (per cluster) without mobility, across multiple regions (single CU per region).

Aim: This UC illustrates coordinated user-centric cell-free operation of multiple regions of CUs and clusters of DUs managing multiple RUs, in a cell-free network, to show a region-edge UE connecting to and operating e2e. This UC may be demonstrated by MARSAL.


Involved entities/resources: Single UE (connecting to multiple clusters and regions). Multiple clusters of RUs connected to DUs in multiple regions of CUs with cell-free operation. 3GPP aligned CU and 5GC will implement cell-free specific extensions to support the ‘UE spread over multiple DUs and CUs’ functionality.

As there are multiple DUs, there will be a need for inter-DU interfaces and coordination. There will also be a need for inter-CU interfaces and coordination. The UP/DL TDD timeslot allocations for the UE will need to be identical across the multiple DU MAC schedulers. The separate cluster fronthaul networks (and RUs/DUs) will require stringent time and frequency synchronization for cell-free operation.

4.1.7 3GPP UC6: INTER CU (REGION) CELL-FREE, MULTIPLE UE, MOBILITY

Definition: Attachment and end-to-end connectivity of a multiple UEs to multiple cell-free clusters of multiple RUs managed by a single DU (per cluster) with mobility (extension of UC5), across multiple regions (single CU per region).

Aim: This UC illustrates coordinated user-centric cell-free operation of multiple regions of CUs and clusters of DUs managing multiple RUs, in a cell-free network, to show a dynamic mix of local and cluster-edge UEs connecting to and operating e2e. Inter-DU and inter-CU mobility will be considered. This UC will not be demonstrated by MARSAL.

Involved entities/resources: Multiple UEs (connecting to multiple clusters and regions). Multiple clusters of RUs connected to DUs in multiple regions of CUs with cell-free operation. 3GPP aligned CU and 5GC will implement cell-free specific extensions to support the UE spread over multiple DUs and CUs’ functionality.

As there are multiple DUs, there will be a need for inter-DU interfaces and coordination. There will also be a need for inter-CU interfaces and coordination. The UP/DL TDD timeslot allocations for the UE will need to be identical across the multiple DU MAC schedulers. The separate cluster fronthaul networks (and RUs/DUs) will require stringent time and frequency synchronization for cell-free operation.

4.1.8 3GPP UC7: INTRA/INTER NG-RAN MOBILITY – CELL FREE BACKWARD COMPATIBILITY

Definition: Attachment and end-to-end connectivity of multiple UEs to multiple gNodeBs, with mobility within one or across multiple regions.

Aim: For similar UCs, such as the aforementioned ones, i.e., mobility optimization, dual connectivity and resource and load coordination between NG-RAN nodes, 3GPP inter NG-RAN and intra NG-RAN communication is employed [35],[36]. Regardless of the cell-free extensions on NG-RAN elements, for backward compatibility of the cell-free architecture with the traditional 5G NG-RAN, these capabilities and the associated protocols should also be supported seamlessly as defined in 3GPP. Some of the UCs may be demonstrated by MARSAL.

Involved entities/resources: Multiple UEs. Cell-free enhanced RUs, CUs, DUs and Core are needed to create end-to-end IP connectivity via multiple PDU sessions. The exact number of sub-systems are based on the legacy 3GPP mobility use case that will be evaluated. E.g., in case of inter-DU Handover (Source and Target gNodeB-DU connected to Common gNodeB-CU) [35] multiple DUs are needed. In case of inter-CU Handover (It can be Xn or N2 based Handover) [36], multiple CUs are needed.

4.2 O-RAN use cases In this subsection, we present O-RAN high level use case definition (Figure 2). Then, we focus on the motivation, as well as on the interaction of the use cases through the Near-RT RIC controller.


Figure 2: O-RAN specified use cases for 5G networks [37]

4.2.1 Use case 1: Context-based Dynamic Handover Optimization Management for V2X

Definition: This use case provides the background, motivation, and requirements for the Context-based Dynamic HO Management for V2X use case, allowing operators to adjust radio resource allocation policies through the O-RAN architecture, reducing latency and improving radio resource utilization.

Aim: This UC aims to present a method to avoid and/or resolve problematic HO scenarios by using past navigation and radio statistics to customize HO sequences on a UE level. The Use Case promises numerous benefits such as increased road safety, reducing emissions, and saving time. These are facilitated by orchestration of the traffic and assistance of individual user decisions based on real-time information on the road and traffic conditions, driver intentions etc.

Involved entities/resources: The UC1 is mainly involved with the following key entities of the O-RAN architecture, Non-RT RIC, Near-RT RIC, V2X application server, and RAN support, respectively. The above-mentioned resources involved on UC1 will collect the data to construct AI/ML models, update the AI/ML models, support data collection, support communication of real-time traffic related data. The measurement reports with RSRP/RSRQ/CQI from serving and neighboring cells, the mobility and handover statics, and the V2X data, e.g., position, velocity, direction, etc., is needed in UC1 to execute dynamic handover management optimization of V2X.

4.2.2 Use case 2: Flight Path-based Dynamic UAV Radio Resource Allocation

Definition: This use case provides the background, motivation, and requirements for the support the use case and allowing operators to adjust RRA policies through the O-RAN architecture. It will also reduce the unnecessary handover and improving radio resource utilization.

Aim: This UC aims to present a flight-based dynamic UAV RRA mechanism supported by the RIC function module that can perform the RRA operation based on considering coverage UAV. Use Case support for flight path based dynamic UAV Radio Resource Allocation, allowing operators to adjust radio resource allocation

O-RAN use cases

Use case 1: Context-based

dynamic HO management

for V2X

Use case 2: Flight path-

based dynamic UAV RRA

Use case 3: RRA for UAV application

scenario

Use case 4: QoE

optimization

Use case 5: Traffic

SteeringUse case 6: mMIMO

beamforming optimization

Use case 7: RAN sharing

Use case 8: QoS based resource

optimization

Use case 9: RAN slicing

SLA assurance

Use case 10: Dynamic spectrum sharing


policies through the O-RAN architecture, reducing unnecessary handover and improving radio resource utilization.

Involved entities/resources: Similarly, the UC2 is involved with the following key entities of the O-RAN architecture, Non-RT RIC, Near-RT RIC, Application server, and RAN support, respectively. The above-mentioned resources involved on UC2 will retrieve the aerial vehicles related measurement metrics to construct the AI/ML models, send policies/intents to the drive the O-RAN support UAV at RAN level, update the AI/ML models, support data collection, and support data collection with UE performance, respectively. The UE level radio channel condition, mobility related metrics and the UAV related measurement metrics collected from SMO is needed in UC2 to execute dynamic UAV RRA based on flight path of the UAV.

4.2.3 Use case 3: RRA for UAV Application

Definition: This use case provides the background, motivation, and requirements for the UAV control vehicle use case. It will adjust RRA policies through the O-RAN architecture and reduce latency and improving radio resource utilization.

Aim: This UC3 objective is to control vehicle scenario and the UE level radio resource configuration by utilizing the delivery of policies and configuration parameters. This Use Case support for UAV control vehicle, allowing operators to adjust radio resource allocation policies through the O-RAN architecture, reducing latency and improving radio resource utilization.

Involved entities/resources: The UC3 is involved with the following key entities of the O-RAN architecture, Non-RT RIC, Near-RT RIC, Application server, and RAN support, respectively. The above-mentioned resources involved on UC3 will be providing RRA requirements, UE-level adjustment request, support related RRA related parameters, terminal registration request, and sending UE-level RRM requirements, respectively. The UE level radio channel condition, mobility related metrics and L2 measurement report, e.g., throughput, latency, packets/second, inter-frame arrival time are needed to provide UC3 related support to the RAN.

4.2.4 Use case 4: QoE Optimization Definition: This use case provides the background and motivation for the O-RAN architecture to support real-time QoE optimization.

Aim: The main objective of UC4 is to ensure QoE optimization within the O-RAN architecture and its open interfaces. The QoE estimation/prediction from application level can help to improve the efficiency of radio resources, and eventually improve user experience. Multi-dimensional data, e.g., user traffic data, QoE measurements, network measurement report, can be acquired and processed via ML algorithms to support traffic recognition, QoE prediction, QoS enforcement decisions.

Involved entities/resources: The UC4 of O-RAN is involved with the following key entities of the O-RAN architecture, Non-RT RIC, Near-RT RIC, and RAN support, respectively. The above-mentioned resources involved on UC4 will retrieve necessary QoE related measurement metrics from network level measurement report, train the AI/ML model for QoE optimization, execute the AI/ML model to solve the application classification, QoE prediction, available bandwidth prediction problems of the RAN, respectively. The UE level radio channel condition, mobility related metrics and L2 measurement report, e.g., throughput, latency, packets/second, inter frame arrival time, RAN protocol stack status: e.g., PDCP buffer status, Cell level information: e.g., DL/UL PRB occupation time data are needed to provide QoE optimization support to the RAN.

4.2.5 Use case 5: Traffic Steering


Definition: This use case provides the requirements for traffic steering use case and allowing the operators to specify the objective of managing the traffic such as optimizing the network/UE performance or achieving balanced cell load in the network.

Aim: The aim of UC4 is to provide proactive load balancing optimization by predicting the network condition and UE performance.

Involved entities/resources: similarly, the UC5 of O-RAN is involved with the following key entities of the O-RAN architecture, Non-RT RIC, Near-RT RIC, and RAN support, respectively. The above-mentioned resources involved on UC5 will specify the policies for different optimization objectives to guide the carrier/band preferences at per/UE, configures the RAN parameters, enforcing the policies from non-RT RIC, support data collection with required granularity to SMO, respectively. The measurement reports with RSRP/RSRQ/CQI information for serving and neighboring cells, the UE connection and mobility/handover statistics, and the Cell load statics e.g., number of active users and connections, number of schedules active users per TTI, PRB/CCE utilization are needed to enable UC5 by utilizing the O-RAN architecture.

4.2.6 Use case 6: mMIMO Beamforming Optimization Definition: This UC6 provides the motivation, description, and requirements for mMIMO beamforming long-term optimization use case. This UC6 contribution about mMIMO system configuration will allows the operator to optimize the network performance and QoS by reducing inter-cell interference.

Aim: The objective of the massive MIMO optimization use case is to improve cell-centric network QoS proactively and continuously and/or user (group)-centric QoE in a multi-cell and, possibly, multi-vendor massive MIMO deployment area with multiple transmission/reception points, depending on specific operator-defined objectives. This optimization is performed at different timescales from near- real-time TTI scale to non-real-time.

Involved entities/resources: The UC6 of O-RAN is involved with the entities/resources of Non-RT RIC, Near-RT RIC, and RAN nodes, respectively to solve the mMIMO beamforming optimization. The above-mentioned resources involved on UC6 will retrieve necessary data from the RAN to build the AI/ML model, retrieve necessary user location related information, optimal beam pattern configuration to SMO, monitor the performance of all the cells, collect and report to SMO KPI’s, e.g., traffic load, coverage, QoS performance, per beam/area, respectively. The environmental data, e.g., the cell-site information (location), ISD, mMIMO system configurations, the measurement reports with RSRP/RSRQ/CQI/SINR per beam information for the UE, the data from application to SMO (i.e., user location related information) is needed to run UC6 by utilizing the O-RAN architecture.

4.2.7 Use case 7: RAN Sharing Definition: The UC7 enables multiple operators to share the same O-RAN infrastructure, while allowing them to remotely configure and control via remote O1, O2, and E2 interface.

Aim: The objective is to make available RAN infrastructure and computing resources to host the virtual RAN functions of multiple operators.

Involved entities/resources: The UC7 of O-RAN is involved with the entities/resources of SMO-sharing APP (site A), SMO-sharing APP (site B), RAN (site A), Non-RT RIC (site B), and Near-RT RIC (site B), respectively to share an interoperable RAN from two separate vendors. The above-mentioned resources involved on UC7 will retrieve SLA monitoring information, remote provisioning and VNF deployment, information related to remote management, supports data collection from the hosted VNFs through “E2 remote” interface, supports


data collection from hosted VNF with UE KPI report, the initial network policy template, e.g., default scheduling policy, of the remote VNFs, respectively. The multi-dimensional data are expected to be handled by the SMO-sharing APP, e.g., the SLA data, SMO needs to manage O1, message sent by the Host operator and converting them in local O1, O2 commands.

4.2.8 Use case 8: QOS-based Resource Optimization Definition: The UC8 provides the background information for the O-RAN architecture to support RAN QoS based resource optimization. This UC also provides functionality that ensures resource isolation between slices as well as functionality to monitor that slice Service Level Specifications (SLS) are fulfilled. Moreover, it will also optimize how RAN resources are allocated between users.

Aim: The aim of UC8 is to drive QoS based resource optimization in RAN in accordance with defined policies and configuration.

Involved entities/resources: The UC8 of O-RAN is involved with the entities/resources of Non-RT RIC, Near RT-RIC and RAN, respectively to provide QoS-based optimization support in the O-RAN architecture. The above-mentioned resources involved on UC8 will monitor necessary QoS related metrics from network and other SMO functions, send policies to derive QoS based resource optimization, support network state and UE performance report to SMO, and support QoS enforcement based on message from E2 interface to influence the RRM behavior of the RAN, respectively. The data related to resource consumption in the area and the information related to user throughput and delay is required to optimize the QoS of the RAN.

4.2.9 Use case 9: RAN Slicing SLA Assurance Definition: The RAN slicing SLA assurance use case is based on RAN specific slice SLA requirements, Non-RT RIC and Near-RT RIC can fine-tune RAN behavior to assure RAN slice SLAs dynamically.

Aim: The main objective of UC9 is to support the needs of the business through the specification of several service needs such as data rate, traffic capacity, user density, latency, reliability, and availability.

Involved entities/resources: The UC9 of O-RAN is involved with the entities/resources of Non-RT RIC, Near RT-RIC and RAN, respectively to provide SLA assurance of RAN slicing. The above-mentioned resources involved on UC9 will retrieve RAN slice SLA target, long-term monitoring of the slice performance measurements, slice control/slice SLA assurance information, slice-aware resource allocation and prioritization information, respectively. The per-UE CSI, per slice performance statistics, and user throughput with delay information are needed from the RAN to assure the SLA of RAN slicing.

4.2.10 Innovations Beyond O-RAN Uses Cases The current O-RAN specification does not provide any cell-free supported UCs for evaluating the RAN performance. MARSAL will demonstrate the beyond SoTA UCs which is not available in current O-RAN specifications. The cell-free enabled UCs discussed in Section 5 will enhance the UC4, US6, and UC8 of current O-RAN specification and the outcomes of MARSAL UCs demonstrate can be suggested to the O-RAN community future specification. To demonstrate the MARSAL PoCs, we need to also consider similar entities/resources suggested in WG111, but we need to modify the procedures of the resources involved in the UCs. Similarly, the RAN nodes of the O-RAN architecture and their corresponding interfaces will be modified in MARSAL to integrate the cell-free solutions in the current O-RAN architecture. The environmental data, e.g., the radio edge site information (location), mMIMO system configurations, the measurement reports with RSRP/RSRQ/CQI/SINR per beam information for the UE, the data from application to (i.e., user location related information) suggested in O-RAN can be used in MARSAL to provide the PoC demonstration.


However, the procedures of collecting the data in a cell-free manner and further process it to enable MARSAL technological pillars is the innovations beyond the scope of current O-RAN specifications.

4.3 ITU-R use case domains Future International Mobile Telecommunications (IMT) systems should support emergent new use cases, including applications necessitating very high data rate communications, a large number of connected devices, and ultra-low latency and high reliability applications. IMT for 2020 and beyond is envisaged to expand and support diverse usage scenarios and applications that will continue beyond the current IMT. Furthermore, a broad variety of capabilities would be tightly coupled with these intended different usage scenarios and related applications. The usage scenarios for IMT for 2020 and beyond include Enhanced Mobile Broadband (eMBB), Ultra-Reliable Low Latency Communications (URLLC) and Massive Machine Type Communications (mMTC) [38].

The objective behind Study on New Services and Markets Technology Enablers (SMARTER) was to develop high level use cases and identify what features and functionality 5G would need to deliver to enable them. It began in 2015 and resulted in over 70 use cases, initially grouped into five essential categories which have, since, been trimmed to three. They are characterised by the performance attributes the particular use cases will require, although there is some overlap. The three fundamental sets of use cases are as follows:

Enhanced Mobile Broadband (eMBB): This “addresses” the human-centric use cases for access to multi-media content, services and data. Actual technology trends purely “demonstrate” that the demand for mobile broadband will continue to increase, thus leading to enhanced Mobile Broadband. The enhanced Mobile Broadband usage scenario will come with new application areas and requirements in addition to existing Mobile Broadband applications for improved performance and an increasingly seamless user experience. This usage scenario covers a range of cases, including wide-area coverage and hotspot, which have different requirements. For the hotspot case that is for an area with high user density, very high traffic capacity is needed, while the requirement for mobility is low and user data rate is higher than that of wide area coverage. For the wide area coverage case, seamless coverage and medium to high mobility are desired, with much improved user data rate compared to existing data rates. However, the data rate requirement may be relaxed compared to hotspot.

Ultra-Reliable Low Latency Communications (URLLC): This set has stringent requirements for capabilities such as throughput, latency and availability. In particular, it implicates for strict requirements on latency and reliability for mission critical communications, such as remote surgery, autonomous vehicles or the Tactile Internet [39]. Some other examples include wireless control of industrial manufacturing or production processes, remote medical surgery, distribution automation in a smart grid, transportation safety, etc.

Massive Machine Type Communications (mMTC): This set is characterised by a very large number of connected devices typically transmitting a relatively low volume of non-delay-sensitive data. These devices are usually located in a small area, which may only send data sporadically, such as Internet of Things (IoT) use cases. Devices are required to be low cost and have a very long battery life.

The MARSAL use cases are thoroughly aligned with the product and services researched and offered by different partners and will have a major and multidimensional impact for beyond 5G telecommunication industries. The developed concepts and technologies in the areas of ML-based cell-free networking and edge computing infrastructure resource management are the key characteristics of the future beyond 5G intelligent systems and thus provide shaping the upcoming consumer industries. The innovative design,


approaches, schemes, and optimization developed in MARSAL will accelerate entering markets that go beyond the wireless access network modes, integrated infrastructures, X-haul configuration, cost-efficient processing, intelligent computing protocols, and enhanced security management. The ML based approaches in the converged optical-wireless infrastructure will provide cost-effective resource pooling by enabling a flexible fronthaul and mid-haul design. In parallel, the cell-free networking concept has the potential to solve the cell-edge problem and at the same time, increase energy efficiency and reduce latency in beyond 5G wireless networks. There is a high potential for research and market development and future growth from the cell-free network architecture, with vendors and operators envisaging to benefit from its deployment due to its excellent suitability for high data-rates and low power, the capability to support larger number of connections, higher reliability and low latency, as well as its increased energy-efficiency. Cell-free massive MIMO inherently supports and promotes the deployment of MEC-based applications while providing excellent support for URLLC and mMTC. Such new innovative technologies, techniques, and infrastructure investigated in the MARSAL will create new commercial opportunities for eMBB-, URLLC- and mMTC-related use cases, and will introduce a new line of products in the smart connectivity domain of beyond 5G networks.

4.4 Use cases and verticals in 5G PPP projects 5G and beyond 5G networks will open the way too many new and exciting applications not possible at the moment due to limited bandwidth and latency. Under the umbrella and scope of 5G PPP, numerous implementations and demonstrations of initial proof of concepts will be achieved. Proof of concepts and Use Case scenarios that will notably benefit from the MARSAL network. In this section, we summarize some of those projects and applications to demonstrate the versatility of the 5G platform.

4.4.1 5G ERA 5G ERA [40] works towards an integrated, open, cooperative, and fully featured network platform running across multiple domains where needed and tailored to specific Use Cases related to robotic application. To demonstrate the specific target, four distinct Use Cases will be explored:

Public Protection and Disaster Relief; 5G enhanced semi-autonomous transport; 5G enhanced healthcare robots; 5G-remote assistance for manufacturing process.

5G ERA’s use cases involved indoors and outdoors activities regarding high bandwidth and low latency. The target applications will initially be implemented with 5G networks but, as the number of robots increases, the need for higher bandwidth and zero perceive latency will increase. The use cases of 5G ERA will immensely benefit from the cell-free network architecture proposed in the MARSAL project.

4.4.2 5GMEDIAHUB Vertical applications need to support the corresponding platform Application Programming Interfaces (APIs) and undergo extensive testing to ensure the meet the service-level Key Performance Indicators (KPIs). To the end, 5GMediaHub [41] will offer a testing infrastructure which is compatible with industry standard DevOps processes, streamlining and simplifying testing and KPI verification. This will be demonstrated through:

Immersive Augmented, Virtual and Extended Reality applications: Immersive 360° VR media experiences; Interactive consumption of 8K and VR media content;


Smart media production: High quality User Generated Content (UGC) uploading dense uploading scenarios; Professional live video production;

Smart media content distribution: Dynamic multi-CDN (multi–Content Delivery Network) selection for 8K IPTV; Smart city co-creation.

Similarly to the MARSAL project, 5GMediaHub will examine use cases related to augmented and virtual reality and dense video consumption. MARSAL elastic edge computing and cloud native technologies will offer zero perceived latency and smart connectivity applications, allowing for immersive media.

4.4.3 VITAL-5G The pan-European Transport & Logistics (T&L) eco-system is considered one of the main adopters of 5G, and as such, the successful transfer of 5G-empowerd services from trial/pilot stages to production depends highly on the availability of flexible and intuitive tools and APIs for design, management, and orchestration of their services. VITAL-5G [42] prioritizes this ambition and plans to overcome, through its intuitive and production-ready NetApps orchestration platform with open repository, various limitations that exist today for industry verticals keen to design and deploy T&L virtualised services in a 5G network. In order to showcase the functionality and added value of the VITAL-5G experimental facility, three Use Cases will be demonstrated focus on:

Automated vessel transport; 5G connectivity and data-enabled assisted navigation using IoT sensing and video cameras; Automation & remote operation of freight logistics.

Assisted navigation using IoT sensing and video cameras required high bandwidth and zero perceived latency. High bandwidth as a vast amount of data from the various sensors and the location data from the GPS needs to be real time transferred and processed and, at the same time, zero perceived latency due to the required accuracy and timing required to navigate a ship in a port. This can be offered in B5G/6G networks based on MARSAL architecture.

4.4.4 5G-PHOS 5G-PHOS [43] is a project focusing on 5G integrated Fiber-Wireless networks that leverage existing photonic technologies towards implementing a high-density SDN-programmable network architecture. Three Use Cases are envisioned by the project based on different dimensioning and network layout for:

Dense area; Ultra Dense area; Hotspot area.

Similarly to the MARSAL project, 5G-PHOS conducts analysis for different coverage areas based on the location of RAN equipment and the related fixed network to achieve xhaul. 5G-PHOS was focused on fronthaul (digital and analogue). MARSAL is more concerned by mid-haul and backhaul.

4.4.5 5G-ROUTES 5G-ROUTES [44] plans to conduct field trial of most representative and innovative CAM applications seamlessly functioning across a designated 5G cross-border corridor spanning across 3 EU member states borders in order to validated the latest 5G features and 3GPPP specifications under realistic conditions, so as


to accelerate the widespread deployment of 5G E2E interoperable CAM ecosystems and services in digitised motorways, railways and shipways throughout Europe. The cross-border use cases design and test by 5G-ROUTES are:

Automated Cooperative Driving; Awareness Driving; Sensing Driving; Uninterrupted infotainment passenger services on the go; Multimodal services.

Similar to VITAL 5G, 5G-ROUTES require high bandwidth and low latency for assisted driving. Similarly, MARSAL architecture and cell-free network can play a significant role in this.

4.4.6 MONB5G MonB5G [45] aims to create a hierarchical, fault-tolerant, automate data driven network management system that incorporates security as well as energy efficiency as key features, in order to orchestrate a massive number of parallel network slices and significantly more diverse types of services in an adaptive and zero-touch way [46]. MonB5G will demonstrate it through:

Zero-Touch Network and service management with end-to-end SLAs: Zero-Touch multi-domain service management with end-to-end SLAs; Elastic end-to-end slice management.

AI-assisted policy-driven security monitoring & enforcement: Attack identification and mitigation; Robustness of learning algorithms in the face of attacks.

MonB5G can offer an alternative or complementary approach to MARSAL security and network management for 5G and B5G networks. It can be integrated with the MARSAL architecture to overcome any potential limitation each path may have.

4.4.7 5G-SOLUTIONS 5G-SOLUTIONS [47] concept is based on i) formulating a wide set of innovative, heterogeneous use cases spanning across five vertical industries in a complementary way, and ii) Provide the technological enablers for facilitating the execution of the field trials. The 5G-SOLUTIONS field trials are categorised into four leaving labs, each with its own Use Cases as below [48]:

Factories of the Future (FoF): Time-critical process optimisation inside digital factories; Non-time-critical communication inside factories; Remotely controlling digital factories; Connected goods; Rapid deployment, auto/reconfiguration, testing of new robots.

Smart Energy: Industrial Demand Side Management; Electrical Vehicle (EV) Smart Charging; Electricity network frequency stability.


Smart Cities & Ports: Intelligent Street Lighting; Smart Parking; Smart city co-creation; Smart buildings / Smart campus; Autonomous assets and logistics for smart harbour/port.

Media & Entertainment: Ultra High-Fidelity Media; Multi CDN selection; On-site Live Event Experience; User & Machine Generated Content; Immersive and Integrated Media and Gaming; Cooperative Media Production.

5G SOLUTIONS will investigate four different leaving labs with applications ranging from smart cities to media consumption and entertainment with the current 5G networks. The MARSAL project can take this a step further by enhancing the QoS for using B5G/6G networks and MARSAL architecture.

4.4.8 5G-EPICENTRE 5G-EPICENTRE [49] aims to deliver an open end-to-end experimentation 5G platform focusing on software solutions that serve the needs of Public Protection and Disaster Relief (PPDR). Towards the final aim, 5G-EPICENTRE will demonstrate its functionality on eight Use Cases:

Multimedia Mission Critical Communication and Collaboration Platform; Multi-agency and multi-deployment mission critical communication and dynamic service scaling; Ultra-reliable drone navigation and remote control; IoT for improving first responder’ situational awareness and safety; Wearable, mobile, point-of-view, wireless video service delivery; Fast situational awareness and near real-time disaster mapping; AR and AI wearable electronics for PPDR; AR-assisted emergency surgical care.

MARSAL can implement the security solutions of 5G EPICENTRE for data protection and anonymity in its network while providing the flexible and elastic network required for the 5G EPICENTRE.

Above, the Use Cases and proof of concepts for 5G networks presented. At the same time, a new and exciting group of projects targeting beyond 5G and 6G networks have launched under the umbrella and scope of 5G PPP, including the MARSAL project. Latency and bandwidth are ever higher requirements for the Use Cases of those projects, offering a root for MARSAL network architecture to create a higher impact. A selected list of projects and their Used Cases are listed below, to present the versatility of beyond 5G networks.

4.4.9 6G BRAINS 6G BRAINS [50] aims to bring AI-driven multi-agent Deep Reinforcement Learning (DRL) to perform resource allocation over and beyond massive machine-type communications with new spectrum links including THz and optical wireless communications to enhance the performance with regard to capacity, reliability and latency for future industrial networks. The use cases for 6G BRAINS will focused over two pillars:


The first use case represents the offloading of the control logic from the industrial controller to a more centralised computing area in a virtualised form as a virtual machine or a container;

The second use case is represented by wireless video cameras to send high quality and high frame rate video to an image analysing system located at the “factory edge”.

Contrary to MARSAL, 6G BRAINS aims to deliver a future network system to an industrial level. While there is a difference in the target application, the method remains similar through network management (through smart contract network slicing or AI-driven multi-agent DRL) for flexible and adaptable networks based on the requirement of the tome.

4.4.10 AI@EDGE The goal of AI@EDGE [51] is to achieve tan EU-wide impact on industry-relevant aspects of the AI-for-networks and networks-for AI paradigms in beyond 5G system. The AI@EDGE will be validated using four Use Cases:

Virtual validation of vehicle cooperative perception; Secure and resilient orchestration of large (I)IoT networks; Edge AI assisted monitoring of linear infrastructures using drones in BVLOS operation; Smart content & data curation for in-flight entertainment services.

AI@EDGE aims to deliver a new smart network architecture with an emphasis on flexibility and security. Similar to MARSAL, AI@EDGE intents to design a new type of network based on edge computing. To an extend it validates the MARSAL approach and adopts the idea for the B5G/6G network’s structure.

4.4.11 HEXA-X The main goal of the HEXA-X project [52] is to define a 6G vision for the world of 2030, which tightly interlinks the human world of our senses, bodies, intelligence and values; the digital world of information, communication and computing; and the physical world of objects, organisms and processes. During HEXA-X lifetime, five demos will be presented:

6G OTA-Waveforms in action; FED-XAI-Federated XAI; Flexible topologies (FLEX-TOP) for efficient network expansion; Extreme performance in handling unexpected situations in industrial contexts; Algorithms for data-driven device-edge-cloud continuum management.

HEXA-X has the vision to define the 6G networks of the future. MARSAL project can be part of this vision through the network architecture developed for augmented reality and machine to human interaction.

4.4.12 5G-TOURS The use cases addressed by 5G-TOURS [53],[54] revolve around the life in a city. The focus is on improving the quality of life of the citizens as well as the experience of the tourists visiting the city, ultimately making the city more attractive to visit, more efficient in terms of mobility, and safer for everybody. To achieve this vision of 5G-driven quality of life improvements for tourists and citizens, the use cases addressed by 5G-TOURS are grouped around three main themes that represent different aspects of the city:

1. Touristic city: Visitors to museums and outdoor attractions use 5G-enabled applications to enhance their experience while visiting the city. These include VR/AR applications to complement the


traditional physical visit with additional content, including interactive tactile communications. The experience of the visitors is also enhanced with robot-assisted services, telepresence to allow for remote visits, as well as live events enabled by mobile communications such as multi-party concerts and content distribution.

2. Safe city: To provide means to better assist health-related care in all the phases of an incident, ranging from health monitoring for prevention and early detection, to diagnosis and intervention at the ambulance, and surgery at the wireless operation room in the hospital.

3. Mobility-efficient city: This involves smart cities, gathering information about the city and using it to improve navigation systems as well as parking. Traveling is also made more enjoyable by providing AR/VR services to passengers, and airports become logistically more efficient by relying on 5G for their operation. The mobility efficient city brings 5G to users in motion as well as to transport- related service providers.

The use case covered in 5G-TOURS can be further improved with the introduction of Beyond 5G concepts like MARSAL’s distributed cell-free massive-MIMO networks supporting massive AP deployments, serving an unprecedented plethora of users (or tourists) in touristic zones at the same time. Besides, MARSAL’s policy-driven security, privacy, and trust in multi-tenant infrastructures will support the decentralized collaboration, via smart contracts, of non-trusted tenants without them having to share their business or operational data. This will boost next generation user experience with improved data security and thread detection.

4.4.13 5G-MEDIA The 5G-MEDIA project [55] aims at innovating media-related applications by investigating how these applications and the underlying 5G network should be coupled and interwork to the benefit of both: to ensure the applications allocate the resources they need to deliver high quality of experience and so that the network is not overwhelmed by media traffic [56]. The project targets three well-defined use cases:

1. Immersive media: Tele-Immersive (TI) applications are immersive media network-based applications that enable multi-party real-time interaction of users located in different parts of the globe, by placing them inside a shared virtual world.

2. Mobile Contribution, Remote and Smart Production in Broadcasting: Due to the steadily rising cost pressure, broadcasters are looking for new, low-cost, and time-saving production methods, which include participatory and user-generated media archives in the production. These production methods are known under the term smart production. Remote production is a subgroup of smart production where the on-site production is handled remotely from the broadcaster’s facilities. In a remote production, the control room is on a fixed location, usually in the facility of the broadcaster. The control of the equipment on the venue itself happens remotely from this room.

3. UHD over CDNs: This use cases covers the access to streaming of UHD media services through various personal devices, both fixed and mobile, while the user is on the move in the 5G network. The focus is on how UHD contents by a Media Service Provider (MSP) can serve users on the go and how the MSP can build media distribution service chains made of software defined media functions to properly serve users attached to the 5G network.

With respect to media-related applications, MARSAL’s Elastic Edge Computing paradigm with Cloud-Native technologies aims to offering zero perceived latency and smart connectivity applications, which are critical in immersive media. Also, MARSAL’s self-driven infrastructure with pervasive ML-driven control will leverage the use of Analytic Engines at all tiers of the Edge infrastructure, and Decision Engines at the two Core-Tier


orchestration subsystem. Such approach enables the utilization of effective multi-objective optimization in media-related services, which can translate from improved user experience to reduced infrastructure OPEX.

4.4.14 5G-COMPLETE 5G-COMPLETE [57] aims to revolutionize the 5G architecture, by efficiently combining compute and storage resource functionality over a unified ultra-high capacity converged digital/analog Fiber-Wireless (FiWi) Radio Access Network (RAN). The project targets five well-defined use cases:

Smart Energy Metering deployed over 5G-COMPLETE Architecture 5G System over Multi-domain Transport Infrastructure 5G Wireless Transport services with MEC capability provided to Network Operators Advanced Surveillance/ Physical Security Services Augmented Reality/Virtual Reality (AR/VR) Service

Similarly to the MARSAL project, 5G-COMPLETE conducts studies about software RAN and Core functions in relation with fixed network technology to achieve xhaul. The issue of location of computing in different nodes and network controller are also discussed in these two projects.

4.5 Industrial initiatives Although MARSAL will tackle only a few verticals in its PoCs, we cover next different industrial initiatives for the sake of completeness. These initiatives are sharing their vision on how to realize innovative and economical 5G (and beyond) services, from which we highlight NGMN, ATIS, 5G ACIA, and NEM. We briefly summarize the use cases proposed by each of them.

4.5.1 NGMN The vision of the Next Generation Mobile Networks (NGMN) Alliance is “to provide impactful industry guidance to achieve innovative and affordable mobile telecommunication services for the end user with a particular focus on supporting 5G’s full implementation, Mastering the Route to Disaggregation, Sustainability and Green Networks, as well as starting work on 6G” [58]. In their latest whitepaper [59], a bunch of vertical industry examples are envisioned, from which tourism, and especially, smart cities are verticals covered in MARSAL’s PoCs.

1. Manufacturing industry: 5G will provide more flexible and convenient wireless connections meeting industry requirements through enablers like low-latency, high-reliability and high-availability connectivity, mobility, and precise positioning of all the devices (e.g., sensors and actuators) for real-time monitoring and control of processes, and end-to-end logistics and asset tracking.

2. Construction industry: 5G will support an autonomous smart system for construction projects by providing, low-latency, high-reliability and high-availability connectivity for sensors and controllers, with a high density in a given geographical area; high-quality real-time video transmission; high-precision location of equipment and humans; local (edge) computing resources for controlling e.g., movement of vehicles.

3. Transport: 5G will provide efficient, safe, environmentally friendly, and comfortable transportation, especially by exploiting the potential of artificial intelligence, to achieve connected and automatic driving through perception, decision and control. NGMN points out different key enablers such as the transmission of high-quality video or images of road condition and roadside facilities to help navigation, remote and automatic driving, as well as identification of blind zones and other vulnerabilities for vehicles.


4. Health: The health industry requires a balanced allocation of medical resources, portable and intelligent medical equipment, improved medical vehicle treatment capabilities, and the transformation of surgical operations from the operating room to multiple regions. Most hospitals currently use wired connection medical equipment, which is reliable but complex and inflexible. There is also an increasing need to provide remote health assistance, in particular to assist the fragile and elderly or to manage emergency situations. 5G can provide more flexible and convenient wireless connections, which can meet most health industry application scenarios, for example wide-area continuous coverage for ambulances, including sending live high quality video and patient vital signs in real time to the command centre in the hospital, or sensors collecting vital signs from the wearable devices of patients or the elderly, wherever they are, helping remote medical staff make timely treatment decisions and administer medication remotely.

5. Smart Cities and Communities: NGMN propose to 5G key enablers to enhance information collection and processing capabilities, to improve urban security governance capabilities, to integrate various intelligent applications for communities, and to improve government efficiency and the quality of life: 1) high quality video transmission, positioning and tracking to improve security monitoring efficiency, and 2) networked intelligent sensors, including video, to realise urban environmental monitoring, supporting enhanced integrated city management and various citizen services such as traffic and transport management, and resource management and planning.

6. Education: Communication technologies enable new educational models and applications, which in turn address the imbalance in the availability of general education and education resources. For the education industry, 5G will enable, among others, virtualised and augmented teaching, using virtual reality and augmented reality services to enhance and animate traditional approaches, or remote interactive teaching, delivering a synchronous experience and remote interaction between teachers and students from different schools, as well as the inclusion of children who would not otherwise be able to attend school.

7. Tourism: The tourism industry can use new 5G network capabilities to deliver more immersive, interactive, and exciting experiences and provide much more in-depth knowledge about the visited region, and its sites, attractions, and facilities. Tourist spots tend to be much more crowded than other locations. For instance, thanks to the higher data throughput and capacity of 5G networks, large numbers of visitors can benefit from augmented reality, offering, for example, a 3D virtual reconstruction of archaeological sites, thematic tours in museums, and annotated city tours enhancing the experience of a city with architecture and historical information.

8. Agriculture: Smart farming constitutes a revolution in the field of agriculture providing a tremendous productivity improvement. The enhanced capabilities of the 5G system support several enablers of smart farming, including agricultural robots such as driverless tractors, precision seeders, automated weed and pest controllers and automated harvesters, which will make it possible to produce more and higher quality crop with less manpower. Besides, UAVs or drones providing a bird’s eye view for imaging, planting and crop spraying to help optimise land and crop management.

9. Finance: The finance industry requires a secure, reliable, and widely adopted platform to enable advanced financial services. 5G will provide such a platform and accelerate the digital transformation for banks, boosting ubiquitous banking operations and delivering better customer service. 5G offers an array of capabilities that can be matched with the finance industry’s needs, including gigabit data rates that enable collection of a vast amount of high-frequency stock market data for fast analysis, ultra-reliable and low latency communication that enables fast response and control on high-


frequency mobile trading, and ensures time-critical buying and selling transactions take place at the edge of the mobile network by using multi-access edge computing technology.

4.5.2 ATIS The Alliance for Telecommunications Industry Solutions (ATIS) “brings together the top global ICT companies to advance the industry’s business priorities” [60]. ATIS is the North American Organizational Partner for the 3rd Generation Partnership Project (3GPP), a founding Partner of the one M2M global initiative, a member of the International Telecommunication Union (ITU), as well as a member of the Inter-American Telecommunication Commission (CITEL). The ATIS 5G Vertical Enablement Platform (5GVP) focus group created a landscape of the 5G-enabled vertical requirements [61]. A survey of ATIS members identified eight types of enterprise vertical use cases that were of priority interest for members.

1. Industrial and machine manufacturing: three use cases are defined. First, 5G non-public networks can be desirable for high QoS, high security, or isolation from other networks, as a form of protection against malfunctions in the public mobile network. Second, site safety monitoring through aerial intelligence provides a reality-modelling platform for physical asset management, addressing business requirements in various industry sectors. Third, AR-assisted factory maintenance will enable remote-assisted support to workers at distant locations, such as providing guidance on maintenance and repairs. This is expected to reduce downtime, increase output, improve safety and, in the long term, enable a more sustainable manufacturing process.

2. Connected vehicle: 5G will support the development of telematics to improve safety, traffic management and in-car entertainment, including 4K video, real-time navigation and AR for drivers and passengers. To fully exploit this immersive experience, data derived from the vehicle, surrounding infrastructure, surrounding environment and passengers will be augmented to provide personalized services wrapped around the connected vehicle use case.

3. Smart cities: ATIS defines two smart cities use cases. First, in smart traffic management, the combination of 5G network advantages, RFID transponders and cloud infrastructure make it possible to create a city traffic monitoring system that helps drivers reach their destination in an optimal time. Second, in citizen movement, 5G will be able to continuously ingest HD video streams with crowd-sourced sensor data to enable smart cities to analyse citizen movement and behaviour to support several services, including traffic control, public transport, retail and commercial planning, lighting, public safety and commercial vehicles for utilities and deliveries.

4. Public safety: three uses cases are defined. First, real-time video surveillance can provide collaborative intelligence that enables first responders to be better prepared and more effective in assessing a situation. Second, firefighter AR headset will use thermal imaging to let firefighters see through smoke, toxic gases, and darkness to find victims and colleagues, and spot falling objects and holes in floors that would otherwise be invisible. Third, wearables and connected vehicle sensors will enable the provision of enhanced criminal and/or patient insights. Police and paramedics will have clothing that can provide real-time video feeds and other sensor-related data about their immediate environment.

5. Healthcare: With telemedicine and remote patient monitoring, 5G eMBB will potentially reduce the number of in-person doctor visits by making virtual care more effective and lifelike. This alternative is particularly valuable for patients who cannot easily travel to their health care providers. Also, connected ambulances are vehicles equipped with advanced multi-technology and communication services that use a 5G network and dedicated communication to receive specialized remote support in real time via HD video while caring for the patient aboard.


6. Media and entertainment production: In broadcast virtual network operator use case, the customer becomes a broadcast virtual network operator (BVNO) that will use the 5G network to acquire all the resources it needs for TV production within a limited timeframe for a live program broadcast. The BVNO model also is a good fit for pandemic and post-pandemic hybrid workflows, where many producers now work from home over 5G low-latency networks. The, with mobile and social news gathering, journalists and TV program creators virtually reach their audience and engage with them on live TV, connecting to contributors, professionals, and others to reach high visibility, as well as enable editors to easily find the right person to be in contact with.

7. Media and entertainment consumption: First, the off-site major event immersive experience, expects 5G networks to broadcast at scale to millions of consumers concurrently, providing an “At Event” fully immersive interaction from a remote viewing location through mobile devices, 4K TVs or AR/VR glasses and headsets. Second, on-site live event experience focuses on large-scale event venues, such as theatres, stadiums, and ballparks to provide customers with a variety of options, such as instant replay, choosing a camera feed and AR. Finally, 5G subscription-less sponsored service bundles enable consumers without a 5G subscription to connect with a sponsored service bundle (i.e., advertisement content) in return for 5G service/access connectivity via a communication service provider (CSP).

8. Education: ATIS envisions three education use cases. First, immersive lessons with AR and VR can make the learning process more fun and much more interesting. This also can bring new experiences for distance learning, enabling the virtual presence of students. Second, virtual classroom with tactile interaction will create new ways of tele-teaching and tele-mentoring, especially for manual training and skill development. Third, personalized assisted learning will connect each learner to intelligent, personalized systems. These can suggest learning pathways, enable aggregated analysis and, through better data capture of learner experiences, enable much better decision-making about all aspects of a student’s education.

MARSAL’s PoCs mainly cover three of the verticals above: media and entertainment production, media and entertainment consumption, and smart cities. In particular, MARSAL’s PoC 1 covers dense UGC distribution and ultra-dense video traffic delivery, coping with media broadcasting and event immersive experiences, respectively. Besides, PoC 2 explores next-generation sightseeing through augmented related, which may apply both to the media verticals and to the ATIS smart city “tourism movement” use case.

4.5.3 5G-ACIA (FOR INDUSTRIAL DOMAIN ONLY) The 5G Alliance for Connected Industries and Automation (5G-ACIA) is the central global forum for shaping 5G in the industrial domain [62]. The paramount objective of 5G-ACIA is to ensure the best possible applicability of 5G technology and 5G networks for industry, particularly discrete and process manufacturing. So, even though MARSAL does not explicitly cover industrial use cases in its PoCs, MARSAL’s advances leave the door open for further application to the industrial domain. In particular, seven use case categories are listed by 5G-ACIA in [63], identifying real-time (RT) requirements, which can be:

1. Non-RT: Cycle times and latency are not critical; several seconds are regarded as sufficient. 2. Soft RT: Cycle times and latency are moderately critical, i.e., approximately one second. 3. Hard RT: Cycle times and latency are highly critical, to within milliseconds or even microseconds.

The industrial domain use cases are as follows:

1. Connectivity for the factory floor (Hard RT). 2. Seamless integration of wired and wireless components for motion control (Hard RT).


3. Local control-to-control communication (Hard RT). 4. Remote control-to-control communication (Soft RT). 5. Mobile robots and automated guided vehicles (AGVs) (Soft RT). 6. Closed-loop control for process automation (Soft RT). 7. Remote monitoring for process automation (Non-RT).

4.5.4 NEM (FOR MEDIA AND CONTENT ONLY) As mentioned in the previous sections about the NGMN and ATIS initiatives, MARSAL PoCs primarily cover media verticals. That is why, initiatives like NEM (New European Media Initiative) are also of interest. NEM was established as one of the European Technology Platform under the Seventh Framework Programme, aiming at fostering the convergence between consumer electronics, broadcasting, and telecoms to develop the emerging business sector of networked and electronic media. NEM defines nine relevant use cases in the media and content domain that will be used to study the requirements that the future 5G media slice will be able to support [64]:

1. Ultra-high-fidelity imaging for medical applications. 2. Immersive and interactive media. 3. Audio streaming in live productions. 4. Remote, cooperative, and smart media production incorporating user generated content. 5. Professional content production. 6. Machine generated content. 7. Collaborative design including immersive communication. 8. Dynamic and flexible UHD content distribution over 5G CDNs. 9. Smart education.


5 MARSAL Use Cases Description This section is focused on the description and definition of the MARSAL PoCs. First of all, we present the MARSAL testbeds, which will be used for the PoCs and the demonstration phase. Then, we focus on the main PoCs. In particular, we present the considered experimentation scenarios and the involved stakeholders. We also map the scenarios with the MARSAL architecture, and we consider the testing and evaluation process.

5.1 MARSAL Testbeds 5.1.1 KUL CELL-FREE MIMO TESTBED (LEUVEN, BELGIUM) The KU Leuven Cell-free massive MIMO testbed consists of 32 National Instruments (NI) USRPs (2942R/USRP RIO) connected with a MIMO processor responsible for data processing or storage (Figure 3). Each USRP (also called sometimes Remote Radio Head – RRH by NI) can represent a two-antenna AP (max 64) or two APs. Several USRPs can be also put together to create a larger AP. Note that antennas can be distributed over an area. Currently we can cover an indoor environment with a significantly high density of cell-free access points. We can also use the testbed for collecting data in an outdoor environment. In this case, we can divide 32 USRPs into two sub-systems, where each of them supports 32 antennas. The two subsystems are connected to the MIMO processor with two 10-meter optical fibers responsible for the backhaul In-phase and Quadrature (I/Q) components processing/storage. Note that since one of the sub-systems is bundled to the MIMO processor in the same rack, the maximum separation of the two sub-systems is 10 m (we are planning an extension to 100 m in Q3 2021).

Figure 3: KUL cell-free MIMO testbed

Let us describe the two main components of the testbed: MIMO processor and APs connected with Data distributor (switch). A master chassis embeds a x64 controller which runs LabVIEW on a Windows 7 64-bit OS and serves three primary functions: (i) it provides a user interface for radio configuration, deployment of FPGA bitfiles, system control, and visualization of the system, (ii) it acts as source and sink for the user data—e.g. HD video streams—sent across the links, and (iii) the MIMO processor measures link quality with metrics such as BER, EVM, and packet-error rate (PER). It connects to four switches in a star fashion. Switches yield no processing but allow data to be transferred between USRPs. The USRPs can work with 40 MHz bandwidth signals in sub-6 GHz frequency band.


5.1.2 ORANGE EXPERIMENTAL PLATFORM (LANNION, FRANCE) The experimental platform that is located at the ORANGE (Lannion) premises is used in PoC1 to evaluate the transportation of a given mobile interface through different sorts of optical access networks and provide intelligent management of such networks to optimize the mobile KPIs. The test bench can be divided into 4 distinct parts (Figure 4). At the first part, the existing vRAN radio plane runs on a virtualized server in a CentOS Openstack virtualization environment with real-time vCUs and vDUs connected through a PDCP/RLC V1 interface. Such interface goes out of the server over Ethernet and through the fixed aggregation and access networks before looping back to the same server, where high layer functionalities of the UE are also implemented. Even though the various mobile nodes are installed on the same server, they are logically separated and can only communicate via the existing mobile interfaces. The second part of the test bench refers to the emulation of the aggregation network. This is done with a network impairment engine that can introduce latency, packet jitter and BER to the RAN interfaces. Those can also be aggregated to other mobile packet-based flows by means of a switch to provide higher bitrates. The RAN flows can be distinguished with different VLAN tags. The third main block of the test bench corresponds to a specific optical access transmission system through which the RAN interface will be transmitted. Different topologies and transmission approaches can be seamlessly integrated to the test bench, spinning from standard point-to-point (PtP) connections with or without wavelength division multiplexing to very disruptive point-to-multipoint (PtMP) approaches based on SFP-OLT modules. Such devices allow PtMP connectivity using a smart XGS-PON SFP+ transceiver and can be connected to any type of layer 2 or 3 network devices such as switches, routers and servers. The final block concerns the software-based abstraction and intelligence associated to the underlying optical devices. Our SDN controller abstracts and controls the switching fabric and the PtMP devices (OLT and ONUs). Moreover, dynamic management of backport interfaces of the OLT (switch + SFP-OLT modules) can be supported via our controller.


Figure 4: ORANGE experimental platform

5.1.3 CTTC EXPERIMENTAL PLATFORM (BARCELONA, SPAIN) The 5G testbed, located at the CTTC premises, is based on OpenAirInterface, Amarisoft 5G, and ETSI OSM MANO. It leverages a Cloud Radio Access Network (C-RAN) architecture with a 5G core and a fully virtualized 5G RAN. Figure 5 illustrates a high-level view of the 5G testbed. Amarisoft Callbox Ultimate is used as 3GPP compliant gNodeB and 5G Core (5GC). Likewise, white box servers can be used to implement all RAN functionalities, i.e., the 5G gNB, Remote Radio Head (RRH), and Base-Band Unit (BBU) with the OAI 5G software stack. Moreover, the 5G testbed supports Fronthaul based on 10 GigE, allowing the 5G RAN functions to be split between the BBU and the RRHs. The RRH units are implemented with Small Form Factor PCs that are connected to programmable radios e.g., USRP X310, which implement all RF functions. Alternatively, 5G gNBs are connected directly to the 5G Core via 10 GigE backhaul connections. The 5G testbed supports all Traffic Types, including eMBB and URLLC for high bandwidth / low latency applications, as well as mMTC traffic associated with document Ref. Ares (2020)7177041 - 29/11/2020 from NB-IoT and LoRA sensors and actuators.


Figure 5: CTTC experimental platform

The testbed's Mobile Edge tier incorporates a virtualized data centre based on OpenStack, which handles part of the 5G baseband processing and also hosts VNFs and application services. An orchestration layer based on ETSI OSM is employed for VNF onboarding and orchestration, allowing the full automation of operational processes and tasks related to the placement and lifecycle management of all services. As for SDN, the CTTC testbed leverages various controllers, including FlexRAN and Amarisoft, OpenDaylight and Neutron, and Service Network Mesh to: i) provide dynamic RAN-related modifications and measurements, ii) implement end-to-end slicing, and iii) provide service-level network management, respectively. Complete dynamic end-to-end slicing support is expected to be supported after an upgrade to 3GPP R.17.

The 5G RAN-Core segment is implemented with the industry Amarisoft 5G Private Node, which provides 5G RAN/Core capabilities as well as management northbound interfaces (NBI) out-of-the-box, enabling the establishment of 5G end-to-end services. Amarisoft NBI exposes a range of telemetry and actuation (e.g., TX/RX gain adjustments, frequency, MCS, etc.) via web sockets, which can be leveraged for emulating a wide variety of scenarios and management algorithms. Edge (i.e., ETSI MEC) and Cloud 5G segments are enabled thanks to Virtualized Infrastructure Managers (VIM) such as OpenStack and Kubernetes. The former is typically used to provide Infrastructure as a Service (IaaS), while the latter helps deploy MEC architecture, state-of-the-art cloud-native applications, and dynamic infrastructure/service support components (aligned with ETSI NFV IFA 029).

The testbed is ready to host the experiments of the PoC2 and it will be upgraded with the latest 3GPP R.17 compliant 5G core, allowing tighter integration of the MEC hosts with the 3GPP network via Local Area Data Network (LADN) reference points. The latest R.17 RAN slicing specifications will also be incorporated at the gNBs.


5.2 PoC1: Cell-free networking in dense and ultra-dense hotspot areas During high-popularity events, both indoors (e.g., music concerts) and outdoors (e.g., a fair), a large number of users tend to stream high volumes of content from multiple handheld devices, thus creating a heavy burden to the network infrastructure both in the uplink and in the downlink. Ericsson shows the practical implications of cell-edge users’ effect for 4G cellular networks where 10 Mbps data rate is available at the center of the cell, at mid-cell drops by 10X to 1 Mbps, and by the cell edge it drops another 10X to 0.1 Mbps [65]. From the example, it is reflecting that the cellular architecture has a 100:1 data rate ratio from cell center to cell edge. Similarly, the concept of network densification that involves small cells with mMIMO capabilities for 5G networks (i.e., a large number of transmitting and receiving antennas with beamforming and precoding techniques) is still limited by inter-cell interference, with performance at the cell edges being a particular concern [12], [66], and [67]. For an example, the small deployment by the assumption of 1 AP serving to 1 user with uplink (UL) power control and pilot assignment can achieve 3 Mbits/s as per-user UL network throughput [12]. Unfortunately, the concept of network densification (very beneficial in multiple aspects) results in a situation where more users become cell-edge. The current SoTA solutions including traditional cellular mMIMO communication systems suffer from severe inter-cell interference negatively affecting service quality for cell-edge users.

In Cell-free massive MIMO, users are simultaneously served by multiple cooperating APs (with a low number of antennas) instead of associating each user terminal to a cell with a gNB equipped with a large number of antenna elements. It relaxes the restriction of cell boundaries, which can significantly reduce or even eliminate inter-cell interference. Therefore, the cell-free mMIMO shows 2.5X improvement over small cell in terms of per-user UL network throughput and it can achieve the performance up to 8 Mbits/s [12]. This makes cell-free networking, an emerging 6G technology, extremely suitable for hotspot areas, as it can offer seemingly infinite capacity and fully mitigates data rate problem of the cell edge users. Like 5G the Small-Cell Network (SCN) with the concept of non-cooperative base stations can only serve up to 200-meter cell-radius, reduce power in signal transmission up to 10 Watts, and achieve the mean spectral efficiency 3 Mbits/s. The network densification through the small-cell deployment even with cellular MIMO still reduce the spectral efficiency performance. This is because the interference generated by the neighboring cells, increasing numbers of uncoordinated and lightly loaded small cell in the network, and in-depth analysis of small cell deployment [68]. We will overcome the Inter-Cell Interference (ICI) problem and uncoordinated deployment of such SCN by the distributed cell-free RAN solution of MARSAL where the dynamic adaptability algorithm will enable AP-DU and DU-DU coordination. Such coordinated solution can further improve the system capacity/spectral efficiency of future 6G networks. However, the cell-free massive MIMO deployment for future 6G network can achieve significantly better performance, since each user can be served by the dedicated APs. Consequently, the PoC1 in MARSAL will investigate the benefits of such cell-free mMIMO concept by disaggregating the traditional cell-free CPU in DUs and a CU in line with the 3GPP NG-RAN disaggregation is already defined in the 3GPP TS 38.401 [69]. The MARSAL’s cell-free innovations which is envisioned for 6G networks will be implemented with existing vRAN elements (e.g., vDU, vO-CU) integrate with O-RAN Near-RT RIC for the first time and provide PoC1 demonstration with an improved performance of the baseline scenarios. Such testcase scenario, KPI definition, targeted improvement to satisfy the requirements of future 6G networks is shown in the later section of the deliverable. The PoC1 in MARSAL with Lannion’s platform will be evaluating MARSAL’s optical wireless cell-free infrastructure, with a distributed serial fronthaul solution and XGS-PON backhauling. The PoC1 innovations of a converged Fixed-Mobile network will deploy distributed cell-free RAN in Serial fronthaul topology for a pre-recorded video content from the regional Edge node.


5.2.1 EXPERIMENTATION SCENARIO 1.1: DENSE USER-GENERATED CONTENT DISTRIBUTION WITH MMWAVE FRONTHAULING

5.2.1.1 INTRODUCTION

The main objective of this scenario is to demonstrate and evaluate MARSAL’s distributed cell-free RAN in terms of increased capacity and spectral efficiency gains, and the adaptivity of dynamic clustering and RRM mechanisms in managing connectivity resources in a dynamic environment with varying hotspots areas (experimentation scenario 1.1). Furthermore, an additional objective of this scenario is to evaluate the Hybrid MIMO Fronthaul in terms of its ability to offer a dynamic AP topology.

The experimentation scenario 1.1 will show the potential of deploying cell-free networking in 6G networks with massive AP deployments. The MARSAL innovations in Scenario 1.1 will focus on distributed processing, with clusters of APs and DUs coordinating via fronthaul links. In this experimentation scenario, we will evaluate the performance of dynamic data driven clustering algorithm, as shown in Figure 6. It will also explore and evaluate inter-DU coordination effect in Spectral Efficiency and propose dynamic adaptability of the coordination levels jointly addressing AP-DU and DU-DU coordination for the first time. The Cell-free vRAN components in this experimentation scenario will validate the design of cell-free enable vO-DU, cell-free MAC scheduler, PHY sub-layer. Moreover, it will also validate the appropriate modification of the CP protocols and O-RAN specified interfaces (i.e., E2, O1) to support practical cell-free operation and fully distributed processing.

Figure 6: Dense user-generated content distribution with mmWave fronthauling scenario

In CF networks, the environment around the user defines the set of APs serving it. These APs may i) be connected to different O-DU and ii) utilize different types of fronthaul links. The latter, in its turn, defines the information to be shared between the APs and their O-DUs. Next, the involved O-DUs share necessary information with each other. The final MARSAL processing will also consider the fronthaul and midhaul constraints (feedback to the cluster formulation block).

5.2.1.2 DESCRIPTION

Ever increasing number or user devices and their traffic demands put a very high stress in the radio interfaces. Densification of the network is necessary; however, it causes high interference in conventional cellular networks. CF networks can eliminate the interference problem in dense outdoor deployments. Moreover, APs can be switched on\off depending on the network needs (i.e., the AP clusters can be formed following the users and their traffic demand). This solution becomes attractive if fronthaul links are less expensive than the fiber connections mostly used in current deployments.

The investigated experimental scenario represents an event with a high density of users and APs. It can be an indoor venue (e.g., a concert) or an outdoor setting (e.g., a football stadium). During this kind of events, it is common for dense UGC to be streamed by spectators via their handheld devices and consumed locally


in real-time. Moreover, there may be users with very different requirements: for instance, a UAV controlled by law enforcement or security agent vs a regular user/spectator. In general, this use case is characterized by high user density and the users generate high traffic.

In this scenario, APs will be interconnected with O-DU nodes serving the users in a coordinated manner. Furthermore, MARSAL’s Hybrid MIMO fronthaul solution will be leveraged for the interconnection of O-RUs and O-DUs. The performance of the cell-free NG-RAN will be evaluated via pre-recorded video content that will be uploaded and downloaded by UEs to/from a video streaming MEC app deployed at the Regional Edge node, to emulate Dense UGC streaming both in the uplink and in the downlink direction.

5.2.1.3 INVOLVED STAKEHOLDERS AND ROLES

We list below a generic description of the stakeholders involved in the MARSAL project:

Cell-free mMIMO infrastructure provider: An infrastructure provider is needed to validate the MARSAL solution, especially for a novel cell-free NG-RAN architecture. The relevant stakeholder, besides providing the infrastructure, can conduct research on the area of cell-free mMIMO and could be providing, e.g., cell-free clustering solutions. The cell-free clustering solutions are needed to cooperate with the CPUs and the O-DUs via algorithms that integrate the cell-free RUs with the O-DUs by using PHY API’s.

Cell-free vRAN VNFs provider: The cell-free vRAN VNFs provider from the industrial stakeholder will be designing the cell-free vRAN architecture. The stakeholder will be further implementing the solution on 3GPP protocol stacks and provide the VNFs. The VNFs can be implemented in radio edge DCs and/or regional edge DCs based on the underlying user’s requirements and their SLA. However, the solution provided by the stakeholders as a form of VNFs (i.e., the vO-DU and vO_CU-UP) can be deployed at the Radio Edge DCs as white-box servers.

RAN intelligent controller and xAPPs1 provider: The RAN intelligent controller and the xAPPs provider will be involving providing the cell-enabled solutions with the modifying the necessary O-RAN specified interfaces. The AI/ML models at the Near-RT RIC provided by the stakeholders will somehow provide the RAN functionalities of the underlying architecture. In the overall PoC1 demonstration, the stakeholder will be providing the resources of Non-RT RIC/SMO, Near-RT RIC, and RAN nodes, and relevant O-RAN specified interfaces to enable cell-free in the traditional O-RAN architecture. The stakeholder solution of the Near-RT RIC and vCU_CP can be deployed at a Regional Edge white-box server, and it can be further connected to the radio edge DCs through optical network unit from other stakeholders.

mmWave fronthaul provider: Stakeholders can enable scalability through the flexible utilization of a distributed cell-free approach and an mmWave fronthaul, by relying on optical technologies (PONs, WDM rings) for their interconnection with other components of MARSAL architecture. The mmWave transceivers and the beamforming solutions from the stakeholder can be integrated in Hybrid MIMO node that will be characterized via experiments (e.g., in terms of EIRP, beam pattern, Field of View and beam directivity).

mmWave transport network: The stakeholder will contribute to design of the 6G x-haul networks and provide SDN-enabled wireless transport for the interconnection of distributed vRAN components

1 Application available at RAN Controller for RAN Configuration, AI/ML model policy execution, radio resource

management model, slice selection, etc., as per O-RAN compliance approved specifications specified by O-RAN WG-1 and WG-3.


of the MARSAL architecture. The P2P mmWave solution of the stakeholders that can enable several deployment options (e.g., Co-located O-CU and O-DU: O-RAN split 7.2x, independent O-RU, O-CU, O-DU locations: O-RAN split 7.2x, and O-RU and O-DU integration on cell site: O-RAN split 2), respectively.

Network equipment manufacturer: Apart from traditional network devices (e.g., routers and switches), VNF-based devices must be designed and manufactured to support the MARSAL architecture, including DCs and MEC platforms.

In this experimentation scenario, KUL will offer the cell-free mMIMO testbed, integrated with the Hybrid MIMO fronthaul from PT, and will implement the dynamic clustering and cooperation algorithms. ISW will provide the vDU and vCU_UP VNFs deployed at Radio Edge white-box servers, ICOM the P2P mmWave solution and ACC the Near-RT RIC and vCU_CP deployed at a Regional Edge white-box server supporting ML driven RRM.

5.2.1.4 MAPPING SCENARIO TO THE MARSAL ARCHITECTURE

To support the abovementioned experimentation scenario, MARSAL CF NG-RAN will include cell-free clustering solution, mmWave fronthaul, CF MAC scheduler as a part of the CF vRAN to provide mMIMO support in the MARSAL architecture. The cell-free vRAN, especially, the vO-DU and vO_CU-UP will be deployed as VNFs at the radio edge data centre. The mmWave transport network will be providing distributed cell-free coordination amongst the VNFs deployed in radio edge DCs and regional edge DCs. A MEC platform be deployed both at the regional and radio edge data centers (DCs) and a pre-recorded video will used during the PoC evaluation. Moreover, Near-RT RIC will be hosted at the regional edge DCs, and it will relate to the XGS-PON of the MARSAL architecture.

Figure 7: Overview of the PoC1 scenario mapped into the MARSAL architecture

Figure 7 depicts the mapping of the scenario into the MARSAL architecture. To deploy the cell-free vRAN solution on the radio edge data centre we can consider the higher and lower layer split suggested by 3GPP, O-RAN, and small cell forum. Such flexible split options together with the utilization of a converged optical wireless network will allow us to deploy the cell-free vRAN solution in a cost-effective manner for future 6G


networks. Furthermore, the O-RAN introduces a near Real-Time RAN Intelligent Controller (RT-RIC), which implements radio resource management, measurement, and telemetry.

The proposed configuration deploys the vO_CU-UP and vO-DU at the Radio Edge, while the vO_CU-CP near-RT RIC is deployed at the Regional Edge. The split inside the physical layer, i.e., the 7.2 split will simplify the data mapping by limiting the required associated control messages and it will also allow us to enable transport bandwidth scalability based allowing usage of a higher number of antennas without asking for extra transport bandwidth. The O-RAN compliant vO-CU and vO-DU components are incorporated in the form of RAN Virtual Network Functions (VNFs), while the APs serve as the RUs.

5.2.1.5 TEST AND EVALUATION OF THE POC1 SCENARIOS

In the PoC1, we will evaluate NG-RAN solution which is consisting of the technology enablers of cell-free mMIMO and future fronthaul technologies to support the vision of 6G networks. According to the PoC1 description in Section 5.2.1.2, we will use the MARSAL architecture and provide the functional test of cell-free vRAN solution for mMIMO, data-driven CF MAC scheduler, CF clustering algorithm, mmWave fronthaul, and near-RT RIC with CF support, respectively. We will compare the performance of our solutions with the network densification concept of current 5G networks that involves with the baseline solution of cellular MIMO, small cell deployment where the performance is limited due to the ICI problem. This makes cell-free networking, an emerging solution for future 6G networks, extremely suitable for hotspot areas, as it can offer seemingly infinite capacity, while it can fully mitigate the cell edge challenges. The following test scenarios will be considered in the PoC1 to demonstrate the performance of MARSAL solutions and compare the performance of the proposed scenario with a baseline scenario:

Test scenario 1 (Data-driven AP clustering): Evaluate per user data rate/system capacity/ per user SE performance of the MARSAL data-driven cell-free AP clustering and compare the metrics with conventional small cell 5G networks (using equivalent number of antennas, mobile users, bandwidth etc.). Additional comparison with distance-based cell-free clustering will be performed to demonstrate the advantages of the AI-based method taking into account realistic computation and fronthaul constraints. We will evaluate the effectiveness of these practical constraints on channel capacity and spectral efficiency at the uplink and downlink. Note that this test scenario does not consider cluster re-formulation.

Test scenario 2 (Dynamic adaptability algorithm): Demonstrate MARSAL’s Dynamic Adaptability (DA) algorithms, showcasing their capabilities in selecting AP clusters based on the optimal AP-DU and inter-DU cooperation levels and how these are affected by fronthaul capacity constraints, and evaluate the effect of the point-to-point Xn-like interface in inter-DU cooperation. The baseline scenario is Test scenario 1, which investigates non-reconfigurable clusters.

Table 1: Preliminary KPI definition for the experimentation scenario 1.1 (Test scenario 1 & 2)

Baseline scenario (with traditional

network architecture)

Name of the KPI KPI value of the

baseline scenario

TS1: Targeted KPI (with MARSAL architecture)


Mean Spectral efficiency [256QAM, 75% capacity is

used for data]

3.19 bits/s/Hz 2X (bits/s/Hz) 3X (bits/s/Hz)


Cellular SISO (5G)

Per user data rate [256QAM, 75% capacity is used for

data]

7.65 Mbits/s 2X (Mbits/s) 3X (Mbits/s)

System capacity [256QAM, 75% capacity is used for

data]

459.14 Mbps 2X (Mbps) 3X (Mbps)

5G small cell Per-User UL Net Throughput [with UL power control and

pilot assignment]

3 Mbits/s 3X (Mbits/s) 3X (Mbits/s)

Traditional cell-free mMIMO

Per-User UL Net Throughput [with UL power control and

pilot assignment]

8 Mbits/s 2X (Mbits/s) 2X (Mbits/s)

Test scenario 3 (mmWave fronthaul: Beam-steering, multicasting, dynamic reconfiguration):

Demonstrate applicability and advantages of the new Hybrid MIMO solution for the fronthaul link and showcase optimized cluster formation leveraging on the adaptive topology. Baseline solution is eCPRI. The first phase will consist of demonstrating that mmWave link satisfies throughput and delay requirements for fronthaul links of a static AP clusters (TS1). Next, we will demonstrate how the array can be reconfigured in order to provide fronthaul links to a dynamic cluster (e.g., the beamsteering directions will be changed since different AP will require the fronthaul link).

Table 2: Preliminary KPI definition for the experimentation scenario 1.1 (Test scenario 3)

Test Scenario

Baseline scenario (with traditional network

architecture) Name of the KPI

KPI value of the baseline scenario

Targeted KPI (with MARSAL architecture)

TS3

eCPRI Throughput 20 Gbps More than 20 Gbps

eCPRI Delay 250 us 250 us

eCPRI Number of APs per TX 1 (non-reconfigurable)

X10 (dynamic)

eCPRI Switching delay - 250 us

Test scenario 4 (Data-driven cell-free MAC scheduler): Evaluate per user data rate/system capacity/

per user SE performance of the MARSAL data-driven cell-free MAC scheduler. We will compare the performance our proposed solution with the baseline MAC scheduler algorithm (e.g., round robin algorithm) which is widely considered for 4G/5G networks [70],[71],[72]. We will compare our proposed scheduler performance with traditional architecture concept of 5G networks (e.g., small cell, cellular mMIMO, and traditional cell-free mMIMO). The data-driven MAC scheduler will be also adopted in the traditional network architecture of 5G networks to evaluate the KPI’s defined in Table 1. We will then compare the performance of MARSAL architecture which is envisioned for 6G networks and demonstrate the targeted KPI value by utilizing cell-free mMIMO infrastructures of KUL.



Test Scenario



Name of the KPI KPI value of the

baseline scenario Targeted KPI (with

MARSAL architecture)

TS4

Cellular SISO architecture with

Round Robin Scheduler

Mean Spectral efficiency [256QAM, 75% capacity is used

for data]

3.19 bits/s/Hz 2X (bits/s/Hz)

Cellular SISO with Round Robin

Scheduler

Per user data rate [256QAM, 75% capacity is used for data]

7.65 Mbits/s 2X (Mbits/s)

Cellular SISO (8 APs, 60 users), with Round Robin

Scheduler

System capacity [256QAM, 75% capacity is used for data]

459.14 Mbps 2X (Mbps)

5G Small Cell (1 AP serving to 1 user)

Per-User UL Net Throughput [with UL power control and pilot

assignment]

3 Mbits/s 3X (Mbits/s)

Traditional cell-free mMIMO

Per-User UL Net Throughput [with UL power control and pilot

assignment]

8 Mbits/s 2X (Mbits/s)

Test scenario 5 (Cell-free vRAN integration with near-RT RIC): To showcase an E2E functional test CF enabled vRAN components and near RT-RIC with CF support we will integrate the technological enablers discussed in Section 3 (Cell-free mMIMO networks and Fronthauling technologies for 6G networks) and connect the CF-enabled vRAN component with near-RT RIC over E2 interface. The current O-RAN specification does not provide any cell-free supported UCs to evaluate system capacity, per user data rate, per user SE, SE performances for 5G networks. To demonstrate the TS5, we will also consider the entities/resources suggested in Section 4.1, but we modify the RAN nodes (i.e., the vRAN components) of traditional O-RAN architecture and their corresponding interfaces to integrate the cell-free solutions in O-RAN which envisioned as one of the key technology enablers for future 6G networks. The data-driven closed-loop control implementation by using the O-RAN Software Community, i.e., near real-time RIC and a virtualized 5G protocol stacks with 4 BSs and 40 UEs in the dense urban scenario of Rome, Italy is investigated in [73] where a deep reinforcement learning (DRL) scheduling algorithm is used to maximize the spectral efficiency of the networks. As we have discussed earlier in TS4, we will also consider cellular SISO, small cell architecture of 5G networks and the traditional O-RAN architecture [14] as a baseline scenario to evaluate the performance of MARSAL solution. The feasibility of a closed-control loop is demonstrated in [73], where a deep reinforcement learning agent running in xApps on the near real-time RIC and select the best-performing radio resource scheduling policy for each RAN slice to evaluate the spectral efficiency performance. Under the TS5, we will demonstrate the cell-free vRAN and integrate it with near-RT RIC over O-RAN complaint E2 interface with CF support. The integration of cell-free in the traditional O-RAN architecture will be demonstrated under TS5 and we will further compare the spectral efficiency/system capacity/per user data rate performance compare.



Test Scenario

Baseline scenario (with traditional network

architecture) Name of the KPI O-RAN RIC architecture

(without CF support)

Targeted KPI (with MARSAL architecture and cell-free enabled

RIC)

TS5

Cellular SISO

Spectral efficiency [ML algorithm]

4 bits/s/Hz

2X (bits/s/Hz)

Spectral efficiency [Round Robin Algorithm]

2 bits/s/Hz

Spectral efficiency [Proportional Fair Algorithm]

4 bits/s/Hz

Spectral efficiency [Water filling Algorithm]

4 bits/s/Hz

5.2.2 EXPERIMENTATION SCENARIO 1.2: ULTRA-DENSE VIDEO TRAFFIC DELIVERY IN A CONVERGED FIXED-MOBILE NETWORK


This scenario will showcase MARSAL’s solution towards Fixed-Mobile Convergence in an ultra-dense indoors context (experimentation scenario 1.2) like campus, stadium, malls... Mobile clients served by a distributed Cell-Free RAN will be sharing the Optical Midhaul with third party fixed clients. The Fixed-Mobile Convergence in an Ultra-dense indoors scenario will be operated based on two operation modes (Figure 8): Fixed operation (Passive Optical LAN) and Mobile operation (small/pico cell with optionally Distributed Antenna System), respectively. The mobile clients served by a distributed cell-free RAN will be sharing the Optical Midhaul with third party FTTH clients. The optical fiber access equipment (Optical Line Terminal) will relate to PON and PtP interfaces where the Network urbanism organization, e.g., OLT and CU co-localized and DU at the end face of the optical termination.

Before beginning the description of this scenario, we would like to highlight the motivation of this work by considering the 5G status and the 6G needs. So first, 5G carriers and equipment are emitted and localized at the regular antenna locations with 2G/3G/4G. The pressure of coverage based on the requirements of the regulator is the main reason to have such deployment engineering rules. Concerning the mobile backhaul, 5G deployment coincides with a massive use of optical fiber to achieve the required backhaul throughput up to 10GEth. The fixed access network is based on PtP (Point-to-Point) topology to achieve the connectivity between antenna site and the first aggregation node (central office). Due to that in parallel of 5G, FTTH is under deployment, we have more and more central offices equipped with OLT (Optical Line Terminal) shelf. 5G backhauling could be addressed either by direct PtP connection to aggregation switch/router or through OLT PtP ports & cards. Now in order to address the increase of 6G cells, the preferred fixed technology to collect multiple spots is the PtMP also named PON (Passive Optical Network) based on “tree” fiber infrastructure. 6G transport challenges concern the coordination between RAN and FAN (Radio & Fixed Access networks) networks to address throughput, latency, availability issues.

5.2.2.2 DESCRIPTION

In large indoor campus (or stadium or mall), visual contents can be captured locally, and high number of users can be connected to real time or streaming videos, generating ultra-high-density video traffic. Figure 9


shows two fixed networks segment dedicated to xhaul: the frontahaul based WDM transmission technologies and the back-/mid-haul based on PtP and PtMP (like XGS PON) technologies.

Firstly, MARSAL’s distributed cell-free RAN Radio Unit will be deployed based on a Serial Fronthaul topology. The Serial Fronthaul allows a large number of cell-free APs to be interconnected in a bus topology, significantly increasing Spectral Efficiency, but with minimal cabling requirements. The technology of transmission considered to support serial fronthaul is Wavelength Division Multiplexing (WDM). A passive fiber network infrastructure is preferred using passive optical multiplexer. We have colorized transceiver at the end faces of this fronthaul networks. Hence, Serial Fronthauling is considered an ideal solution for indoor venues.

Secondly, in this scenario, Radio Edge nodes, that host the vRAN elements, are interconnected via PtP and PtMP (PON technology like G-PON (gigabit capable) and XGS-PON (10 gigabit capable)) midhaul links with the Regional Edge. The Regional Edge nodes, interconnected in a WDM ring topology, will host the Near-RT RIC and vCU_CP VNFs. The SDN-Transmisison controller and Near-RT RIC SDN function will also be deployed at the Regional Edge nodes. The performance of this scenario will be evaluated via pre-recorded 4K/HDR video that will be uploaded and downloaded by UEs to/from a video streaming MEC application deployed at the Regional Edge node. Fixed clients will be included as well in this scenario that will be served by the same infrastructure via PtMP (PON) links sharing capacity with the Radio Edge node.

Figure 8: Mobile operation based on fixed operation facilities for large indoor CAMPUS (or stadium, mall,

etc.) use-case


Figure 9: Experimentation set-up for the use case about content delivery in a converged Fixed-Mobile

network using PtP & WDM for fronthaul and PtP & PON for mid/back-haul


Optical access networks and fiber infrastructure provider: The stakeholder of this experimentation scenario 1.2 will be providing the ONU and optical fiber infrastructure that will connect the Radio Edge nodes of the MARSAL architecture which is hosting the vRAN elements. It will further enable an interconnection via PtP and PtMP (PON technology) midhaul links with the Regional Edge. The Regional Edge nodes of the MARSAL architecture will be interconnected in a WDM ring topology, will host the Radio Intelligent controller and the vCU_CP VNFs from the NVFs provider.

Radio intelligent controller provider: The stakeholder will be providing Radio intelligent controller including the Near-RT RIC, SMO, relevant O-RAN specified interfaces to enable cell-free in the traditional O-RAN architecture.

The RAN components provider: The RAN components provider will be providing the CU and DU as monolithic VNFs that will be integrated with the cell-free RUs (APs). Furthermore, the stakeholder of the VNFs providers will modify and implement necessary interfaces to enable cell-free NG-RAN for the PoC1 experimentation validation.

Network equipment manufacturer: Apart from traditional network devices (e.g., routers and switches), VNF-based devices, the ONU, optical fiber gateway must be designed and manufactured to support the MARSAL architecture.

In this experimentation scenario, ORANGE testbed, with a WDM ring interconnecting the Regional Edge Nodes, and PtP / PtMP links towards the Radio Edge nodes. ORANGE will also provide the SDTN controller. KUL will provide cell-free APs; each will be connected to a Radio Edge node where ISW’s vCU_UP and vDU VNFs will be deployed. ACC’s Near-RT RIC and SDN control function will be deployed at the Regional Edge.



To support this experimentation scenario, MARSAL ULTRA-DENSE VIDEO TRAFFIC DELIVERY IN A CONVERGED FIXED-MOBILE NETWORK will include an Optical Access Network to serve the connectivity between radio edge data centers (DCs), and cell sites equipped with DUs and RUs. We also propose a coordination between the controllers dedicated to Fixed and Radio access networks.

Figure 10 depicts the mapping of the scenario into the MARSAL architecture. It must be noted that we will also use existing and next-generation PON technologies, to take advantage of their lower CAPEX (due to higher reuse of commodity FTTx equipment) and shared OPEX options, to demonstrate 5G midhaul and backhaul traffic over PON. The PON solution will also provide to the network operators a management interface. When used for mobile networks, PONs already make use of a Dynamic Bandwidth Allocation algorithm to prioritize certain flows. The development of network slicing could be proposed by dynamic creation/edition/deletion of slice instances.

The project will demonstrate an on-the-fly reconfiguration of the PON DBA parameters, to be managed from an SDN controller in coordination between fixed and radio networks, used for mid and backhaul of 5G traffic with integrated RU/DU.

We also consider PTPv2 and SyncE protocols (transportation over the different transport domain mentioned above will be demonstrated. To support this demonstration, video services of various flavors (e.g., downstream HD/4K streams and/or conversational video) will been selected as representative vertical services that the Edge and Core DCs will also host the 5G NFs (e.g., the 5G Core VNFs), while resource sharing will be accomplished via MARSAL’s innovative MEO. This will allow disaggregating the AR, scene analysis, and activity recognition application functions in multiple tiers (i.e., Regional Edge, Radio Edge, on-device).

Figure 10: Mapping scenario with MARSAL architecture


5.2.2.5 TEST AND EVALUATION

The objective of this experimentation scenario is to demonstrate architecture based on O-RAN and to provide transportation of mobile traffic flows of disaggregated 5G/6G RAN deployments, over different optical transport technologies, including point-to-point, WDM and PON, while maintaining the latency and throughput constraints. The orchestration framework will handle the dynamic provisioning and configuration of the different virtualized resources, enabling network slicing over wireless and transport network resources as well as over edge compute resources.

In this experimentation scenario, the following innovations will be showcased and evaluated:

Demonstrate load balancing between fixed and mobile traffic through the coordination of the two SDN controllers, leveraging unused capacity from the PtMP link to serve traffic from the mobile clients of the cell-free RAN. Evaluate the distributed control plane’s capability to adapt to workload variations or protection mechanism in near-real time.

Demonstrate adaptive cooperation between multiple cell-free serial fronthaul based on WDM transport technology, leveraging on the midhaul links for information sharing. Showcase the effect of midhaul capacity limitations on the levels of cooperation and achieved Spectral Efficiency.

Evaluate the energy savings that can be achieved via traffic aggregation on a limited number of wavelengths and shutdown of unused SFP+ modules in light load conditions, under the control of the SDTN controller.

MARSAL will integrate the orchestration framework and supports procedures for the slicing and configuration of the transport network. In details, this experimental scenario will demonstrate and validate the following aspect. We will also use existing and next-generation PON technologies, to take advantage of their lower (CAPEX) (due to higher reuse of commodity FTTx equipment) and shared OPEX options, to demonstrate 5G midhaul and backhaul traffic over PON. The PON solution will also provide to management interface compatible with controller and orchestration. When used for mobile networks, PONs already make use of a Dynamic Bandwidth Allocation algorithm to prioritize certain flows. However, the development of network slicing requires a dynamic creation/edition/deletion of slice instances. The project will demonstrate an on-the-fly reconfiguration of the PON DBA parameters, to be managed from an SDN controller, used for mid and backhaul of 5GS traffic with integrated RU/DU. We will also focus our interest on protection mechanism to increase availability. The following tables present a set of preliminary requirements and KPIs for our scenario. These KPIs will be reconsidered during the project.

Table 5: Preliminary network requirements for the experimentation scenario 1.2.

KPI Campus use-case High-density area use-case

DL throughput 10 Gbps for fixed networks (xhaul)

1 Gbps for mobile user/device

10 Gbps for fixed networks (xhaul) 1 Gbps for mobile user/device

UL throughput 10 Gbps for fixed networks (xhaul)

100 Mbps for mobile user/device

10 Gbps for fixed networks (xhaul)

100 Mbps for mobile user/device

E2E delay < 2 ms < 2 ms

Density thousands per km2 ~hundreds per km2

Coverage 1 km2 0.5 km2


Mobility < 10 km/h N/A

Reliability 99.999 % 99.9 %

Location accuracy < 10 m < 0.5 m

Slice/service deployment < 30 minutes < 90 minutes

Table 6: Preliminary KPI definition for the experimentation scenario 1.2 (baseline scenario)



Name of the KPI KPI value of the baseline

scenario


TS2: Targeted KPI (with MARSAL

architecture) 5G backhaul Throughput

backhaul & midhaul

10GEth PtP XGS-PON or 25GEth 50G-PON (option)

5G fronthaul Throughput fronthaul

25Geth (eCPRI) Passive WDM (400Gbit/s capacity)

WDM (1Tbit/s capacity)

5G backhaul Reliability No protection mechnism (single PtP

fiber)

Protection mechanism with two fibers PtP (recovery latency <1s)

Protection mechanism with cooperation PON

and PtP

(recovery latency <300ms)

5G backhaul Manageable xhaul Static flow control

PONs propose a Dynamic Bandwidth Allocation (DBA)

algorithm to prioritize certain flows. However, the development

of network slicing requires fast, dynamic creation/ edition/ deletion of slice instances.

(traffic allocation dynamic <1s)

(traffic allocation dynamic <3ms)

5G backhaul Energy efficiency 1W/Geth 0.2W/GEth 0.1W/GEth (including

transport sleep modes)

5.3 PoC2: Cognitive assistance and its security and privacy implications in 5G and Beyond

Augmented Reality (AR) enhances the perception of the real world around us by introducing extra information over the users view, by combining and synchronizing computer generated data and real objects. To enhance the information provided to the users, AR devices can support haptic feedbacks, audio output as well as voice inputs. Cognitive assistance is taking AR on a step further by relying on real-time video, scene


analytics, and activity recognition to provide personalized feedback for activities the user might be performing.

Next-generation cognitive assistance devices are using multiple inputs like cameras, GPS, microphones and more to process and analyze the information to provide real time feedback to the user. The high computational load and massive data sets required for scene analysis and activity recognition makes on-device execution infeasible, requiring large amount of data to be transferred over the network [78]. Apart from the high bandwidth requirements from the network, zero perceived latency is also needed to ensure a satisfactory user experience, making real cognitive assistance not currently technically feasible in 5G networks. Or, to put it in ITU terminology, none of the three main 5G flavors (eMBB, URLLC, mMTC) can cope with such technical requirements since next-generation cognitive assistance requires not only ultra-high bandwidth (eMBB) but also low latency (URLLC) [79]. This mix of verticals is envisioned to result in 6G’s so-called mobile broadband and low latency (MBBLL) [80].

In more technical terms, the human eye typically requires low motion-to-photon (MTP) delay, where MTP is the time needed for a user movement to be fully reflected on a display screen [81]. Practically, MTP’s upper bound should not exceed 15 or 20 ms in simple settings. However, the response time of 5G is normally 25 ms in ideal functioning conditions. So, apart from higher throughputs to support ultra-high-definition video and other immersive features such as human gestures, next-generation cognitive assistance demands much lower response time for real-time voice-based commands (< 0.1 ms) and prompt control receptions (< 1 ms). In other words, for a completely immersive experience, 6G is needed to provide a latency of 0.1 ms and data rates of multi Gbps [82].

In this regard, AI is currently envisioned as the most prominent 6G enabling technology [79]. The reason lies in the increasingly complex and heterogeneous tasks of wireless networks, which are normally intractable through traditional approaches and point out the need for advanced ML techniques. Therefore, to support use cases like next-generation cognitive assistance, integrating AI and Machine Learning into mobile communication networks [83] is paramount to process massive amounts of data to obtain meaningful information [84]. Processing such massive amounts of data requires real-time access to powerful computational facilities in order to cope with demanding SLAs in a flexible and dynamic manner. MEC approaches like edge intelligence (EI) are envisioned as well to be key enablers for 6G systems (e.g., [85]). These technologies will endure MARSAL’s paradigm of Elastic Edge Computing, aiming to overcome the isolation and underutilization of resources deployed at Edge nodes, and offering zero perceived latency to smart connectivity applications.

The introduction of B5G/6G networks for cognitive assistance in a multi-tenant environment, without assuming trust, will raise security concerns that need to be addressed [78]. In fact, many security and privacy implications are inherent in applications that process personal data and Personally Identifiable Information (PII) as per the GDPR regulations, including video streams with users’ field of view and tracked location. Hence, privacy and security mechanisms that guarantee the isolation of slices and ensure collaboration of participants in multi-tenant B5G/6G infrastructures without assuming trust, need to be developed and demonstrated. Hereof, to ensure the end-to-end security in 6G, AI techniques will also play a critical role in protecting the network, user equipment, and vertical industries from unauthorized access and threats [79]. Besides, blockchain is envisioned as another key technology in 6G privacy/security given its decentralized operation, immutability, and enhanced security [86].

The effectiveness of the MARSAL solutions in B5G/6G Cognitive Assistance will be demonstrated through the implementation of the two experimentation scenarios of MARSAL PoC2. By bringing state-of-the-art


technologies in a novel way, MARSAL aims to address security and privacy related issues in B5G/6G networks, while optimizing user experience and enhancing confidentiality and trustworthiness.

5.3.1 EXPERIMENTATION SCENARIO 2.1: COGNITIVE ASSISTANCE AND SMART CONNECTIVITY FOR NEXT-GENERATION SIGHTSEEING


5G introduced real-time, interactive, Next-Generation Internet (NGI) applications that support human-centered interaction via novel interfaces (e.g., vision and haptics) including augmented reality applications. However, as it has been described in the previous section, current 5G networks do not have the capabilities to unleash the full potential of these applications due to, e.g., the required huge data throughputs and low latency for scene analysis and activity recognition. In this context, this experimentation scenario motivates the need for going beyond 5G through a high-level definition of two real-time, interactive, cloud-native applications for B5G/6G outdoors sightseeing. Through the use of augmented reality (AR) - a technology that superimposes a computer-generated image on a user's view of the real world, thus providing a composite view – these types of applications will support cognitive-assistance and human-centered interaction features.

Before moving forward to the description of the scenario (experimentation scenario 2.1), let us briefly define the key concepts involved from a user perspective. First, cognitive assistance is "a systematic approach to increasing human intellectual effectiveness"[74] that assumes "computational assists to human decision making are best when the human is thought of as a partner in solving problems and executing decision processes, where the strengths and benefits of machine and humans are treated as complementary co-systems” [75]. Second, human-centered computing is a computer system engineering methodology that uses a combination of methods from computer science, social science, and management studies to understand and model work practice, technology use, and technology gaps. Its objective is developing computer systems that fit human capabilities and practices by exploiting and improving AI programming methods [76].

To support these applications while respecting the correct performance of others, operators must provide a flexible network architecture capable of adapting to the demand. So, this experimentation scenario puts the focus on how to cope with the challenging requirements imposed by the operation of the proposed applications while keeping the SLAs of the rest of applications running in parallel. Load balancing according to priority services will be implemented at the same time MEC approaches will be conducted to lighten the network burden of the AR applications and leverage the use of edge computing nodes.

5.3.1.2 DESCRIPTION

In this scenario, the deployment of two real-time and interactive cloud-native applications for outdoors sightseeing supporting human-centered interaction via 3D cameras is envisioned in the MARSAL’s multi-tenant Elastic Edge Infrastructure. These applications would be offered to users equipped with untethered AR glasses. Both applications would endure an enhanced strolling experience by showing overlaid information relevant to their surroundings (APP#1) and enabling virtual artifacts manipulation (APP#2), while considering background traffic from other applications and services.

Augmented reality, or AR, is a digital technology, which makes changes to a person’s perception of their physical surroundings, when viewed through a particular device. The technology has similarities with virtual reality (VR), but AR does not replace the real-world environment, but augments it by overlaying digital components. To date, perhaps the most notable example of an AR app is Pokémon Go [77]. However, AR goes beyond the gaming world and tackles other scopes such as manufacturing or tourism. In the later, tourism providers or marketers will usually use AR to add graphics or useful information to an environment viewed


through a compatible device. A detailed description of B5G/6G use cases that feature AR can be found in [78].

AR may be realized in different ways depending on the user equipment. There are, nonetheless, two main types of AR end-devices: smartphones (or tablets) and smart glasses. While the former devices have a much more relevant penetration in the market, smart glasses are gaining attention given the complete immersive experience they provide. In essence, smart glasses are wearable computer-capable glasses that overlay additional (pre-processed) information in the form of images or animations to the scenes viewed by the end-user. This experimentation scenario envisaged within MARSAL leverages smart glasses to furnish a full B5G/6G sightseeing experience.

An enhanced strolling experience with overlaid information relevant to their surroundings and activities (e.g., restaurant ratings, nearest ATMs or bus-stations, touristic information, etc.) is proposed as application 1 (APP#1). This sightseeing application would apply real-time video analytics on the user’s field-of-view to detect user intent or activity and offer visual guidance in the form of relevant information that is optically super-imposed. Figure 11 shows a conceptual example of AR for outdoors sightseeing.

Figure 11: Example of outdoors sightseeing [87]

Besides, at certain attraction points throughout the strolling experience, IoT nodes equipped with novel interfaces (i.e., 3D cameras) and B5G/6G connectivity could be deployed to facilitate interaction with the user. In this regard, APP #2 is a cognitive assistance application that would encourage the user to manipulate in real-time the virtual representation of an artefact (also termed Digital Twin in B5G/6G terminology), projected at his/her AR glasses. Gesture recognition can be implemented via real-time analysis of the 3D


cameras2 streams, while the application may offer cognitive visual guidance, superimposing information at the users’ field of view explaining how the exhibit (or artefact) is used in real time. Figure 12 shows a conceptual example from Microsoft Hololens [88].

Figure 12: Microsoft Hololen [88]

To better showcase the scope of the proposed scenario, let us conceive a general use case consisting of a next-generation sightseeing tour enhanced with AR in the city of Barcelona. Throughout the tour, the user is provided with information of interest as she/he walks by the streets of the city (APP#1). The tour also proposes a predefined route where the user visits multiple points of interest (POI) empowered with artifact manipulation applications (APP #2). The whole concept is shown in Figure 13.

From the conception of smartphones on, tourists can search for almost any kind of information on the Internet while sightseeing. However, this experience can be highly improved by providing the user with overlayed information of potential interest in a proactive way, so the user can keep his/her focus just on the experience, getting rid of inconvenient searches among the copious amount of information online. In this regard, APP#1 would provide guidance and information concerning touristic places, nearest ATMs/bus-stations, restaurants, museums, hotels, and stores nearby the user by integrating location-based AR and image-recognition AR, which recognizes images and markers [89]. The intensity, ratings, comments, current social media data, and price information about these areas will be provided simultaneously on the user glasses [90]. Two examples showcasing the type of information provided by APP#1 are shown in Figure 13. At the right-most, information about the neighbourhood being visited by the user is displayed accessing to public domain services like Wikipedia. The second example, in the upper centre of the image, shows

2 A 3D camera is an imaging device that enables the perception of depth in images to replicate three dimensions as experienced through human binocular vision. Some 3D cameras use two or more lenses to record multiple points of view, while others use a single lens that shifts its position. The combination of the two perspectives, as with the slightly different perspective of two human eyes, is what enables depth perception. 3D photography can enable an immersive frozen-in-time moment for stills or video content that seems real enough to touch.


condensed information of a disco club near the user provided by Google Maps. This condensed information includes ratings, average price, and pictures of the place.

Figure 13: Concept of the preliminary use case (experimentation scenario 2.1)

Apart from overlaying information of potential interest to the user glasses, this scenario also considers artifact manipulation supported through gesture recognition via real-time analysis of the 3D camera stream to further enrich the sightseeing experience (e.g., [91] or [92]). In particular, APP#2 could be supported by multiple POIs (highlighted in red dots) with 3D cameras, where the user can interact in-situ with virtual artifacts through different puzzles or games. This way, gamification is put to the table to boost the user experience and potentially aiding the local commerce. A couple of examples are showcased in the conceptual diagram. At the left-most, a user visiting the Footbol Club Barcelona’s Camp Nou stadium is proposed a game in which she/he has to save a virtual shot by a virtual footballer. In order to save the shot, the user should move its arms toward the direction of the virtual football. The users that save the shot are given vouchers for the store or any other type of reward. Likewise, at the bottom centre of the picture, users visiting the Cristóbal Colón monument are suggested to indicate with their fingers the direction to America. In this case, the 3D cameras would capture the indications of the users and those pointing correctly to America may unlock different historical tips of interest.

The description of experimentation scenario 2.1 above would fit within the “Immersive smart city” use case under the category “Massive twinning” and also within the “Fully-merged cyber-physical worlds” use case under the category “Immersive telepresence for enhanced interactions” as described in [78]. In our vision, (i) the sightseeing application (APP#1) could be part of the different services that would be managed under the Immersive smart city, e.g., under the ambience/environment (for example, climate, air quality) and cultural aspects and (ii) the virtual artifact manipulation (APP#2) would be linked with the enhanced interactions to


be addressed by B5G/6G networks. As explained in [78], the use of B5G/6G networks are needed in these use cases as they improve some existing KPIs of 5G networks (such as service availability, coverage, and network energy efficiency), but also because B5G/6G networks bring new capabilities leading to new KPIs among which we highlight integrated sensing, local compute integration, integrated intelligence and flexibility (ability of the system to be adapted and tailored to specific use cases and environments as a consequence of disaggregation, softwarization, and automation/orchestration, which are concepts dealt with in MARSAL). The improvements in existing KPIs and the new KPIs that B5G/6G networks will bring are required in order to be able to guarantee the necessary quality of experience that MARSAL’s experimentation scenario 2.1.


Below, we list a generic description of the stakeholders involved in the MARSAL project:

End-user: The end-user is the consumer of the applications, participating as a spectator and active player, in APP#1 and APP#2, respectively.

User equipment manufacturer: These are the companies in charge of designing and manufacturing user equipment for VR/AR devices like smart glasses.

Network equipment manufacturer: Apart from traditional network devices (e.g., routers and switches), VNF-based devices must be designed and manufactured to support the MARSAL architecture, including DCs and MEC platforms.

Connectivity provider: Large enterprises (telecom operators) play this role by offering resources (network, computing, and storage) in large-scale, cloud-based environments for service deployment. It is their duty to safeguard the integrity and long-term viability of their infrastructures in order to accommodate virtualized services. NFV Infrastructures (NFVIs) are provided/leased by connectivity providers, where VIMs are in charge of the infrastructures, connected with the Service Virtualisation Platform (SVP) and controlled by the Service Platform Provider. The NFVIs address not only the requirements for a static, centralised cloud environment, but also the dynamic and mobility-related requirements.

Service provider: This is a generic role that can be assigned to many types of companies, covering a plethora of services in vertical domains that may potentially be benefited from the MARSAL architecture. Service Providers are companies that offer services to end-users or other companies, considered as the consumers of a service, for example, a company offering a MEC platform for processing of video files captured through the smart glasses or charging end-users.

Service Virtualisation Platform Operator: This stakeholder is in charge of running and maintaining a secure, scalable, and efficient SVP for the deployment of media services. The SVP Operator's responsibilities include general control of daily operations, provisioning and maintenance processes, security practices, disaster recovery planning and execution. The SVP operator is also responsible for analysing and optimizing resource allocation, as well as ensuring that charges are made in accordance with agreed-upon SLAs.

Applications developer: Companies that develop novel 6G-related applications (or services or functions that can be integrated to make a service) like APP#1 and APP#2. They may use the SVP's Service Development Kit (SDK) for deploying, configuring, and managing their services.

In this experimentation scenario, CTTC provides the 5G testbed, with a MEC platform deployed at a two-tier virtualized infrastructure with Regional and Radio Edge nodes and a 5G Core. CTTC will also offer the NFVO with the NSaaS subsystem, and ICCS will offer the Decision Engine with multi-objective optimization, and NEC


the Analytic engine with Context Representation. IQU will offer the cloud-native sightseeing applications, supporting real-time video analytics and 3D cameras for human interaction.


To support the abovementioned APPs, MARSAL Virtual Elastic Infrastructure will include a MEC platform be deployed both at the regional and radio edge Data Centers (DCs), and centralized orchestrators, i.e., the MEO and the NFVO, which will be deployed at a Core-tier Data Centre. Figure 14 depicts the mapping of the scenario into the MARSAL architecture. It must be noted that the Edge and Core DCs will also host the 5G NFs (e.g., the 5G Core VNFs), while resource sharing will be accomplished via MARSAL’s innovative MEO. This will allow disaggregating the AR, scene analysis, and activity recognition application functions in multiple tiers (i.e., Regional Edge, Radio Edge, on-device).

Figure 14: Overview of the experimentation scenario 2.1 mapped into the MARSAL architecture.

Two novel key MARSAL components are highlighted yellow in Figure 14. First, the Analytic Engine (AE) is the module in charge of creating the Context representation of the system. To this end it will interpret the data collected from the different MARSAL architecture components in the form of knowledge graphs and will create vectors representations that on the one hand avoid the sharing of private information and on the other hand make easy to downstream ML algorithms to understand the status of the system. From an information flow perspective, the Analytic engine will be placed between the data collection and the Decision Engine. It will generate an embedding of each node in the context knowledge graph using tools such as EP


[93] or GraphSage [94]. The embeddings will be used then by the Decision Engine to optimize the resource usage.

Second, MARSAL will design and implement Decision Engines (DEs) at the two Core-Tier orchestration subsystems (i.e., the NFVO, which manages end-to-end slices via the NSaaS module, and the Multi-access edge orchestrator) towards a Self-Driven Virtual Elastic Infrastructure. The Decision Engines will use as input the resulting embeddings of the Analytic Engine that represent the current state of the MARSAL infrastructure in a highly compressed form and feed them to different multi-objective downstream ML algorithms that will be implemented in the Decision Engines. Through these mechanisms the Decision Engines will jointly orchestrate Network Slices, Network Services and MEC applications continuously and automatically evaluating the current context under the required policy, while also delegating data-driven local control decisions to the lower tiers of the hierarchy.

5.3.1.5 TEST AND EVALUATION

As described above, this experimentation scenario will demonstrate MARSAL’s virtual elastic infrastructure, showcasing its ability to ensure high reliability and quality of experience for next-generation human-cantered applications with new terminal types (AR glasses), while sharing resources with high-priority 6G network functions (e.g., emergency calls). It will target two real-time and interactive Cloud-native applications for outdoors sightseeing: AR sightseeing application and Cognitive Assistance application. Both applications will rely on MARSAL’s Virtual Elastic Infrastructure to optimize and disaggregate their AR, scene analysis and activity recognition application functions in multiple tiers (i.e., Regional Edge, Radio Edge, on-device).

The main challenges are twofold:

1. Zero-perceived latency for satisfactory user experience, and for timely feedback in time-sensitive activities.

2. High computational load for scene analysis and activity recognition.

In particular, the following tests will be carried out:

Test scenario 1 (Placement of application functions): Demonstrate and evaluate the capabilities of the MEO to derive the optimal placement of the (containerized) application functions at the Radio Edge or Regional Edge DCs, achieving optimized distribution of latency budgets. The computational requirements and latency constraints of application functions will be derived from at the applications’ manifest files. This will result in imperceptible latency of the untethered AR applications, comparable to tethered AR, which will be validated by user tests.

Test scenario 2 (Inter-DC traffic steering): Demonstrate the collaborative interaction of the MEC system with the 5G UPF for real-time inter-DC traffic steering for load balancing purposes, evaluating the effect on resource utilization. Unbalanced demand will be emulated in the coverage area of certain Regional Edge nodes, and the ability of the MEC system to uniformly re-direct traffic will be showcased.

Test scenario 3 (Analytic and decision engines): Demonstrate the Analytic and Decision engines of MARSAL’s Self-Driven infrastructure and evaluate their ability to derive accurate context representations while successfully driving the NFVO and MEO. Evaluate their effectiveness in achieving a set of objectives related to SLA requirements (e.g., which of the two AR applications to prioritize) and cost considerations (e.g., related to OPEX, energy costs, etc.).

For the abovementioned test scenarios, the following baseline scenario will be considered:


Baseline scenario: Consider as a baseline the current SoTA of cloud-native technologies, where inter-DC placement of containerized application functions is not possible. Thus, time-sensitive functions can be placed in sub-optimal Regional Edge nodes, unnecessarily increasing latency, and non-time sensitive functions can be placed in Radio Edge nodes, unnecessarily wasting resources. Moreover, dynamic load balancing of traffic via the UPF will not be considered in the baseline scenario.

In the following tables, we list the preliminary set of KPIs from PoC 2.1 to be evaluated in each test scenario. In particular, we specify in the table columns i) the main characteristics concerning the baseline test scenario, ii) the KPI name, and iii) the target KPI value. For the baseline scenario, as it has been described above, we will compare the KPIs reached with MARSAL in each test scenario against its corresponding pre-6G system version featuring current SoTA cloud-native technologies. That is, we will rely on a centralized 5G cellular system as baseline by disabling the MARSAL features (e.g., inter-DC load balancing) from CTTC’s Experimental Platform (see Section 5.1.3). This way, by evaluating the performance in both cases (with and without the MARSAL features) we will be able to accurately corroborate the gains expected with the MARSAL system.

Besides, notice that OPEX reduction and Reduction of SLA violations appear in more than one test scenario given its generic nature. As for the later, SLAs include metrics that quantitatively characterize the communication service to be granted. According to the 5G architecture specification provided by the 5G PPP Initiative [95] the SLA requirements could be based on: i) end-to-end latency and bandwidth requirements, necessary for the service to function, ii) metrics related to the dimensioning of the service (number of users supported, area of coverage), iii) availability and reliability, and iv) optimization targets for service deployment, that could include deployment time and energy efficiency. In particular, MARSAL test scenarios of PoC 2.1 will focus on SLA based on end-to-end slice latency, energy efficiency, throughput, resource utilization, and critical-application reliability.


Test Scenario Baseline scenario Name of the KPI


TS 1

Centralized system Delay of time-critical functions < 1ms

Centralized system Scale-out of containerized functions < 1s

Centralized system Reduction of SLA Violations ≥ 20%


Test Scenario

Baseline scenario Name of the KPI


TS 2 No inter-DC load balancing Resource utilization 50% increase

No inter-DC load balancing OPEX reduction ≥ 20%

No inter-DC load balancing Reduction of SLA Violations ≥ 20%



Test Scenario

Baseline scenario Name of the KPI


TS 3

No multi-objective optimization / AI-driven

approaches

No. of variables in multi-objective optimization

≥ 4


approaches

No. of independent elements sharing data representations

≥ 5


approaches

OPEX reduction ≥ 20%

No distributed knowledge graph

Reduction of SLA Violations ≥ 20%

5.3.2 EXPERIMENTATION SCENARIO 2.2: DATA SECURITY AND PRIVACY IN MULTI-TENANT INFRASTRUCTURES


The goal of this scenario is to demonstrate and evaluate MARSAL’s privacy and security mechanisms. They guarantee the isolation of slices and ensure collaboration of participants in multi-tenant 6G infrastructures without assuming trust. These mechanisms will also be evaluated in terms of their ability to mitigate the increased privacy risks of NGI applications that process Personally Identifiable Information (PII).

To this end, this scenario will demonstrate the application of security and privacy mechanisms in four different layers, namely: secure and private sharing of information among tenants, legal security using smart contracts, security of the data stored in the cloud, and security of the final users.

The development of each of the different layers presents different challenges ranging from the implementation of smart contracts among different tenants to the real time analysis of network data to allow the protection of the final users in a timely manner. In this subsection, we keep the same outline as in the previous experimentation scenarios, but we investigate the testing and evaluation process separately for each domain.

5.3.2.2 DESCRIPTION

In MARSAL we approach the security and privacy in 6G networks in a holistic way. Contrary to the previous scenarios, offering a solution tailored for the MARSAL architecture, in this scenario we present a modular design to offer four different layers of security and privacy that could be applied in very different contexts. Moreover, we adapt them to work in the context of the Cognitive Assistance scenario in 5G and beyond described above within PoC2.

This scenario assumes a multi-tenant infrastructure with one MNO and two MVNOs, each serving an OTT application provider. MARSAL technology will ensure the isolation of the different slices while offering the possibility of collaboration among the different tenants. To this end, we will demonstrate how the usage of smart contracts can be paired with the private representation of data, allowing the sharing of information


among different tenants and the owner of the infrastructure that can be interested in the optimization of different ML models.

Moreover, the scenario will also cover the security (and privacy) at different levels of the MARSAL architecture. First, we will demonstrate how policies can be used to safely store data in the cloud (either at the core or the edge of the network), testing different allocation strategies that ensure the perpetual security of the data.

Then, we move our focus to the network, and we will demonstrate how the browsing patterns of users can be analysed in real time to alert final users against malicious behaviours they may have, before they get in trouble.

Following, we present the involved stakeholders and their roles, as well as the mapping of the scenario to the MARSAL architecture. Then, we provide a more detailed design of the different components involved in this demonstration scenario.


Below, we present the stakeholders who are involved in the security and privacy of the network:

Network operator: The network operator is in charge of the last 2 layers of security, the security of the final users and the security of the network itself. It will be in charge of deploying the engines in charge of the protection.

Cloud providers: The cloud providers should apply the storage policies defined by the users and tenants to ensure the security of the data.

Slice tenants: The slice tenants will be responsible for the correct execution of the privacy transformation of the data and the signature of the smart contracts.

Network Equipment Manufacturers: They should be able to generate equipment able to execute complex tasks such as Machine learning models execution or on-line rule adaptation to allow the protection of the final users.

Final users: Eventually, finally users are the most benefited from the execution of the different privacy and security layers that allows them to use the MARSAL architecture in a safer way.

In this experimentation scenario, CTTC offers the 5G testbed, with a DCS solution contributed by OTE, and a MLNX dataplane with distributed ML-based Threat Detection engines deployed at the SDN switch Network OS. UNIMI will offer the storage resource fragmentation solution, and policy-driven data anonymization and integrity protection via the NFS gateway. NEC will offer the privacy-preserving representation learning algorithms, and the Threat Analysis engine.


The different components of the demonstrator provide security and privacy to many parts of the MARSAL architecture, as depicted in Figure 15.

In the backend, we provide both, security to the data stored in the computing nodes and to the information shared among different tenants to improve the performance of the slicing infrastructure.

Then, the Network Infrastructure Security is directly applied to the network equipment by using the SDN paradigm. Finally, a thread detection engine able to protect the final users will also be demonstrated by collecting network data at the edge that can be then processes in the backend.


Figure 15: Mapping experimentation scenario 2.2 to MARSAL architecture

The main components of the demonstrator are:

Secure and Private Information Sharing among tenants and Blockchain-based Smart-contracts platform for network slicing: The first two components are in charge of allowing a secure information sharing among tenants. Both the algorithms for the secure and private information sharing and the smart-contract platform will be closely related and acting together with the Orchestrator in order to provide a better slicing with the collaboration of the different tenants.

Security and Privacy for the Data stored in the Cloud: The third component defines the policies to ensure the data is stored in a secure and private way. As such, this component is an intrinsic part of the cloud storage.

Users’ Network Security: Provide security to the final users from the network is a complex task that requires the collaboration of different parts of the network. In MARSAL, this security layer will involve the Data Centers in the regional edge and the network equipment (typically smart NICS) placed on the regional edge.

5.3.2.5 TEST AND EVALUATION OF SECURE AND PRIVATE INFORMATION SHARING AMONG TENANTS

The number of AI applications that are used in production real-world environments has rocketed in the past years, following the amazing advances obtained in different areas. These Machine Learning (ML) based applications range from the personalization of services or the improved healthcare offered to final users to the automatic management of networks by Telco operators in the new 5G architectures. However, these applications rely on input data coming from possibly heterogeneous sources (either human or other machines) and spread through platforms owned by different actors that may not be fully trusted. Clearly, this poses different privacy and confidentiality issues.


In MARSAL we aim to research, develop and evaluate data transformation methods to remove all private information from the data while allowing the application of downstream ML methods over the privatized data.

In this demonstration, we will first consider how different methods can be used to generate the data representation that is useful for Machine Learning even when it is impossible (or very difficult) to go back to the original data. To this end, different algorithms of general use such as PCA, Autoencoders or noise addition will be tested. Moreover, algorithms specifically designed for this purpose such as AutoGANs [96] will also be tested to understand if the complexity they add is justified for the improved performance.

Figure 16 describes the target scenario. MARSAL will develop the Transformation algorithm and will use a standard ML task to measure the accuracy obtained when using the Transformed data. Moreover, the privacy of the data will be measured by comparing the original data with the data obtained after the transformation.

Figure 16: Target scenario to measure the privacy vs. accuracy tradeoff

Finally, we will demonstrate the integration of the privacy sharing algorithms developed during the project in the MARSAL architecture. Contrary to the previous algorithms, designed to work over images or numeric data, we will demonstrate algorithms able to work over knowledge-graph data. The algorithms will be integrated as part of the Analytic Engine and demonstrated together with the experimentation scenario 2.1. In this Scenario 2.2 we will demonstrate how the different configurations affect the performance of the whole algorithms and how different tenants may apply different privatization strategies depending on the data they handle. In particular:

Test scenario 1 (Privacy vs. accuracy trade-off): In the first scenario we will evaluate of the privacy vs. Accuracy trade-off for different methods that transform data into vectors without including the original information. This evaluation will be used with standard datasets typically used by the AI community, such as the MNIST or the CelebA datasets.

Test scenario 2 (Assisted network slicing): Evaluation of the algorithms in the data generated by the MARSAL components in the context of the Cognitive Assistance scenario. To this end, the representations generated by the Analytic Engine will be combined with the privacy preserving algorithms. Moreover, in this scenario we will evaluate the performance of the data privatization algorithms.


Table 10: Preliminary KPI definition for the experimentation scenario 2.2 (secure and private information sharing among tenants)

KPI Security Layer Targeted KPI (with MARSAL architecture)

Number of data formats Private Data Sharing >2

Accuracy vs. Privacy trade-off Private Data Sharing >50% privacy with <15% accuracy penalty

Privatization Performance Private Data Sharing >100MB/s

5.3.2.6 TEST AND EVALUATION OF BLOCKCHAIN-BASED SMART-CONTRACTS PLATFORM FOR NETWORK SLICING

Multi-tenancy is foreseen as a pillar of 6G networks and will help recoup the sizable investments required by telecom operators through network slicing. In multi-tenant infrastructure, there is a distinction among Mobile Network Operators (MNOs), Infrastructure providers, and Mobile Virtual Network Operators (MVNOs). MNVOs and Over the Top (OTT) application providers generally lack spectrum licenses and rely on MNOs to lease Network Slices [97].

Toward this, MARSAL project will leverage a decentralised blockchain platform, a distributed ledger of transactions that can assure various security properties using consensus protocols and cryptography algorithms [98] that supports Network Slicing transactions via Smart contracts.

In this demonstrator, MARSAL’s blockchain platform will be implemented with a private, permissioned blockchain solution based on Hyperledger fabric [100]. The blockchain platform will become the base for the smart contract-based network slicing operation.

Following the creation of the blockchain platform, we will create smart contracts for the network slicing negotiation. The blockchain platform will offer smart contract creation, modification and execution. The available resources will auction with an English style auction system, with the highest bidder to take the assets. The auction will have a pre-set time limit, and upon the end of the auction, a smart contract will be created, including important information about the transaction. No personal identification information will be added to the smart contracts, rather than an encryption key, to limit the exposure of sensitive information of the users. The smart contract will be a binding contract among the tenants, alleviate the need for legal documents and time-consuming negations.

For the transaction to be validated and a new block to be added to the blockchain, an agreement needs to be reached among the tenants. Towards this, we will implement a voting consensus protocol as Practical Byzantine Fault Tolerance (PBFT) or RAFT, which require fewer computational resources. With voting consensus protocols, the decision is taken with the approval of the majority.

Finally, it will be integrated with a decentralized resource broker for keeping track of resource ownerships and Network Slice permissions. The latter is updated automatically upon Smart Contract execution and irrevocably committed to the blockchain.

Important parameters are the system latency and the number of blocks per unit time (throughput). These will depend on the number of users and nodes of the platform and the consensus protocol used. To evaluate our blockchain platform, we are planning three test scenarios:


Test scenario 1 (Evaluate blockchain functionality): In the first test scenario, we are going to evaluate the blockchain platform of the MARSAL project. Using platforms like Hyperledger Caliper [101] and Hyperledger Explorer [102], we are going to examine whether the platform works and performs as expected based on the supply inputs. The benchmark at this stage is if the blockchain platforms work. The performance metrics of throughput and latency will be measured at this stage.

Test scenario 2 (Smart contracts interactions): For the second test, we are going to test the functionality of our smart contracts and the interaction between them. We are going to examine whether smart contracts can be created and executed with the parameters provides. Furthermore, we are going to investigate if the inheritance and dependency injection mechanisms are functioning. i.e., Whether the needed information is given from one contract to another with the benchmark to be the creation of the new blocks.

Test scenario 3 (Cooperation of NVNOs): In the third and final testing scenario, we are going to examine different parameters towards the functionality of MARSAL’s Network slicing. The benchmark in this scenario is to demonstrate cooperation and trust among NVNO through the negotiation of the network resources with the MARSAL’s blockchain platform.

Table 11: Preliminary KPI definition for the experimentation scenario 2.2 (blockchain-based smart-contracts platform for network slicing)


Blockchain throughput Blockchain smart contracts >100MB/s

Blockchain latency Blockchain smart contracts 10ms

5.3.2.7 TEST AND EVALUATION OF SECURITY AND PRIVACY FOR THE DATA STORED IN THE CLOUD

The need of supporting the collaboration among different participants in multi-tenant scenarios can require the sharing and processing of data that can be under the control of different authorities and can be stored in different domains with different levels of trust. These data are often sensitive, and their improper access, analysis, and share may result in major privacy and security violations. The proper protection of data is therefore an enabling factor for modern 5G scenarios that make use of these data. In MARSAL, special attention will be then devoted to the design of solutions for regulating the (cross-domain) flow of data involved in collaborative analysis, also considering that such analysis over large data collections can be expensive.

MARSAL will design a policy-driven approach for controlling information sharing and flows among different service providers offering storage and/or computational resources and for ensuring their cooperation in the execution of computations. MARSAL will provide a model for expressing and enforcing data sharing constraints and will design an approach for computing an assignment of operations to nodes to execute collaborative computations compliant with such constraints. Data sharing constraints will specify what data can be accessed by a given party and the level of visibility (e.g., plaintext or encrypted). The proposed approach will allow each data authority to independently specify data sharing constraints on its own data, without the need for cooperation with others for policy specification. Economic aspects will be considered by partially delegating a computation to a computing node whenever this choice provides for an economic advantage (e.g., the computing node offers computational services at competitive prices) without violating the data sharing restrictions. Data will be also dynamically protected (e.g., through the application of on-the-fly encryption) when they will be transferred to an economically convenient node that is however not


authorized to access the data in their original format. MARSAL will also investigate the possibility of enriching the model through the consideration of trusted execution environments, which could be modelled as an additional service provider with specific characteristics. We will demonstrate the working of our solutions in the context of MARSAL architecture, considering computations that involve multiple distributed edge nodes. We will show how the different characteristics (and trust assumptions) of the edge nodes will affect the execution of collaborative computations. In particular:

Test scenario 1 (Economic evaluation): Evaluation of the economic benefits enabled by the designed solution. The goal is to assess whether the availability of different levels of visibility over the data can enable the assignment of operations to non-fully trusted (but potentially less expensive) service providers.

Test scenario 2 (Model applicability): Demonstrate the applicability of the proposed model through the design of a component that determines a minimum cost assignment for the operations of a computation, evaluating advantages (e.g., in terms of economic costs) of its application.

Table 12: Preliminary KPI definition for the experimentation scenario 2.2 (security and privacy for the data stored in the cloud)


Economic cost Security and privacy for data < 1 (normalized cost)

# Parties involved in computations

Security and privacy for data > 1 (involvement of providers)

5.3.2.8 TEST AND EVALUATION OF FINAL USERS’ NETWORK SECURITY

In the Cognitive Assistance Use Case, the traffic of several thousand users of different services is handled by the network that has access to the encrypted streams of traffic. Similar to the current situation, with users browsing the web, cognitive assistance devices will continuously access different sources of data while the user it is walking around the city. For example, it is expected that augmented reality glasses will access the website of a retailer shop to inform the user about the products that he could find inside the store.

In this scenario, where users do not control directly the hosts accessed by their devices, it is of key importance to detect and predict suspicious communications that could lead to the infection of the user device, and, in turn, the infection of the network. However, network traffic volumes have been growing at exponential rates, and new network technologies, such as 6G, are expected to further speed up such growth. Performing complex security analysis on the entire volume of data is usually not possible or anyway expensive. Moreover, the encryption of user traffic makes impossible for network operators to apply traditional Deep Packet Inspection.

In MARSAL, we will demonstrate a scalable solution to identify potentially malicious network entities, i.e., domain names and IP addresses, in large volumes of network traffic. This allows network administrator to focus more expensive analysis techniques only on the subset of network traffic that contains the identified potentially malicious entities.

MARSAL will apply apply recently developed Natural Language Processing [99] algorithms in order to model the communication of a network host and to assess the security sensitiveness of such host. Our solution will focus on an aspect of the network communication that does not change depending on the protocol and is


only partially affected by encryption. Indeed, our approach is to look at the sequence of contacted network destinations (e.g., domain names, IP addresses).

Figure 17: Example of malware detection from the sequence of hosts contacted

MARSAL will demonstrate how it is possible to detect and block dangerous connections (I.e., the download of malware) before they happen. To this end, it will analyse the traffic in real time using the logs of network equipment and acting over the firewall rules (or other equipment) to block the malicious traffic. In particular:

Test scenario 1 (Cyber threat detection): In this scenario we will evaluate the number of threats detected by our system before they happen. Providing the possibility to block the connection. Moreover, we will also evaluate the number of false positives (normal connections classified as malicious).

Table 13: Preliminary KPI definition for the experimentation scenario 2.2 (final users’ network security)


Users’ Security Recall Users’ Network Security Discover 20% more threats over Static Black list

Users’ security Accuracy Users’ Network Security >1% False positives

Users’ security performance Users’ Network Security >1Gbps traffic analysed.


6 Conclusions The evolutions of today’s 5G mobile networks will be soon available to handle all types of applications and to provide service to massive numbers of users. In this complex and dynamic network ecosystem, an end-to-end performance analysis and optimization will be key features, in order to effectively manage the diverse requirements imposed by multiple vertical industries over the same shared infrastructure.

To enable such a vision, MARSAL targets the development and evaluation of a complete framework for the management and orchestration of network resources in 5G and beyond, by utilizing a converged optical-wireless network infrastructure in the access and fronthaul/midhaul segments. At the network design domain, MARSAL targets the development of novel cell-free based solutions that allows the significant scaling up of the wireless APs in a cost-effective manner by exploiting the application of the distributed cell-free concept and of the serial fronthaul approach, while contributing innovative functionalities to the O-RAN project. In parallel, in the fronthaul/midhaul segments MARSAL aims to radically increase the flexibility of optical access architectures for Beyond-5G Cell Site connectivity via different levels of fixed-mobile convergence. At the network and service management domain, the design philosophy of MARSAL is to provide a comprehensive framework for the management of the entire set of communication and computational network resources by exploiting novel ML-based algorithms of both edge and midhaul DCs, by incorporating the Virtual Elastic Data Center/Infrastructure paradigm. Finally, at the network security domain, MARSAL aims to introduce mechanisms that provide privacy and security to application workload and data, targeting to allow applications and users to maintain control over their data when relying on the deployed shared infrastructures, while AI and Blockchain technologies will be developed in order to guarantee a secured multi-tenant slicing environment.

As it has been extensively described in the previous sections, in MARSAL, we focus on a wide range of experimentation scenarios. The first domain includes a set of use cases focused on cell-free networking in dense and ultra-dense hotspot areas. The second domain includes use cases related to cognitive assistance, as well as security and privacy implications in 5G and beyond. In particular:

PoC 1: Cell-Free Networking in Dense and Ultra-Dense Hotspot Areas o Experimentation Scenario 1.1: Dense User-Generated Content Distribution with mmWave

Fronthauling. o Experimentation Scenario 1.2: Ultra-Dense Video Traffic Delivery in a Converged Fixed-Mobile

Network. PoC 2: Cognitive Assistance and its Security and Privacy Implications in 5G and Beyond

o Experimentation Scenario 2.1: Cognitive Assistance and Smart Connectivity for Next-Generation Sightseeing.

o Experimentation Scenario 2.2: Data security and privacy in multi-tenant infrastructures.

In this deliverable, we presented the first outcomes from Task 2.1, which were focused on the definition of PoC requirements, the specifications and the corresponding KPIs. We defined the PoC requirements and objectives, including their components, applications and business KPIs, addressing their target values. The PoCs in all facilities were investigated, determining the technical parameters that may affect their implementation and refining them to the level of detail required for conducting the project effectively. The detailed PoC definitions and requirements will act as blueprints for their implementation in WP6.


References

[1] Cisco Visual Networking Index: Forecast and Trends, 2017–2022 White Paper. [2] Ericsson Mobility Reports, Available at: https://www.ericsson.com/en/mobility-

report/reports/november-2019. [3] United Nations, Revision of World Urbanization Prospects. Available at:

https://www.un.org/development/desa/publications/2018-revision-of-world-urbanization-prospects.html.

[4] Megacities - Statistics & Facts, Available at: https://www.statista.com/topics/4841/megacities/. [5] 3GPP, “Telecommunication management; Study on management and orchestration of network slicing

for next generation network”, 3rd Generation Partnership Project (3GPP), Technical Specification (TS) 28.801, September 2017.

[6] 5G PPP Software Network Working Group, Cloud Native and 5G Verticals’ services, [Online], Available: https://5g-ppp.eu/wp-content/uploads/2020/02/5G-PPP-SN-WG-5G-and-Cloud-Native.pdf, February 2020.

[7] STARLINGX Platform website. Available at: https://www.starlingx.io/. [8] ETSI NFV-IFA 029. Available at: https://www.etsi.org/deliver/etsi_gr/NFV-

IFA/001_099/029/03.03.01_60/gr_NFV-IFA029v030301p.pdf. [9] J. Liu, M. Sheng, L. Liu and J. Li, “Network Densification in 5G: From the Short-Range Communications

Perspective,” in IEEE Communications Magazine, vol. 55, no. 12, pp. 96-102, December 2017. [10] E. Björnson and L. Sanguinetti, “Cell-Free versus Cellular Massive MIMO: What Processing is Needed

for Cell-Free to Win?,” 2019 IEEE 20th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), pp. 1-5, 2019.

[11] N. Rajatheva, et al., “Broadband Connectivity in 6G”, White Paper, arXiv: Signal Processing, 2020. Available at: https://arxiv.org/pdf/2004.14247.pdf.

[12] H. Q. Ngo, A. Ashikhmin, H. Yang, E. G. Larsson, and T. L. Marzetta, “Cell-Free Massive MIMO Versus Small Cells,” IEEE Transactions on Wireless Communications, vol. 16, no. 3, pp. 1834–1850, 2017.

[13] 3GPP, “NG-RAN; Architecture description,” 3rd Generation Partnership Project (3GPP), Technical Specification (TS) 38.401, January 2020, version 16.0.0.

[14] “O-RAN Architecture Description,” Technical Specification (TS) O-RAN.WG1.O-RAN-Architecture-Description-v04.00, March 2021.

[15] G. Interdonato, P. Frenger, and E.G. Larsson, “Scalability Aspects of Cell-Free Massive MIMO,” in ICC 2019 IEEE International Conference on Communications (ICC), pp. 1–6, 2019.

[16] S. Buzzi, C. D’Andrea, A. Zappone and C. D’Elia, "User-Centric 5G Cellular Networks: Resource Allocation and Comparison With the Cell-Free Massive MIMO Approach," in IEEE Transactions on Wireless Communications, vol. 19, no. 2, pp. 1250-1264, February, 2020.

[17] D. Wang, C. Zhang, Y. Du, J. Zhao, M. Jiang, and X. You, “Implementation of a Cloud-Based Cell-Free Distributed Massive MIMO System,” IEEE Communications Magazine, vol. 58, no. 8, pp. 61–67, 2020.

[18] S. Blandino, C. Desset, G. Mangraviti, A. Bourdoux, and S. Pollin, “Phase-Noise Mitigation at 60 GHz with a Novel Hybrid MIMO Architecture”. In Proceedings of the 2nd ACM Workshop on Millimeter-Wave Networks and Sensing Systems (mmNets'18), 2018.

[19] S. Verdú, Multiuser Detection. Cambridge University Press, 1998. [20] A. García-Durán, M. Niepert, “Learning graph representations with embedding propagation”. In

Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS'17). Curran Associates Inc., Red Hook, NY, USA, 5125–5136, 2017.


[21] K. Deb, “Multi-objective Optimization”. In: Burke E., Kendall G. (eds) Search Methodologies. Springer, Boston, MA, 2014.

[22] P. Hernandez-Leal, B. Kartal, M.E. Taylor, “A survey and critique of multiagent deep reinforcement learning”. Auton Agent Multi-Agent Syst 33, 750–797, 2019.

[23] Q.V. Pham, F. Fang, V. Nguyen, Md. J. Piran, M. Le, L. B. Le, W.J. Hwang, Z. Ding, “Survey of Multi-Access Edge Computing in 5G and Beyond: Fundamentals, Technology Integration, and State-of-the-Art,” in IEEE Access, vol. 8, pp. 116974 – 117017, June 2020.

[24] S. Vimercati, S. Foresti, G. Livraga, P. Samarati, “Data Security and Privacy in Smart Cities,” in Smart Cities Policies and Financing Handbook, Morgan Kaufmann, 2021.

[25] G. Salvaneschi, M. Köhler, D. Sokolowski, P. Haller, S. Erdweg, M. Mezini, “Language-Integrated Privacy- aware Distributed Queries”, in Proc. of ACM OOPSLA, 2019.

[26] S. Vimercati, S. Foresti, S. Jajodia, G. Livraga, S. Paraboschi, P. Samarati, “An Authorization Model for Multi-Provider Queries,” in Proc. of the VLDB Endowment, vol. 11, n. 3, November 2017.

[27] S. Vimercati, S. Foresti, S. Jajodia, S. Paraboschi, P. Samarati, “Efficient Integrity Checks for Join Queries in the Cloud, “ JCS, vol. 24, no. 3, 2016, pp. 347–378.

[28] B. Zhang, B. Dong, W. H. Wang, “CorrectMR: Authentication of Distributed SQL Execution on MapReduce,” IEEE TKDE, vol. 33, no. 3, pp. 897 – 908, March 2021.

[29] T. Kohlberger, Y. Liu, M. Moran, P. H. C. Chen, T. Brown, J. Hipp, M. C. Stumpe, “Whole-slide image focus quality: Automatic assessment and impact on AI cancer detection.” Journal of pathology informatics, 2019.

[30] J. A. Ayala-Romero, A. Garcia-Saavedra, M., Gramaglia, X., Costa-Perez, A., J. J. Alcaraz, “vrAIn: A deep learning approach tailoring computing and radio resources in virtualized RANs”, in 25th Annual International Conference on Mobile Computing and Networking, pp. 1-16, October 2019.

[31] A. Garcia Duran, M. Niepert, “Learning graph representations with embedding propagation. Advances in neural information processing systems”, 30, pp. 5119-5130, 2017.

[32] J. Liu, Z. Liu, “A survey on security verification of blockchain smart contracts”. IEEE Access, 7, 77894-77904, 2019.

[33] G. Siracusano, M. Trevisan, R. Gonzalez, R. Bifulco, (2019, November). Poster: on the application of NLP to discover relationships between malicious network entities. In Proceedings of the 2019 ACM SIGSAC Conference on Computer and Communications Security (pp. 2641-2643).

[34] 3GPP, “NR; Multi-connectivity; Overall description; Stage-2”, 3rd Generation Partnership Project (3GPP), Technical Specification (TS) 37.340, January 2018.

[35] 3GPP TS 38.401 V16.6.0, 3rd Generation Partnership Project, Technical Specification Group Radio Access Network, NG-RAN, Architecture description (Release 16), July 2021.

[36] 3GPP TS 23.502 V17.1.0, 3rd Generation Partnership Project, Technical Specification Group Services and System Aspects, Procedures for the 5G System (5GS), Stage 2 (Release 17), June 2021.

[37] O-RAN.WG1.Use-Cases-Detailed-Specification, Technical specification, v03.00, July 2020. [38] 3GPPP Specification 22.891: Study on New Services and Markets Technology Enablers – Technical

Report (TR) - Release 14. Available at: https://portal.3gpp.org/desktopmodules/Specifications/SpecificationDetails.aspx?specificationId=2897.

[39] International Telecommunication Union – Radiocommunication Sector (ITU-R): Recommendation M.2083-0 (09/2015): IMT Vision – Framework and overall objectives of the future development of IMT for 2020 and beyond. ITU, Geneva, Switzerland.

[40] 5G ERA Website, https://www.5g-era.eu/USE-CASES/ [Online], Accessed July 2021.


[41] 5GMedia-Hub Website, https://www.5gmediahub.eu/USE-CASES/ [Online], Accessed July 2021. [42] VITAL-5G Website, https://www.vital5g.eu/TARGET-USE-CASES/ [Online], Accessed July 2021. [43] 5G-PHOS Website, http://www.5g-phos.eu/ [Online], Accessed July 2021. [44] 5G-ROUTES Website, https://www.5g-routes.eu/ABOUT/CAM-TRIALS/ [Online], Accessed July 2021. [45] MonB5G Website, https://www.monb5g.eu/ [Online], Accessed July 2021. [46] Techno-economic analysis of the beyond 5G environment, use case requirements and KPIs [Online],

Available: https://www.monb5g.eu/wp-content/uploads/2020/11/Mon-B-5G_D2-2_Final-NEW.pdf, Accessed July 2021

[47] 5G-SOLUTIONS website, https://5gsolutionsproject.eu/ [Online], Accessed July 2021. [48] Definition and analysis of use cases/scenarios and corresponding KPIs based on LLs [Online],

Available: https://5gsolutionsproject.eu/wp-content/uploads/2021/06/D1.1A-Resubmission-Definition-and-analysis-of-use-cases-scenarios-and-corresponding-KPIs-based-on-LLs_FNET_v1.2.4.pdf, Accessed July 2021.

[49] 5G-EPICENTRE Website, https://www.5gepicentre.eu/PILOT-EXPERIMENTS/ [Online], Access July 2021.

[50] 6G BRAINS Website, https://6g-brains.eu/ [Online], Access July 2021. [51] AI@EDGE Website, https://aiatedge.eu/?page_id=82 [Online], Access July 2021. [52] HEXA-X9 Website, https://hexa-x.eu/ [Online], Access July 2021. [53] 5G-TOURS Website, http://5gtours.eu/ [Online], Access July 2021. [54] 5G-TOURS, Deliverable D2.1 Use cases, deployment and techno-economic requirements - high level

description, http://5gtours.eu/documents/deliverables/D2.1.pdf [Online], Access July 2021. [55] 5G-MEDIA Website, https://www.5gmedia.eu/ [Online], Access July 2021 [56] 5G-MEDIA, D2.2: 5G-MEDIA Requirements and Use Case Refinement,

http://www.5gmedia.eu/cms/wp-content/uploads/2018/07/5G-MEDIA-D2.2-v1.0.pdf [Online], Access July 2021.

[57] 5G-COMPLETE Website, https://5gcomplete.eu/, [Online], Access July 2021. [58] NGMN Website, https://www.ngmn.org/ [Online], Access July 2021. [59] 5G White Paper 2. Available at: https://www.ngmn.org/wp-content/uploads/NGMN-5G-White-

Paper-2.pdf. [60] Alliance for Telecommunications Industry Solutions (ATIS) Website, https://www.atis.org/about

[Online], Access July 2021. [61] ATIS 5G Vertical Platform Assessment Report, https://www.atis.org/resources/5g-vertical-platform-

assessment-report/ [Online], Access July 2021. [62] 5GACIA Website, https://5g-acia.org/ [Online], Access July 2021. [63] 5GACIA White paper "Key 5G Use Cases and Requirements from the Viewpoint of Operational

Technology Providers". [64] NEM White paper "5G Media slice definition" (version 1.2), September 2019. [65] “Ericsson mobility report. On the pulse of the networked society”, Jun. 2013, Fig. 21.

http://www.ericsson.com/res/docs/2013/ericsson-mobility-report-june-2013.pdf. [66] Zhang, Peng, and Frans M. J. Willems. “On the Downlink Capacity of Cell-Free Massive MIMO with

Constrained Fronthaul Capacity.” Entropy (Basel, Switzerland) vol. 22,4:418, April 2020, doi:10.3390/e22040418.

[67] Z. H. Shaik, E. Björnson and E. G. Larsson, "Cell-Free Massive MIMO with Radio Stripes and Sequential Uplink Processing," 2020 IEEE International Conference on Communications Workshops (ICC Workshops), 2020.


[68] M. M. Mowla, I. Ahmad, D. Habibi and Q. V. Phung, “A Green Communication Model for 5G Systems,” in IEEE Transactions on Green Communications and Networking, vol. 1, no. 3, pp. 264-280, September 2017, doi: 10.1109/TGCN.2017.2700855.

[69] 3GPP, “NG-RAN; Architecture description,” 3rd Generation Partnership Project (3GPP), Technical Specification (TS) 38.401, January 2020, version 16.0.0.

[70] C.F. Müller, G. Galaviz, Á.G. Andrade, I. Kaiser, W. Fengler “Evaluation of Scheduling Algorithms for 5G Mobile Systems”. In M. Sanchez, L. Aguilar, M. Castañón-Puga, A. Rodríguez-Díaz (eds) Computer Science and Engineering—Theory and Applications. Studies in Systems, Decision and Control, vol. 143. Springer, Cham, 2018.

[71] M. Minelli, M. Ma, M. Coupechoux and P. Godlewski, "Scheduling Impact on the Performance of Relay-Enhanced LTE-A Networks," in IEEE Transactions on Vehicular Technology, vol. 65, no. 4, pp. 2496-2508, April 2016.

[72] I. Angri, A. Najid and M. Mahfoudi, “Performance Evaluation of Newly Implemented Resource Blocks (RB) Allocation Schemes on NS-3 simulator for mMTC 5G NR (New Radio) Femtocells,” 2020 5th International Conference on Cloud Computing and Artificial Intelligence: Technologies and Applications (CloudTech), pp. 1-8, 2020.

[73] Bonati and S. D'Oro and M. Polese, S. Basagni and T. Melodia, “Intelligence and Learning in O-RAN for Data-driven NextG Cellular Networks,” arXiv:2012.01263v2, July 2021.

[74] D. C. Engelbart, “Augmenting human intellect: A conceptual framework.” Menlo Park, CA, 1962. [75] National Research Council. Complex operational decision making in networked systems of humans

and machines: a multidisciplinary approach. National Academies Press, 2014. [76] NASA Website, Available at: https://www.nasa.gov/centers/ames/research/technology-

onepagers/hc-computing.html. [77] P. Rauschnabel, A. Rossmann, M. C. Dieck, “An adoption framework for mobile augmented reality

games: The case of Pokémon Go”, Computers in Human Behavior 76, 276-286, 2017. [78] Hexa-X D1.2 v2 “Expanded 6G vision, use cases and societal values – including aspects of

sustainability, security and spectrum”, April 2021. Available online: https://hexa-x.eu/wp-content/uploads/2021/05/Hexa-X_D1.2.pdf.

[79] W. Jiang, B. Han, M. A. Habibi, H. D. Schotten, “The road towards 6G: A comprehensive survey”. IEEE Open Journal of the Communications Society, 2, 334-366, 2021.

[80] C. De Alwis, A. Kalla, Q. V. Pham, P. Kumar, K. Dev, W. J. Hwang, M. Liyanage, “Survey on 6G frontiers: Trends, applications, requirements, technologies and future research”. IEEE Open Journal of the Communications Society, 2, 836-886, 2021.

[81] M. S. Elbamby, C. Perfecto, M. Bennis, K. Doppler, “Toward low-latency and ultra-reliable virtual reality”. IEEE Network, 32(2), 78-84, 2018.

[82] M. Giordani, M. Polese, M. Mezzavilla, S. Rangan, M. Zorzi, “Toward 6G networks: Use cases and technologies”. IEEE Communications Magazine, 58(3), 55-61, 2020.

[83] Q. Qi, X. Chen, C. Zhong, Z. Zhang, “Integration of energy, computation and communication in 6G cellular internet of things”. IEEE Communications Letters, 24(6), 1333-1337, 2020.

[84] Z. Na, Y. Liu, J. Shi, C. Liu, Z. Gao, “UAV-supported clustered NOMA for 6G-enabled Internet of Things: Trajectory planning and resource allocation”. IEEE Internet of Things Journal, 2020.

[85] A. Alshahrani, I. A. Elgendy, A. Muthanna, A. M. Alghamdi, A. Alshamrani, “Efficient multi-player computation offloading for VR edge-cloud computing systems”. Applied Sciences, 10(16), 5515, 2020.


[86] R. Sekaran, R. Patan, A. Raveendran, F. Al-Turjman, M. Ramachandran, L. Mostarda. Survival Study on Blockchain Based 6G-Enabled Mobile Edge Computation for IoT Automation. Specia session on Human-Driven Edge Computing (HEC), August 2020.

[87] Zumoko Website, http://zumoko.com [Online], Access July 2021. [88] Microsoft Hololens Website, https://whatis.techtarget.com/definition/3D-camera [Online], Access

July 2021. [89] R. Sasaki, K. Yamamoto, “A Sightseeing Support System Using Augmented Reality and Pictograms

within Urban Tourist Areas in Japan,” ISPRS International Journal of Geo-Information 8.9: 381, 2019. [90] Ö. F. Demir, E. Karaarslan, “Augmented reality application for smart tourism: GökovAR”, 6th

International Istanbul Smart Grids and Cities Congress and Fair (ICSG), IEEE, 2018. [91] R. Iacoviello, D. Zappia, “HoloCities: A Shared Reality application for Collaborative Tourism”, IOP

Conference Series: Materials Science and Engineering. Vol. 949. No. 1. IOP Publishing, 2020. [92] 5G-TOURS, D4.1 Robotic, Media and Smart Cities solutions for Touristic Cities, 2019. [93] A. Garcia Duran, M. Niepert, “Learning graph representations with embedding propagation”,

Advances in neural information processing systems, 30, 5119-5130, 2017. [94] W.L. Hamilton, R. Ying, J. Leskovec, “Inductive Representation Learning on Large Graphs”.

arXiv:1706.02216 [cs.SI], 2017. [95] 5G-PPP, “Living document on 5G PPP use cases and performance evaluation models”. Available

online: https://5g-ppp.eu/wp-content/uploads/2014/02/5G-PPP-use-cases-and-performance- evaluation-modeling_v1.0.pdf.

[96] H. Nguyen, D. Zhuang, P. Y. Wu, M. Chang. “Autogan-based dimension reduction for privacy preservation”, Neurocomputing, 384, pp. 94-103, 2020.

[97] B. Nour, A. Ksentini, N. Herbaut, P. A. Frangoudis and H. Moungla, “A Blockchain-Based Network Slice Broker for 5G Services,” in IEEE Networking Letters, vol. 1, no. 3, pp. 99-102, September 2019.

[98] P. Álvares, L. Silva, N. Magaia, “Blockchain-Based Solutions for UAV-Assisted Connected Vehicle Networks in Smart Cities: A Review, Open Issues, and Future Perspectives”, Telecom, 2, 108-140, 2021.

[99] G. Siracusano, M. Trevisan, R. Gonzalez, and R. Bifulco, “On the application of NLP to discover relationships between malicious network entities”. In Proceedings of the 2019 ACM SIGSAC Conference on Computer and Communications Security, pp. 2641-2643, November 2019.

[100] Hyperledger Fabric Website, https://www.hyperledger.org/use/fabric [Online], Access July 2021.

[101] Hyperledger Caliper Website, https://www.hyperledger.org/use/caliper [Online], Access July 2021.

[102] Hyperledger Explorer Website, Available at: https://www.investopedia.com/terms/h/hyperledger-explorer.asp [Online], Access July 2021.