FREE I FAST I FOR ALL

Volume 1 (2020), Issue 1

Volume 1 (2020), Issue 1

Inaugural issue
 FREE I FAST I FOR ALL

The ITU Journal on Future and Evolving Technologies (ITU J-FET) is an international
journal providing complete coverage of all communications and networking
paradigms, free of charge for both readers and authors.

The ITU Journal considers yet-to-be-published papers addressing fundamental and

applied research. It shares new techniques and concepts, analyses and tutorials, and
learnings from experiments and physical and simulated testbeds. It also discusses
the implications of the latest research results for policy and regulation, legal
frameworks, and the economy and society. This publication builds bridges between
disciplines, connects theory with application, and stimulates international dialogue.
Its interdisciplinary approach reflects ITU’s comprehensive field of interest and
explores the convergence of ICT with other disciplines.

The ITU Journal welcomes submissions at any time, on any topic within its scope.

Publication rights
©International Telecommunication Union, 2020

Some rights reserved. This work is available under the CC BY-NC-ND 3.0 IGO license:
https://creativecommons.org/licenses/by-nc-nd/3.0/igo/.

SUGGESTED CITATION: ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1.

COMMERCIAL USE: Requests for commercial use and licensing should be addressed to ITU Sales
at [email protected]

THIRD-PARTY MATERIALS: If the user wishes to reuse material from the published articles that is attributed to
a third party, such as tables, figures or images, it is the user’s responsibility to determine whether permission
is needed for that reuse and to obtain permission from the copyright holder. The risk of claims resulting from
infringement of any third-party-owned component in the work rests solely with the user.

GENERAL DISCLAIMERS: The designations employed and the presentation of the material in the published
articles do not imply the expression of any opinion whatsoever on the part of ITU concerning the legal status of
any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries.
The mention of specific companies or of certain manufacturers’ products does not imply that they are endorsed
or recommended by ITU in preference to others of a similar nature that are not mentioned.

ADDITIONAL INFORMATION
Please visit the ITU Journal website at:
https://www.itu.int/en/journal/j-fet/Pages/default.aspx

Inquiries should be addressed to

Alessia Magliarditi at: [email protected]
 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Foreword
The new ITU Journal on Future and Evolving
Technologies will deliver unique value to academia
and the broader ITU community.
It seeks global representation in its published papers
and teams of editors and reviewers, and it welcomes
an interdisciplinary approach to the broad scope of
topics addressed by ITU.
The ITU Journal embodies the inclusive character of
ITU and it is this inclusivity that will be the defining
factor in the Journal’s path to global impact.
ITU is the United Nations specialized agency for
information and communication technologies
(ICTs). Our global membership includes 193
Member States and over 900 companies,
universities, and international and regional organizations.
In contributing to the Journal and the work of ITU, leading minds in science and engineering are
providing decision-makers in the public and private sector with unique insight into the latest
developments in ICT and prospects for future innovation.
ITU works in service of the public interest, aiming to ensure that all the world’s people share in the
benefits of the ICT advances changing our world. This mission is very much aligned with the mission
of the global academic community. Therefore, our Journal is free of charge for both readers and authors.
I express my gratitude to all contributors to this first issue of the ITU Journal and I would especially
like to thank our Editor-in-Chief, Professor Ian F. Akyildiz, for the great dedication and conviction that
define his leadership.
I am very much looking forward to our work together to build a journal focused on ensuring that
breakthroughs in research impact our lives for the better, on a global scale.

Houlin Zhao
Secretary-General
International Telecommunication Union

– iii –
ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Foreword
The ITU Journal on Future and Evolving Technologies
is the latest initiative to contribute to the growing
strength of ITU’s relationship with academia.
Researchers participate alongside policymakers and
industry-leading engineers in ITU expert groups
responsible for radiocommunication, standardization
and development. Contributions from research
communities bring greater strength to the work of ITU,
and participation in ITU helps these communities to
increase the impact of their research.
Researchers play a key part in ITU’s Primetime Emmy
winning standardization work for video coding. They
are also an important driving force in ITU
standardization work in fields such as network
orchestration, AI and machine learning, video gaming,
blockchain, digital finance, digital health, autonomous
driving, and quantum information technology.
We are entering new frontiers in information and communication technology (ICT) and supporting ITU
standardization work. Research communities are integral to this movement and this movement is certain
to grow stronger with the support of the ITU Journal.
I would like to thank all contributors to this first issue of the ITU Journal as well as our Editor-in-Chief,
Ian F. Akyildiz, for creating a first issue that addresses topics of great strategic importance to ITU.
I am very glad to welcome Professor Akyildiz to the ITU community. We share a vision focused on the
future and a firm belief that the ITU Journal will deliver unique value to readers and authors worldwide.
Isaac Newton famously said that if he had seen further, it was by standing on the shoulders of giants.
Academia is helping the ITU community to see further. I thank you with all my heart for your support.

Chaesub Lee
Director
ITU Telecommunication Standardization Bureau

– iv –
 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Editor-in-Chief’s Message
On behalf of the International Telecommunication Union (ITU),
I welcome you to the inaugural issue of the ITU Journal on Future
and Evolving Technologies (ITU J-FET), an international, archival
and open access journal providing a publication vehicle for complete
coverage of all topics of interest to those involved in all aspects of
communications and networking from academia, industry and
governments, highlighting standardization perspectives.

The ITU Journal will share new techniques and concepts, analyses
and tutorials, and learnings from experiments and physical and
simulated testbeds and discuss the implications of the latest research
results for policy and regulation, legal frameworks, and the economy
and society. As a globally representative publication with a broad
scope of interest and prospective impact, the Journal is well
positioned to deliver unique value to academia and industry and
presents a unique opportunity to share their work with an
international audience invested in the broad scope of issues under the remit of ITU.

Free, fast, for all, the Journal aims to promote accessibility of research to academics and industry
researchers across the world. The publication is free of charge for both readers and authors, highlighting
the true sense of the term, ‘open access’.

ITU J-FET will be committed to the timely publication of very high quality, peer-reviewed, original
papers that advance the state of the art and applications of communications and networking paradigms.
Survey and roadmap papers reviewing the state of the art of relevant topics will also be considered.

A strong editorial board is the key to success. We have created an Editorial Board with truly outstanding,
experienced and distinguished people who are in the forefront of the telecommunications research
world. We are determined to give detailed, constructive feedback on submitted papers, as well as a fast
turn-around time.

All submissions will be handled electronically and must be in PDF format. Please submit your paper
(no page limits) through EDAS here.

The submitted papers should be original, unpublished work, and not currently under review for any
conference or journal. We understand how much effort goes into creating papers and, accordingly, we
aim at a careful, fair treatment of all papers. All of us are committed to doing everything to satisfy our
authors’ expectations! Regular issues will be accompanied by special issues focusing on particular
topics of broad interest within the telecommunications field. If you have any ideas for timely special
issues please do not hesitate to contact us.

I want to express my sincere thanks to all the authors of papers who accepted our invitation and
published their papers in this inaugural issue which includes ten papers that provide an in-depth analysis
of evolving technologies.

Examining the full potential of backscattering in the realization of Internet of Things, “Backscatter
communication with passive receivers: From fundamentals to applications” presents an overview of
recent innovations in hardware architecture for backscatter modulation. “Non-coherent massive MIMO-
OFDM for communications in high mobility scenarios” proposes the use of non-coherent demodulation
schemes to improve performance under scenarios of high mobility while also highlighting new potential
areas for research.

–v–
ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

The paper on “MSICA: Multi-Scale Signal decomposition based on independent component analysis
with application to denoising and reliable multi-channel transmission” examines the invaluable tools
needed in digital signal processing while “SDN-based Sociocast group communications in the Internet
of Things” looks into the literature around introducing new disruptive network-layer solutions to address
the challenges related to traditional group communication solutions which tend to lack control policies
on involved endpoints. Introducing a new paradigm, “The Internet of Metamaterial Things (IoMMT)
and other software enablers” explores how artificial materials with real time tunable physical properties
will significantly enrich the Internet of Things ecosystem. “Design and analysis of a reconfigurable
intelligent meta-surface for vehicular networks” also introduces a new paradigm, this time for vehicular
communications, through the manipulation of electromagnetic waves.

With the COVID-19 pandemic currently ongoing, “A blueprint for effective pandemic mitigation” offers
a timely focus on contact tracing as a traditional method for mitigating pandemics and the associated
challenges. This paper provides a framework on how to contain the spread of a pandemic through the
use of wireless technologies, providing numerical results to show the efficacy of the testing strategy.

Considering the central feature in 5G and beyond systems, network slicing, the paper “Machine
learning-assisted cross-slice radio resource optimization: Implementation framework and algorithmic
solution” provides a description of a feasible implementation framework for deploying ML-assisted
solutions for cross-slice radio resource optimization. Looking towards future networks including 6G,
authors of the paper “6G vision: An ultra-flexible perspective” provide an overview of the potential 6G
key enablers from the flexibility perspective and give a general framework to incorporate these enablers
into future networks. The paper further considers the role of artificial intelligence and integrated sensing
as key enablers within this framework.

Finally, “On the evolution of infrastructure sharing in mobile networks: a survey” provides a complete
picture of infrastructure sharing both over time and in terms of research branches that have stemmed
from it such as performance evaluation and resource management. This survey also highlights the
relation between infrastructure sharing and the decoupling of infrastructure from services, wireless
network virtualization and multi-tenancy in 5G networks. Such relation reflects the evolution of
infrastructure sharing over time and how it has become a commercial reality in the context of 5G.

We do hope that readers will enjoy these papers in the inaugural issue. More details about the journal
can be found at: https://www.itu.int/en/journal/j-fet/Pages/default.aspx

Many individuals have contributed to the launching of this journal and the preparation of the inaugural
issue. Primarily, I would like to express my gratitude to the ITU Secretary-General, Houlin Zhao for
entrusting me with the responsibility to lead this Journal as Editor-in-Chief. I would also like to thank
the Director of the ITU Telecommunication Standardization Bureau, Chaesub Lee for his support, and
Reinhard Scholl and Alessia Magliarditi who believed in my vision and keenly supported this journal
from day one. The incredible efforts of Alessia cannot be described in words. Her dedication and very
hard work helped to launch this journal successfully. Other team members, Erica Campilongo and
Simiso Dlodlo are the engine of the entire operation and their diligent work is always appreciated.

Our journal’s administration truly reflects the solid base for the success of this new journal.

– vi –
 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Our objective is to become the premier international forum for addressing all aspects of evolving and
future technologies in the telecommunications field. The ITU Journal will continue to publish online all
year round, welcoming papers at any time, on all topics within its scope with the aim of build bridges
between disciplines, connect theory with application, and stimulate international dialogue around the
future and evolution of the digital transformation underway across our economies, with five special
issues already underway for 2021. Achieving a significant impact factor is also our goal for this Journal
and will derive from the relevance of journal papers to the priorities of academia, industry and
governments, leading the way to new frontiers in research.

We all look forward to serving you and again special thanks for your interest to make this journal the
premier journal in our research community.

Dr. Ian F. Akyildiz

Ken Byers Chair Professor in Telecommunications Emeritus
Director of Broadband Wireless Networking Lab
School of Electrical and Computer Engineering
Georgia Institute of Technology
Atlanta, GA 300332, USA
[email protected]

– vii –
 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

EDITORIAL BOARD
Editor-in-Chief Daniel Denkovski, University Ss Cyril and
Methodius, North Macedonia
Ian F. Akyildiz, Georgia Institute of
Technology Emeritus Professor, USA Zafer Dogan, Koc University, Turkey
Hongzhi Guo, Norfolk State University,
Editors
USA
George C. Alexandropoulos, National and
Konstantinos Katsanos, National and
Kapodistrian University of Athens, Greece
Kapodistrian University of Athens, Greece
Marilia Curado, University of Coimbra,
Marios Lestas, Frederick University,
Portugal
Cyprus
Liljana Gavrilovska, Ss. Cyril and
Zhangyu Li, University at Buffalo, USA
Methodius University, Macedonia
Elena Simona Lohan, Tampere University,
Tolga Girici, TOBB University of
Finland
Economics and Technology, Turkey
Peng Lu, Intel Inc., USA
Ozgur Gurbuz, Sabanci University, Turkey
Augusto Neto, Federal University of Rio
Albert Levi, Sabanci University, Turkey
Grande do Norte (UFRN), Brazil
Andreas Pitsillides, University of Cyprus,
Sandro Nizetic, University of Split - FESB,
Cyprus
Croatia
Joel J.P.C. Rodrigues, Federal University
Valentin Rakovic, Ss. Cyril and Methodius
of Piauí (UFPI), Brazil
University in Skopje, North Macedonia
Zhi Sun, University at Buffalo, The State
Ashwin Rao, University of Helsinki,
University of New York, USA
Finland
Sasu Tarkoma, University of Helsinki,
Viktoriia Shubina, Tampere University,
Finland
Finland
The full list of the ITU J-FET Editors is
Cristiano Silva, Universidade Federal de
available at https://www.itu.int/en/journal/j-
São João Del-Rei, Brazil
fet/Pages/editorial-board.aspx.
Petar Solic, University of Split, Croatia
Reviewers
Amee Trivedi, University of Massachusetts,
Mustafa Akkaş, Abant İzzet Baysal USA
University, Turkey
Alexey Vinel, Halmstad University,
A. Ozan Bicen, Sabanci University, Turkey Sweden
Vittorio Cozzolino, Technical University of
ITU Journal Team
Munich, Germany
Alessia Magliarditi, ITU Journal
Carl James Debono, University of Malta,
Coordinator
Malta
Erica Campilongo, Collaborator
Murat Demirtaş, TOBB University of
Economics and Technology, Turkey Simiso Dlodlo, Collaborator

– ix –
 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

TABLE OF CONTENTS
Page

Foreword by the ITU Secretary-General .............................................................................................. 3

Foreword by the TSB Director ............................................................................................................. 4

Editor-in-Chief’s Message .................................................................................................................... 5

Editorial Board ...................................................................................................................................... 9

List of Abstracts .................................................................................................................................... 13

Selected Papers
1. Backscatter communications with passive receivers: From fundamentals to
applications ................................................................................................................................. 1
Milutin Stanaćević, Akshay Athalye, Zygmunt J. Haas, Samir R. Das, Petar M. Djurić
2. Non-coherent massive MIMO-OFDM for communications in high mobility scenarios ............ 13
Kun Chen-Hu, Yong Liu, Ana Garcia Armada
3. MSICA: Multi-scale signal decomposition based on independent component analysis
with application to denoising and reliable multi-channel transmission ...................................... 25
Abolfazl Hajisami, Dario Pompili
4. SDN-based Sociocast group communications in the Internet of Things .................................... 37
Luigi Atzori, Claudia Campolo, Antonio Iera, Giuseppe Massimiliano Milotta,
Giacomo Morabito, Salvatore Quattropani
5. The Internet of MetaMaterial Things and their software enablers ............................................. 55
Christos Liaskos, Georgios G. Pyrialakos, Alexandros Pitilakis, Ageliki Tsioliaridou,
Michail Christodoulou, Nikolaos Kantartzis, Sotiris Ioannidis, Andreas Pitsillides, Ian
F. Akyildiz
6. Design and analysis of a Reconfigurable Intelligent Meta-surface for vehicular
networks...................................................................................................................................... 79
Mohammad Ojaroudi, Valeria Loscrí, Anna Maria Vegni
7. A blueprint for effective pandemic mitigation ........................................................................... 89
Rahul Singh, Wenbo Ren, Fang Liu, Dong Xuan, Zhiqiang Lin, Ness B. Shroff
8. Machine learning-assisted cross-slice radio resource optimization: Implementation
framework and algorithmic solution ................................................................................ 103
Ramon Ferrús, Jordi Pérez-Romero, Oriol Sallent, Irene Vilà, Ramon Agustí
9. 6G vision: An ultra-flexible perspective.......................................................................... 121
Ahmet Yazar, Seda Doğan Tusha, Huseyin Arslan
10. On the evolution of infrastructure sharing in mobile networks: A survey ...................... 141
Lorela Cano, Antonio Capone, Brunilde Sansò
Index of Authors ........................................................................................................................ 159

– xi –
 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

LIST OF ABSTRACTS
Backscatter communications with passive receivers: From
fundamentals to applications
Pages 1-11
Milutin Stanaćević, Akshay Athalye, Zygmunt J. Haas, Samir R. Das, Petar M. Djurić
The principle of backscattering has the potential to enable a full realization of the Internet of Things.
This paradigm subsumes massively deployed things that have the capability to communicate directly
with each other. Based on the types of excitation and receivers, we discriminate four types of
backscattering systems: (i) Dedicated Exciter Active Receiver systems, (ii) Ambient Exciter Active
Receiver systems, (iii) Dedicated Exciter Passive Receiver systems, and (iv) Ambient Exciter Passive
Receiver systems. In this paper, we present an overview of bacskscattering systems with passive
receivers which form the foundation for Backscattering Tag-to-Tag Networks (BTTNs). This is a
technology that allows tiny batteryless RF tags attached to various objects to communicate directly with
each other and to perform RF-based sensing of the communication link. We present an overview of
recent innovations in hardware architectures for backscatter modulation, passive demodulation, and
energy harvesting that overcome design challenges for passive tag-to-tag communication. We further
describe the challenges in scaling up the architecture from a single link to a distributed network. We
provide some examples of application scenarios enabled by BTTNs involving object-to-object
communication and inter-object or human-object dynamic interactions. Finally, we discuss key
challenges in present-day BTTN technology and future research directions.
View Article

Non-coherent massive MIMO-OFDM for communications in high

mobility scenarios
Pages 13-24
Kun Chen-Hu, Yong Liu, Ana Garcia Armada
Under scenarios of high mobility, the traditional coherent demodulation schemes (CDS) have a limited
performance, due to the fact that reference signals cannot effectively track the variations of the channel
with an affordable overhead. As an alternative solution, non‑coherent demodulation schemes (NCDS)
based on differential modulation have been proposed. Even in the absence of reference signals, they are
capable of outperforming the CDS with a reduced complexity. The literature on NCDS laid the
theoretical foundations for simplified channel and signal models, often single‑carrier and spatially
uncorrelated flat‑fading channels. In this work, the most recent results assuming orthogonal frequency
division multiplexing (OFDM) signaling and realistic channel models are explained, and the impact of
some hardware impairments such as the phase noise (PN) and the non‑linear high power amplifier (HPA)
are also considered. Moreover, new potential research lines are also highlighted.
View Article

– xiii –
ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

MSICA: Multi-scale signal decomposition based on independent component

analysis with application to denoising and reliable multi-channel transmission
Pages 25-35
Abolfazl Hajisami, Dario Pompili
Multi-scale decomposition is a signal description method in which the signal is decomposed into
multiple scales, which has been shown to be a valuable method in information preservation. Much
focus on multi-scale decomposition has been based on scale-space theory and wavelet transform. In this
article, a new powerful method to perform multi-scale decomposition exploiting Independent
Component Analysis (ICA), called MSICA, is proposed to translate an original signal into multiple
statistically independent scales. It is proven that extracting the independent components of the even and
odd samples of a digital signal results in the decomposition of the same into approximation and detail.
It is also proven that the whitening procedure in ICA is equivalent to a filter bank structure.
Performance results of MSICA in signal denoising are presented; also, the statistical independency of
the approximation and detail is exploited to propose a novel signal-denoising strategy for multi-channel
noisy transmissions aimed at improving communication reliability by exploiting channel diversity.
View Article

SDN-based Sociocast group communications in the Internet of Things

Pages 37-54
Luigi Atzori, Claudia Campolo, Antonio Iera, Giuseppe Massimiliano Milotta, Giacomo Morabito,
Salvatore Quattropani
The new applications populating the Future Internet will increasingly rely on the exchange of data
between groups of devices, dynamically established according to their profile and habits (e.g., a
common interest in the same software updates and services). This will definitely challenge traditional
group communication solutions that lack the necessary flexibility in group management and do not
support effective control policies on involved endpoints (i.e., authorized senders and intended receivers).
To address the cited issues, the idea of introducing new disruptive network-layer solutions has emerged
from recent literature. Among them, Sociocast has been theorized as an enabler of flexible interactions
between groups of devices tied by social relationships. In this paper we start from the concept of
Sociocast and propose a solution based on Software Defined Networking (SDN) for its implementation
at the network layer in the Internet of Things. The performance of Sociocast is studied and compared
to methods running at the application layer that provide similar features. Experimental results, achieved
through an emulation-based playground, confirm that the Sociocast approach allows for significant
reduction of signaling and data packets circulating in the network with respect to traditional approaches.
View Article

– xiv –
 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

The Internet of MetaMaterial Things and their software enablers

Pages 55-77
Christos Liaskos, Georgios G. Pyrialakos, Alexandros Pitilakis, Ageliki Tsioliaridou, Michail
Christodoulou, Nikolaos Kantartzis, Sotiris Ioannidis, Andreas Pitsillides, Ian F. Akyildiz
A new paradigm called the Internet of MetaMaterial Things (IoMMT) is introduced in this paper where
artificial materials with real-time tunable physical properties can be interconnected to form a network
to realize communication through software-controlled electromagnetic, acoustic, and mechanical
energy waves. The IoMMT will significantly enrich the Internet of Things ecosystem by connecting
anything at any place by optimizing the physical energy propagation between the metamaterial devices
during their lifetime, via “eco-firmware” updates. First, the means for abstracting the complex physics
behind these materials are explored, showing their integration into the IoT world. Subsequently, two
novel software categories for the material things are proposed, namely the metamaterial Application
Programming Interface and Metamaterial Middleware, which will be in charge of the application and
physical domains, respectively. Regarding the API, the paper provides the data model and workflows
for obtaining and setting the physical properties of a material via callbacks. The Metamaterial
Middleware is tasked with matching these callbacks to the corresponding material-altering actuations
through embedded elements. Furthermore, a full stack implementation of the software for the
electromagnetic metamaterial case is presented and evaluated, incorporating all the aforementioned
aspects. Finally, interesting extensions and envisioned use cases of the IoMMT concept are discussed.
View Article

Design and analysis of a Reconfigurable Intelligent Meta-surface for vehicular

networks
Pages 79-88
Mohammad Ojaroudi, Valeria Loscrí, Anna Maria Vegni
In this work, a new paradigm for vehicular communications based on Reconfigurable Intelligent Meta-
surfaces (RIMs) is presented. By using the proposed RIM, we are able to manipulate electromagnetic
waves in the half-space, since the element is reflective. The unit cell consists of a U-shaped designed
microstrip structure equipped with a pin diode and via a hole. In this study, two different reflection
modes are achieved for 1-bit data transferring in each state. By incorporating these two different
configurations together, the reflected phases in the proposed RIM surface can be controlled respectively
in 0∘ and 180∘. The proposed unit cell can provide a usable double negative functional characteristic
around 5.3 GHz. The main goal of this paper is the use of a multifunctional behavior RIM for vehicular
communications to code the transmitted wave. A novel phase distribution diagram is generated to
propagate in each angle. Moreover, two major electromagnetic modulation functions, beam forming
and space coding have been demonstrated. Finally, we show how the RIM can be employed for
vehicular communications, acting as a coated access point along the street. We derive the instantaneous
data rate at the receiver node, the outage probability and the channel capacity, as affected by different
beam widths, distances and vehicle speed.
View Article

– xv –
ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

A blueprint for effective pandemic mitigation

Pages 89-101
Rahul Singh, Wenbo Ren, Fang Liu, Dong Xuan, Zhiqiang Lin, Ness B. Shroff
Traditional methods for mitigating pandemics employ a dual strategy of contact tracing plus testing
combined with quarantining and isolation. The contact tracing aspect is usually done via manual (human)
contact tracers, which are labor-intensive and expensive. In many large-scale pandemics (e.g., COVID-
19), testing capacity is resource limited, and current myopic testing strategies are resource wasteful. To
address these challenges, in this work, we provide a blueprint on how to contain the spread of a
pandemic by leveraging wireless technologies and advances in sequential learning for efficiently using
testing resources in order to mitigate the spread of a large-scale pandemic. We study how different
wireless technologies could be leveraged to improve contact tracing and reduce the probabilities of
detection and false alarms. The idea is to integrate different streams of data in order to create a
susceptibility graph whose nodes correspond to an individual and whose links correspond to spreading
probabilities. We then show how to develop efficient sequential learning based algorithms in order to
minimize the spread of the virus infection. In particular, we show that current contact tracing plus testing
strategies that are aimed at identifying (and testing) individuals with the highest probability of infection
are inefficient. Rather, we argue that in a resource constrained testing environment, it is instead better
to test those individuals whose expected impact on virus spread is the highest. We rigorously formulate
the resource constrained testing problem as a sequential learning problem and provide efficient
algorithms to solve it. We also provide numerical results that show the efficacy of our testing strategy.
View Article

Machine learning-assisted cross-slice radio resource optimization:

Implementation framework and algorithmic solution
Pages 103-120
Ramon Ferrús, Jordi Pérez-Romero, Oriol Sallent, Irene Vilà, Ramon Agustí
Network slicing is a central feature in 5G and beyond systems to allow operators to customize their
networks for different applications and customers. With network slicing, different logical networks, i.e.
network slices, with specific functional and performance requirements can be created over the same
physical network. A key challenge associated with the exploitation of the network slicing feature is how
to efficiently allocate underlying network resources, especially radio resources, to cope with the spatio-
temporal traffic variability while ensuring that network slices can be provisioned and assured within
the boundaries of Service Level Agreements / Service Level Specifications (SLAs/SLSs) with
customers. In this field, the use of artificial intelligence, and, specifically, Machine Learning (ML)
techniques, has arisen as a promising approach to cater for the complexity of resource allocation
optimization among network slices. This paper tackles the description of a feasible implementation
framework for deploying ML-assisted solutions for cross-slice radio resource optimization that builds
upon the work conducted by 3GPP and O-RAN Alliance. On this basis, the paper also describes and
evaluates an ML-assisted solution that uses a Multi-Agent Reinforcement Learning (MARL) approach
based on the Deep Q-Network (DQN) technique and fits within the presented implementation
framework.
View Article

– xvi –
 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

6G vision: An ultra-flexible perspective

Pages 121-140
Ahmet Yazar, Seda Doğan Tusha, Huseyin Arslan
The upcoming sixth generation (6G) communications systems are expected to support an unprecedented
variety of applications, pervading every aspect of human life. It is clearly not possible to fulfill the
service requirements without actualizing a plethora of flexible options pertaining to the key enabler
technologies themselves. At that point, this work presents an overview of the potential 6G key enablers
from the flexibility perspective, categorizes them, and provides a general framework to incorporate
them in the future networks. Furthermore, the role of artificial intelligence and integrated sensing and
communications as key enablers of the presented framework is also discussed.
View Article

On the evolution of infrastructure sharing in mobile networks: A survey

Pages 141-157
Lorela Cano, Antonio Capone, Brunilde Sansò
Infrastructure sharing for mobile networks has been a prolific research topic for more than three decades
now. The key driver for Mobile Network Operators to share their network infrastructure is cost
reduction. Spectrum sharing is often studied alongside infrastructure sharing although on its own it is a
vast research topic outside the scope of this survey. Instead, in this survey we aim to provide a complete
picture of infrastructure sharing both over time and in terms of research branches that have stemmed
from it such as performance evaluation, resource management etc. We also put an emphasis on the
relation between infrastructure sharing and the decoupling of infrastructure from services, wireless
network virtualization and multi-tenancy in 5G networks. Such a relation reflects the evolution of
infrastructure sharing over time and how it has become a commercial reality in the context of 5G.
View Article

– xvii –
 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

BACKSCATTER COMMUNICATIONS WITH PASSIVE RECEIVERS: FROM

FUNDAMENTALS TO APPLICATIONS

Milutin Stanaćević1 , Akshay Athalye1 , Zygmunt J. Haas2,3 , Samir R. Das4 , Petar M. Djurić1
1
Electrical and Computer Engineering, Stony Brook University, Stony Brook, NY 11794, 2 Computer Science,
University of Texas at Dallas, Richardson, TX 75080, 3 School of Electrical and Computer Engineering, Cornell
University, Ithaca, NY 14853, 4 Computer Science, Stony Brook University, Stony Brook, NY 11794,
NOTE: Corresponding author: Milutin Stanaćević, [email protected]

Abstract – The principle of backscattering has the potential to enable a full realization of the Internet of Things.
This paradigm subsumes massively deployed things that have the capability to communicate directly with each other.
Based on the types of excitation and receivers, we discriminate four types of backscattering systems: (i) Dedicated
Exciter Active Receiver systems, (ii) Ambient Exciter Active Receiver systems, (iii) Dedicated Exciter Passive Receiver
systems, and (iv) Ambient Exciter Passive Receiver systems. In this paper, we present an overview of bacskscattering
systems with passive receivers which form the foundation for Backscattering Tag-to-Tag Networks (BTTNs). This
is a technology that allows tiny batteryless RF tags attached to various objects to communicate directly with each
other and to perform RF-based sensing of the communication link. We present an overview of recent innovations in
hardware architectures for backscatter modulation, passive demodulation, and energy harvesting that overcome design
challenges for passive tag-to-tag communication. We further describe the challenges in scaling up the architecture
from a single link to a distributed network. We provide some examples of application scenarios enabled by BTTNs
involving object-to-object communication and inter-object or human-object dynamic interactions. Finally, we discuss
key challenges in present-day BTTN technology and future research directions.
Keywords – Backscatter-based communication, baterryless tags, Internet of Things, protocols, tag-to-tag networks

1. INTRODUCTION rise to different classes of systems and networks. In the

broad literature, backscatter systems are classified based
on the source of excitation into two types:
The promise of the Internet of Things (IoT) has fo-
mented research into a wide array of wireless technolo- 1. Dedicated exciter (DE) systems: a source of ex-
gies and devices capable of providing the required ubiq- citation is deployed specifically for the purpose of
uitous connectivity at a very large scale. In order to enabling backscatter transmissions, and
maximize the everywhereness and scalability of the IoT,
such devices should satisfy the following two key re- 2. Ambient exciter (AE) systems: backscatter trans-
quirements: (i) very low power consumption allowing missions leverage preexisting sources of excitation
for batteryless operation and (ii) direct communication in the environment such as TV towers, WiFi APs
with one another and networking without the need for and cell phone towers.
a central master controller. It is in this context that Independent of the excitation source, we posit that an
backscattering technology has seen a rapid emergence equally important classifying feature of backscatter sys-
in recent years, beyond its traditional uses in radar and tems is the type of receiver. Based on this, we identify
more recently in Radio Frequency Identification (RFID). the following two subclasses of backscatter systems:
Backscattering is a form of wireless transmission based
on modulated reflection of external RF signals. Since 1. Active receiver (AR) systems: the receiver is
the source of the RF signal is external, such transmis- a device with an on-board radio transceiver ca-
sion does not require an ‘active’ radio transceiver, al- pable of IQ demodulation and carrier cancella-
lowing devices to function in an extremely low power tion resulting, typically, in a very high sensitivity
regime (under 10 𝜇𝑊). The power needed to operate (down to −110 dBm for data rate of 20 kb/s and
the transmitter can be harvested from the external RF 30 mW power consumption for commercially avail-
signal itself, and thus it is possible for such devices to able transceivers [1]), and
be batteryless.
2. Passive receiver (PR) systems: the receiver is a
While the backscattering transmitter is a necessary con- radio-less passive device using an envelope detec-
stituent of every backscatter system, the type of RF ex- tor for signal demodulation resulting in a much
citation source and the type of receiver can vary, giving lower sensitivity (−56 dBm for data rate of 8 kb/s

© International Telecommunication Union, 2020 1

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

and 236 nW power consumption as in a Bluetooth be used as devices to ubiquitously tag “things” on a
wakeup receiver [2]). large scale.
In summary, the BTTN paradigm utilizing DE or AE
can provide a common medium or language for direct
Passive Receiver(PR)

er
cit
Ex Backscattering Tag-to-Tag Networks (BTTN)
peer-to-peer communication between all constituent ob-
Ambient jects of the IoT irrespective of cost, volume or density
of object population. In a sense, the excitation source
source

Tag Tag Tag in BTTN can be thought of as simply illuminating an

Tag
area, and the tags can “see” each other in this “light.”
Whether the illumination comes from a natural source
Monostatic
(AE approach) or a light bulb (DE approach), the com-
munication between the tags is unaltered.
Active Receiver(AR)

g
Exciter + Ta
Ambient Tag
Active Rx.

1.1 Scalability and practical implications of

source

Bista
tic Active
the AE approach
Exciter Active
Generally, AE backscatter systems are considered to be
Rx.
Rx.

extremely scalable because they can theoretically en-

Dedicated Exciter(DE) Ambient Exciter(AE) able communication without any deployment. However,
there are important practical considerations that can
Fig. 1 – Classification of backscatter systems. All such systems
adversely affect this in a real-world IoT deployment.
contain a backscattering transmitter (Tag). They are classified
based on the source of excitation signal and the type of receiver
• AEAR systems: There are widespread efforts in the
employed. literature on these systems. The systems are based
on the idea of building passive tags that can use
Combining the above criteria, we classify all backscatter ambient excitation to synthesize backscatter pack-
systems into four types as follows: ets that are compatible with commodity standards
such as WiFi, Bluetooth or ZigBee [4, 5, 7]. A
(i) Dedicated Exciter Active Receiver (DEAR) [3], corresponding commodity receiver can then receive
this signal and communicate with the tag. This
(ii) Ambient Exciter Active Receiver (AEAR) [4, 5, 6, approach significantly complicates the transmit cir-
7], cuitry on the tag. Further, such tags can only be
built to synthesize one kind of backscatter pack-
(iii) Dedicated Exciter Passive Receiver (DEPR) [8, 9, ets using one kind of excitation signal. Under this
10], and approach, the tags cannot communicate with each
other.
(iv) Ambient Exciter Passive Receiver (AEPR) [11].
• AEPR systems: These systems fall under the
This classification is shown in Fig. 1. As pointed out
BTTN umbrella. While this, in theory, enables
in [12], DEAR systems can be either monostatic where
maximum scalability, it is important to note the
the exciter and receiver functions are on the same device
ambient power level requirement. As shown in [13],
(e.g., traditional RFID) or bistatic with these functions
in order to enable a practical link distance, the re-
being on different devices.
quired power level is of the order of −25 dBm. Most
We focus specifically on passive receiver (PR) backscat- ambient excitation signals in general indoor envi-
ter systems which form the basis of the so-called ronments from various sources including TV tow-
Backscattering Tag-to-Tag Networks or (BTTNs). PR ers, cell phone towers or WiFi APs are far below
backscatter systems present fundamentally different this value.
challenges than AR backscatter systems. The challenges
stem from the need to passively receive the backscat- Thus, a practical implementation of BTTN might in-
ter signals in the presence of the interfering excitation variably call for a dedicated exciter. However, such an
with only an envelope detector. However, if these chal- exciter is simply an autonomous, RF transmitter that
lenges are overcome, then BTTN can fundamentally is not part of the communication network and does not
transform the capabilities of the IoT by enabling all pas- centralize the tag-to-tag communications (refer to the
sively tagged “things” to talk directly with each other light bulb analogy above). Furthermore, the BTTN
without any central active controller or master. In AR backscatter modulator is designed for tag-to-tag com-
systems whether of DEAR or AEAR variety, this capa- munication as opposed to tag to commodity receiver
bility is impractical since the high cost and high power- communication. This keeps the “language” and thereby
requirement of active receivers means that they cannot the design of the transmitter much simpler. Unlike the

2 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

AEAR approach where backscattering is tuned to one

format of the excitation signal, BTTN tags can talk to

Rx Signal Amplitude (mV)

each other irrespective of the excitation signal format Ideal link
whether it is a CW or a modulated signal. The only
Passive Receiver (PR) link in BTTN
requirement is that in the case of a modulated exciter
signal, the bandwidth of the tag-to-tag backscatter is
lower than the bandwidth of the excitation signal.

Most of the review papers in the literature on the pas-

sive backscatter communication focus on the backscatter
systems with active receiver(AR) [14] or group both the
AR and PR systems in the same category [12] without
addressing specific issues that exist in PR systems. In Tx to Rx Distance (m)
enabling a single BTTN link and further scaling up to
a larger network, a vast array of challenges needs to be
Fig. 2 – Characteristic of backscatter signal amplitude with in-
overcome. These stem mainly from having to process
creasing Tx to Rx distance in BTTN (PR) link compared to an
received signals and mitigate interference in passive re- ideal link without phase cancellation.
ceivers, to operate in an extreme low power regime (e.g.,
[15]), and to communicate in inherently high volume backscatter signal and the excitation signal causes a phe-
and high density networks. In the rest of this paper, we nomenon referred to as phase cancellation. This is seen
provide an overview of recent advances in BTTNs, chal- in Fig. 2 (the red solid line), where the received sig-
lenges, applications of BTTNs, and future directions for nal instead of monotonically decreasing with distance
research. undergoes alternating peaks and nulls, with decreasing
peak values [16, 17]. We note that this phenomenon will
also occur in so-called bistatic AR systems [12] where
2. THE FUNDAMENTALS the exciter and receiver are separate. On the other hand,
in monostatic AR systems like traditional RFID, the re-
One of the most challenging requirements of a PR is ceiver (reader) is able to cancel out the excitation sig-
to receive the inherently weak backscatter signal in the nal and use IQ demodulation for the received backscat-
presence of a much stronger interfering excitation with- ter. In this case the received signal amplitude decreases
out IQ demodulation or carrier cancellation capability. monotonically with distance (the blue dash-dot line).
We illustrate this challenge with a basic BTTN link con- Phase cancellation and low modulation index are two of
sisting of two tags in an area that sees a sufficient level of the most fundamental challenges in enabling basic com-
excitation signal whether DE or AE. In this basic link, munication in a PR BTTN link. The phase cancellation
at any given time, one of the tags transmits (Tx) and can be addressed using a multi-phase backscatter mod-
the other one receives (Rx). All BTTN tags are identi- ulator, while signals with low modulation indices are
cal, and they switch between Tx and Rx roles based on processed with demodulators with innovative architec-
the MAC-layer and network-layer protocols. tures [17, 16]. We note that in a link with AR, there
are alternative ways to avoid phase cancellation – by
The Tx tag generates the modulated backscatter sig- providing a frequency shift to the backscattered signal,
nal by altering the antenna’s reflection cross section. thereby avoiding interference with the excitation signal
The signal seen at the Rx tag is a superposition of the altogether [14].
excitation signal and the modulated backscatter signal
from the Tx tag. In the absence of an on-board radio
transceiver, the Rx tag has to demodulate the backscat- The maximum communication range of a BTTN link
ter signal using envelope detection. The received signal (Tx to Rx distance) depends critically on the excitation
has a very low modulation index due to the small ampli- power available at the Tx tag regardless of whether the
tude of the backscatter signal combined with the much excitation source is DE or AE. Distances up to about
larger magnitude of the exciter signal. Additionally, as 3 m have been reported with −20 dBm power available
the two signals combine at the Rx tag, the modulation at the Tx tag (5 kbps, BER below 10−3 ) [16]. Further
index is significantly impacted by the relative phase dif- improvement is possible using coding techniques such as
ference between the excitation signal and the backscat- CDMA, but at the expense of data rate. For example,
ter signal. [18] has reported distance up to about 10 m with sim-
ilar power levels but providing much slower bit rates,
In Fig. 2, we see the characteristic of the received base- in the order of 100 bps. Innovations in the demodula-
band backscatter signal amplitudes as a function of the tor design can improve the distance and/or improve the
Tx to Rx distance for an ideal link and a PR backscat- data rate. Other innovations are also possible including
ter link which constitutes BTTNs. In the PR BTTN multiple antennas on the tag [18] or beamforming using
link, the relative phase difference between the received multiple tags via a collaborative arrangement. The con-

© International Telecommunication Union, 2020 3

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

RF energy harves�ng module tenna between different values (or states). This in turn
changes the amplitude and/or the phase of the reflected
Power Power
Supercapacitor signal in accordance with the data to be transmitted.
Harves ng Management
Vdd
This is the conventional backscatter modulation process.
In a monostatic AR backscattering system like standard
Demodulator
Computa onal RFID, the impedance is typically varied between two
Logic
Comm values selected so as to maximize the modulation index
Control of the received signal at the reader in the two states. In
Modulator Memory a BTTN link on the other hand, the backscatter modu-
lation index depends on the relative phase difference be-
Communica�on module Computa�on module tween the exciter signal and the backscatter signal seen
at the Rx tag. In order to overcome this problem, it was
Fig. 3 – The architecture of a backscattering RF tag.
proposed in [8, 17] that the backscatter modulator has
Comm &
the ability to introduce a variable phase offset into the
Data Ctrl backscattered signal. At some value of the phase offset,
the backscatter signal and the excitation signal will be
Antenna
ZL1 Antenna
Ctrl
ZL2
in phase at the Rx tag resulting in the maximum re-
ceived backscatter amplitude. When the phase offset is
Envelope Amplifier + Comparator rxData
Detector BP filter

ZLn (b) shifted by 𝜋/2 from this value, the received backscat-
RF Switch
ter amplitude is minimum. The variable phase offset is
(a) achieved by switching the tag antenna impedance be-
tween a range of systematically designed values; each
Fig. 4 – Circuit implementation of (a) modulator and (b) demod-
ulator backscattering-based RF tag.
such impedance corresponds to one phase in a set of
phases that span the range from −𝜋/2 to 𝜋/2, as illus-
tributions to the backscatter PR tag-to-tag (T2T) links trated in Fig. 4(a). The number of different phases is a
are summarized in Table 1. trade-off between the achieved voltage difference in the
received signal, communication data rate and the tag
3. TAG HARDWARE form factor.
The overall architecture of the BTTN tag is shown in
Fig. 3. It has three modules, a communication mod- 3.1.2 Demodulator architecture
ule, an energy harvesting module, and a computation Demodulating the weak backscatter signal is a funda-
module. The tag optionally interfaces to an external mental challenge in PR backscatter systems because in
near-zero power sensor. While the sensing and computa- the absence of an active radio, the tags need to rely on a
tional module greatly depend on the application, the en- passive envelope detector for demodulation. The Rx tag
ergy harvesting and communication modules are similar must resolve a weak backscatter signal from the presence
across a wide range of BTTN tags and will be described of a much stronger external excitation signal resulting
in greater detail. The control logic manages the opera- in a low modulation index input signal to the demod-
tion of the tag while the computational logic, based on ulator. The communication distance of the BTTN link
the collected data, deduces information on the tag’s en- is directly related to the modulation index that a de-
vironment. The power consumption of the BTTN tag modulator can resolve [13]. The demodulator uses an
is on the order of a few 𝜇W as the operating frequency envelope detector that serves as an analog front-end for
typically does not need to exceed 1 MHz due to a 10s extraction of the baseband signal. For the detection
of kbps data rate in a tag-to-tag communication link. and demodulation, this analog front-end is followed by
The critical resource that requires careful optimization a comparator. Because of the much smaller modulation
on the system level is memory, both volatile and non- index in the received signal, using conventional RFID
volatile. tag demodulator architecture leads to short distances
of communication [9]. By inserting an amplifier with
3.1 Communication Module high-pass filtering after the envelope detection, as illus-
The communication module of the BTTN tag incorpo- trated in Fig. 4(b), a tailored demodulator for a tag-
rates the passive backscattering transmitter and the PR. to-tag link can demodulate signals with a modulation
These operations are implemented, respectively, by the index as low as 0.5% [13]. The sensitivity of this ar-
modulator and demodulator sections. chitecture is related to the power consumption of the
amplifier. The ripple voltage in the baseband signal is
a critical parameter that determines the performance of
3.1.1 Modulator architecture
the demodulator. To reduce the ripple voltage, higher-
The modulator of the BTTN tag generates the backscat- order adaptable low-pass filtering could be integrated in
ter signal by varying the impedance of the tag an- the envelope detector prior to signal amplification at a

4 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Table 1 – Summary of the contributions to backscatter PR tag-to-tag (T2T) links

Article Key contribution Frequency Experimental Results

[9] T2T communication concept 915 MHz T2T link at 10 cm

[11] T2T link with ambient exciter 539 MHz 1 kbps at 0.75 m and -8 dBm
[18] coding technique to extend link distance 915 MHz 0.003 kbps at 6 m and -20 dBm
multi-antenna tag for increased data rate 539 MHz 1000 kbps at 2.1 m and -20 dBm
[17] phase cancellation in T2T link 915 MHz
[13] theoretical analysis of T2T link 915 MHz
[16] demonstration of multi-hop network 915 MHz 5 kbps at 3 m and -20 dBm
[19] M-PSK for increased data rate 539 MHz 20 kbps at 0.75 m
[10] MAC protocol 915 MHz multi-hop T2T links reach 5.65 m

cost of chip area and power consumption. tion between the tags and computation are of the same
order of magnitude. Further, the different energy costs
3.2 Energy harvesting architecture for performing different operations on the tags lead to a
unique power management paradigm for BTTN.
The RF energy harvesting module acquires energy from
the external excitation signal. A power harvesting cir-
4. SCALING FROM A SINGLE LINK
cuit comprises rectification of the incident AC voltage,
followed by multiplication and regulation that provides TO A FULL NETWORK: ROUT-
stable DC supply voltage for the operation of the tag. ING FOR IOT APPLICATIONS
The energy efficiency of the conventional power har-
4.1 From a Link to a Network
vesting circuit is optimized for a certain range of in-
put power. As the input power can exceed the power Extending a single link communication to a tag network
consumption of the tag, the extra energy can be stored is far from trivial [20]. Two issues need to be consid-
using a supercapacitor. This enables the operation of ered: topology formation and routing. The topology
the tag when the harvested energy is lower than the formation involves selection of network links for com-
instantaneous power consumption. The size of the su- munication based on the energy states of the individual
percapacitor is limited by the form factor of the tag. tags. This decision typically involves tags beyond local
The power management logic optimizes the charging of neighborhoods and may require dynamic operation as
the supercapacitor based on the incident RF power and the tags’ energy states continuously vary.
the power needs of the tag operation. For communication across a tag-to-tag link, there must
be enough RF power reaching the Rx tag to power up
Based on the incident RF power, the stored energy and the receiving tag for effective demodulation and then
the operation of the tag, e.g., backscatter, receive or to do any needed post-demodulation computation (e.g.,
compute, the power management module directs the MAC, routing decisions). The power needed for effective
tag’s operation. The operation of such tags powered by demodulation is dependent on the modulation index,
RF harvested energy and low capacity supercapacitors which in turn depends on the wireless channel condi-
introduces some unique challenges compared to those tions that determine the powers reaching the tags. This
of a traditional sensor node. Sensor nodes incorporate is heavily influenced by the tag and exciter locations.
active radios that dominate the power budget. Though As mentioned in Section 3.2, the energy management
significant steps have been made in reducing their power module decides the power split among the various oper-
consumption at the receiver end [1], the transmit power ations. The Rx tag in a weak link may have to decide
still dominates the operation as the radios must gener- whether to receive a packet at all if it may not be able
ate the RF carrier signal used for communication. The to forward it immediately for a lack of enough avail-
principal difference between the power budgets of con- able power. Similarly, a more “energy-rich” tag may be
ventional sensor nodes and the RF tags is that the tags able to take up more responsibilities for routing or MAC
operate at orders of magnitude of lower power consump- protocol decisions.
tion due to the low energy cost of the communication, as
the energy cost of their communication can be orders of
4.2 Routing and MAC
magnitude lower than for the nodes comprising active
radios. This is because tags only reflect (backscatter) The challenges of designing routing and MAC protocols
externally supplied RF signals and do not generate any for BTTN arise from the unique characteristics of the
signal on their own. However, in BTTN tags there is backscattering environment, including the extreme low-
no such dominance – the energy costs for communica- power operation and from the intended BTTN applica-

© International Telecommunication Union, 2020 5

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

tions [21]. In this section, we discuss these challenges preserve connectivity and to maintain optimal routes.
and propose some solutions that have been considered Furthermore, the fact that these changes occur with lit-
in this field. tle delay, and thus little advanced warning, and the fact
that these changes may occur frequently, even more ex-
First, in BTTNs there is strong dependence of the range acerbate the problem.
over which a tag can communicate on the distance be-
tween the tag and the source of its backscattered RF Third, the backscattering tags experience high level
energy, whether an ambient source or an RF exciter is of interference at various protocol layers. Interference
used. In other words, a tag which is located close to is generally not a problem in sparse networks or net-
the RF energy source will be able to backscatter over a works with infrequent communications among the net-
larger distance, than the same tag if it is placed further work nodes. But in the envisioned applications of BTTN
away from the RF energy source. This is, of course, dif- ([21]), such as those for densely deployed IoT systems,
ferent than a typical sensor/ad hoc networks, where the even a simple query might cause at least some mes-
length of a link depends on the node itself and does not sage flooding among the tags, significantly affecting the
change with the location of the node. This has a num- throughput of big portions of the network. This problem
ber of implications in the design of the routing and MAC is further intensified in real-time IoT applications.
protocols, including the fact that the set of destination Fourth, in some configurations where the RF exciter is
of a node in a BTTN depends on the location of the tasked with at least part of the computational functions,
RF energy source relative to the node. Because of this some part of the routing and MAC processing could be
phenomenon, there is also a larger likelihood of unidirec- done by the exciter [20], offloading some of the complex-
tional links being present between two communicating ity from the tags. On the other hand, in the case of a
tags, when the two tags are at different distances from zero-intelligence exciter or when an ambient RF source
the RF source. (In general, this likelihood depends on is utilized, all the computations need to be performed
the tags and the distribution of RF sources.) As BTTNs distributively by the tags themselves. Thus, to adjust
tend to be distributed in their operation (i.e., there is to different operational scenarios, the routing and MAC
no central element that coordinates the MAC access or protocols may need to adapt to the division of processing
the routing discovery operations) the need to perform between the tags and the exciters. Furthermore, in the
these operations over unidirectional links is a more dif- case of a zero-intelligence exciter, distributed processing
ficult problem than in undirected networks ([22]), often is especially a challenge, since the BTTN tags operate at
leading to network partitions in the unidirectional graph extremely low energy levels, significantly limiting their
type. Of course, preservation of network integrity is crit- processing capabilities. Depending on the limited pro-
ical for most networking environments. This is unlike cessing capabilities of the tags and their extreme low-
other typical wireless networks where each node is pow- power operation, there is a need for new approaches to
ered by its own battery, thus creating links with similar design very simple MAC and routing protocols [20, 21],
capabilities in the two directions, a fact that is quite of- such as by trading the protocols’ performance for pro-
ten relied upon in the design of the protocols. (E.g., if cessing complexity.
node A sends a message to node B, it is given that node
B expects to be able to reply to node A on the link in In applications such as IoT, the network of tags should
the reverse direction.) facilitate interactions among smart objects, each tagged
with a passive tag that stores information about the
Second, as multiple tags are usually powered by a single object, such as the object’s identity, its capabilities, at-
RF source, any increase or decrease in the source’s RF tributes, and past history of interactions with other ob-
power is likely to drastically affect large portions of the jects. As an example, if the BTTN is designed to track
network topology. Thus, movements of nodes (of the RF infectious contacts among individuals, as to alert them
power sources) or changes in the RF propagation impair- of possible infection ([21]), the lists of contacts need to
ments of the RF sources could significantly, and more be stored and maintained in the tags. Routing among
problematically nearly instantaneously, affect large por- such passive tags, each being associated with a particu-
tions of the network topology. Similarly, a movement of lar object, requires creation of a suitable routing infras-
another RF source into the network coverage area would tructure and appropriate protocols. More specifically,
increase the lengths of at least some of the network links the routing functions consist mainly of: finding paths
– and typically of all the links in a particular area – between specific tags or among related tags; ensuring
thus, creating a topology with richer connectivity. This that the communication among the tags is expedited
is unlike other wireless networks (e.g., typical ad hoc or and takes information priority into consideration; maxi-
sensor networks), where the changes of topology caused mizing the network throughput, i.e., concurrent commu-
by a movement of a single node are mostly limited to in- nications among the tags; and reducing the interference
dividual nodes in the neighborhood of the moving node among the selected paths. Although the basic operation
only. Such sizable changes of BTTN topology require a of the routing protocol is to facilitate communication
much more robust and adaptable routing approach to among the tags, i.e., finding multi-hop routes among

6 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

the tags, routing should also facilitate higher-level op- the accumulated route. (Any node which receives the
erations, e.g., searching for a particular object, such as RREQ and knows the route to the destination, can cre-
other associated/related tags in the network (e.g., for ate an RREP by appending the known route to the
all the tags that were in close contact with a tag carried accumulated route in the RREQ and forwarding the
by an infected individual); querying to identify all the RREP back to the source node through reversing the
objects with certain attributes or certain historical val- accumulated route.) Unfortunately, the above process
ues, thus creating “communities of interest” among ob- will not work in a BTTN, because many links are uni-
jects to facilitate interactions and information exchange directional only, thus reversing the route will create an
among such member objects, etc. infeasible path. First, we note that our route discov-
ery operates between a source tag and a community of
As an example of an approach to routing in BTTN, we tags, rather than a single destination node. Second, a
now briefly discuss how to address two of the specific new RREQ/RREP process could be introduced, where
challenges of routing in BTTN: (a) routing scalability in a message Forward Route Request (FREQ) is broadcast
a densely-deployed network and (b) route discovery in by the source and propagates (with route accumulation).
the presence of unidirectional links in the network. When the FREQ is received by any member of the com-
In a massively deployed network, such as is envisioned munity of tags, such a node now becomes the destination
for IoT applications, it is difficult to discover whether node. The destination node, upon receipt of the FREQ,
a particular tag is reachable by another tag. To com- initiates a new Backwards Route Request (BREQ), by
bat this problem, the tags can establish loose associa- appending the forward route from the FREQ and broad-
tions, creating communities of interest – a collection of casting the BREQ back to the source. When the BREQ
related objects, which are interspersed by other objects arrives at the source, it now contains both, the forward
in the network. For example, all books in a library by and the backward routes, where the routes in the two
a particular author could be an example of a commu- directions are not necessarily the same. The source then
nity of interest. In this way, as further explained below, creates an RREP message with the backward route and
rather than routing a message to a particular book (i.e., uses the forward route to send the RREP to the desti-
a particular tag), a message is anycasted to the “com- nation.
munity of books by the author,” rather than unicasted
to a specific tag. Routing in the network is then per- 5. APPLICATIONS OF BTTNS
formed based on the attributes of a community. When In this section, we first explain a fundamental operation
a node moves away or changes its attributes, it removes of two tagged objects that will facilitate many applica-
itself from the particular community of interest. Once tions based on object interactions, then we describe an
a message is delivered to any member of a community application that involves human interactions, and finally
of interest (i.e., anycasted), based on the attribute of we list a number of possible applications of BTTNs.
the community, the member will then share the message
with all the other members of its community through
5.1 Object interactions
intra-community routes. In other words, we proposed a
two-level distributed routing hierarchy, where each tag By object interaction we mean exchange of information
maintains a route to some members of its community between two objects with attached tags that are in the
of interest, so that delivery to a particular member of a proximity of each other and that is used for some pur-
community of interest requires only delivery to one (i.e., pose. For example, tagged objects can localize them-
any) member of the community. The notion of commu- selves relative to one another or even in an absolute
nities of interest addresses a major challenge in routing sense if some tagged objects serve as anchors, that is,
in the network of tags – routing scalability. Instead of their locations are known. Tagged objects can also track
discovering routing paths between every pair of tags in other tagged objects in their neighborhood.
the network, routing within only a much smaller com-
munity of tags is needed. The central problem here is the estimation of distances
between communicating tags. One technique for dis-
We now discuss the second challenge – discovering rout- tance estimation is based on multiphase backscattering,
ing paths in unidirectional graphs. One approach to where a tag changes the phase offset of the signal that is
path discovery in ad hoc networks and sensor networks being backscattered in a systematic manner [23]. Sup-
is through broadcasting Route Request Query (RREQ), pose there are two tags, Tag 1 and Tag 2, where Tag
which is a message sent from the source node to the des- 1 acts as Tx tag with different phases. It can readily
tination node. As the RREQ propagates through the be shown that the square of the estimated amplitude
network, the nodes append their ID to the message, un- of the Rx Tag 2 at the output of the envelope detector
til the message reaches the destination. The destination is a sinusoid that is a function of the used phase offsets
extracts the accumulated route in the RREQ and cre- and a fixed parameter that carries information about the
ates the Route Reply Message (RREP), which is then distance between the two tags. When the roles of Tag
forwarded back to the source node through reversing 1 and Tag 2 are reversed, i.e., Tag 1 receives and Tag

© International Telecommunication Union, 2020 7

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

2 transmits, the same phenomenon occurs. Tag 1 now activities in such a fashion is already available in vari-
obtains a sinusoid but with a different fixed parameter. ous other radio technologies, BTTN provides a unique
It turns out that when the respective parameters of the approach due to its entirely batteryless operation, pos-
sinusoids are added, their sum is equal to 4𝜋𝑑/𝜆, where sibility of ubiquity and hence ability to measure a large
𝑑 is the distance between the tags and 𝜆 is the wave- number of tag-to-tag channels for very fine grain mea-
length of the excitation signal. From this relationship, surements.
the distance can readily be determined.

Further, with the same line of reasoning, the tags can

5.3 From smart cities to biomedicine
estimate Doppler shifts due to moving tags. Experimen- Since the introduction of the RFID technology in the
tal results suggest that tags can estimate Doppler shifts supply chain area about 15 years ago, the technical lit-
with about the same accuracy as that obtained by active erature has provided numerous articles that promote the
conventional RFID readers. Also, the median tracking concept of smart homes and smart cities. One can eas-
error based on data from two tags can be as low as 2.5 ily imagine a smart home with BTTNs, where the tags
cm [23]. equipped with sensors pepper the space of the home
and where many of them are placed on various types of
5.2 Human interactions objects. The location and tracking of such objects will
then readily be enabled by the functionality described in
An interesting application of BTTNs is related to hu- Section 5.1. Applications in smart cities include use on
man interactions [8]. Here we present a setting where structures like buildings, streets, bridges, and parking
BTTNs serve as a ‘device-free’ activity recognition sys- spaces. The tags (with attached sensors) can be tasked
tem [8]. Namely, when the tags in the network commu- to monitor air pollution, traffic, and availability of park-
nicate with each other, the backscatter channel state is ing spaces. If the tags’ density is high, these operations
influenced by the surrounding environment. The chan- can be completed with high spatial resolution. The BT-
nel state thus carries information that can be used for TNs can also be applied to perform the structural moni-
classification of dynamic activities that take place in the toring of buildings and bridges where abnormalities can
proximity of the tags. As explained earlier, with multi- be detected without actual sensing devices and instead
phase backscattering, the communication between two based on the changes in the backscattered signals due to
tags becomes more reliable. It turns out that this is not the developed abnormalities, (e.g., cracks can be found
the only advantage of the scheme. Multiphase backscat- by detecting changes in distances between two tags be-
tering also helps to quantify channel state information fore and after the appearance of a crack). BTTNs will
that can serve as a unique signature of activities which also find a number of applications in medicine, environ-
in turn allows for their accurate classification. mental sensing, precision farming, and manufacturing.
More specifically, when a Tx tag backscatters the exter- For more details and other applications, see a recent
nal signal with different phases, the Rx tag can compute review on ambient backscatter communication [12].
features of these signals. These features vary accord-
ing to the dynamic alterations of the multipath wire- 6. FUTURE RESEARCH DIREC-
less channel between the tags. When there is no one TIONS
near the communicating tags, the amplitudes of the re-
ceived signals with different phases have features that BTTNs offer a unique system to enable ubiquitous
can serve as no-activity features. Similarly, when a per- massively-deployed IoT. Being batteryless and small
son performs an activity near the tags, the signature of form factor, they can easily blend with everyday ob-
the features takes its own value and carries information jects and thus almost everything can become part of
about the activity. Clearly, it is important to identify the network. Current research has successfully proto-
good features that allow for accurate classification. For typed and evaluated single BBTN links, explored their
example, it has been found that the backscatter chan- ability to characterize the intervening wireless channel
nel phase, the backscatter amplitude, and the change in (RF sensing) with applications to localization, tracking
excitation amplitude between two multiphase probings and activity recognition. Current research has also pro-
have a high discriminatory power for classification [8]. duced theoretical studies on large-scale network routing
issues. But much still needs to be done to make BTTNs
Experimental results suggest that with signals provided practical and their applications realizable. One key issue
by a BTTN, one can recognize human activities with an is effective power harvesting and associated power man-
average error of about 6%. This was accomplished with agement, so that the optimal power is allocated to activ-
8 different activities and 9 individuals. Interestingly, ities such as communication, sensing, and computation
this level of performance is similar to that achieved by at all times. This may limit the computation needed
systems that use powered, active radios. The classifica- for routing and other application level signal processing
tion results were obtained by convolutional neural net- due to a limited power budget. These are trade-offs that
works (for details, see [8]). While the ability to recognize need to be explored in very dense deployments, e.g., tags

8 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

in a neighborhood can time multiplex various activities IEEE Circuits and Systems Magazine, vol. 15,
to achieve a network-level power management. Also, ef- no. 3, pp. 6–17, 2015.
fective distributed computing techniques are needed to
address processing limitation issues. [2] N. E. Roberts, K. Craig, A. Shrivastava, S. N.
Wooters, Y. Shakhsheer, B. H. Calhoun, and D. D.
These are quite exciting times of the IoT era. The in- Wentzloff, “A 236nw -56.5 dbm-sensitivity Blue-
vention of tags that can form BTTNs presents a spring- tooth low-energy wakeup receiver with energy har-
board for launching the concept of IoT to new heights. vesting in 65nm CMOS,” in 2016 IEEE Inter-
The possibility for connecting every tagged object in a national Solid-State Circuits Conference (ISSCC).
network that turns into a part of the IoT has finally IEEE, 2016, pp. 450–451.
become a reality.

BTTNs offer a range of research challenges. For exam- [3] J. F. Ensworth and M. S. Reynolds, “BLE-
ple, one of them is in energy harvesting and involves Backscatter: Ultralow-Power IoT Nodes Compat-
the design of an energy harvester with high energy ef- ible With Bluetooth 4.0 Low Energy (BLE) Smart-
ficiency over a wide input power range. Scalability and phones and Tablets,” IEEE Trans. on Microwave
routing discovery in the presence of unidirectional links Theory and Techniques, 2017.
are not trivial routing tasks. Further, in security, future
research should focus on balancing the security needs of [4] P. Zhang, D. Bharadia, K. Joshi, and S. Katti,
BTTNs with limited resource use on the tags. Signal “Hitchhike: Practical backscatter using commod-
processing on the tags is also difficult due to the limited ity WiFi,” in Proceedings of the 14th ACM Con-
computing power of the tags. Future work will reveal ference on Embedded Network Sensor Systems CD-
efficient ways of processing increasing amounts of data ROM, ser. SenSys ’16, New York, NY, USA, 2016,
by the tags and in a cooperative manner. p. 259–271.

The prospects of BTTNs are quite promising, creating a [5] D. Bharadia, K. R. Joshi, M. Kotaru, and S. Katti,
driving force for their further development. The number “Backfi: High throughput WiFi backscatter,”
of BTTN applications is simply staggering. In the near ACM SIGCOMM Computer Communication Re-
future, the hardware and computational aspects of the view, vol. 45, no. 4, pp. 283–296, 2015.
tags will continue to improve. Novel machine learning
methods, possibly designed for the tags only, will be [6] V. Iyer, V. Talla, B. Kellogg, S. Gollakota, and
developed, and more novel techniques in the networking J. Smith, “Inter-technology backscatter: Towards
of the tags will be invented. With all the anticipated internet connectivity for implanted devices,” in
progress, one may argue, the BTTNs will become the Proceedings of the 2016 ACM SIGCOMM Confer-
true backbone of the IoT and will bring to fruition many ence, 2016, pp. 356–369.
of the benefits that have been envisioned by the IoT
paradigm. [7] P. Zhang, C. Josephson, D. Bharadia, and S. Katti,
“Freerider: Backscatter communication using com-
7. CONCLUSIONS modity radios,” in Proceedings of the 13th Interna-
tional Conference on emerging Networking EXper-
In this article, we presented an overview of backscatter- iments and Technologies, 2017, pp. 389–401.
based communication among batteryless tags, the hard-
ware of the tags, the scaling from a single link to a full [8] J. Ryoo, Y. Karimi, A. Athalye, M. Stanaćević,
network, and the signal processing that is carried out S. R. Das, and P. M. Djurić, “BARNET: Towards
by the tags. Further, we listed a number of important activity recognition using passive backscattering
applications with networks composed of such tags. We tag-to-tag network,” in 16th Annual International
also discussed challenges that the tags and the network Conference on Mobile Systems, Applications, and
present, including challenges in energy harvesting, com- Services. ACM, 2018, pp. 414–427.
puting, networking, security, and distributed signal pro-
cessing and decision making. [9] P. V. Nikitin, S. Ramamurthy, R. Martinez, and
K. Rao, “Passive tag-to-tag communication,” in
ACKNOWLEDGEMENT 2012 IEEE International Conference on RFID
The authors are thankful for the support of NSF under (RFID). IEEE, 2012, pp. 177–184.
Awards CNS-1901182, CNS-1763843 and CNS-1763627.
[10] A. Y. Majid, M. Jansen, G. O. Delgado, K. S.
Ytidtnm, and P. Pawetczak, “Multi-hop backscat-
REFERENCES
ter tag-to-tag networks,” in IEEE INFOCOM 2019-
[1] J. Blanckenstein, J. Klaue, and H. Karl, “A survey IEEE Conference on Computer Communications.
of low-power transceivers and their applications,” IEEE, 2019, pp. 721–729.

© International Telecommunication Union, 2020 9

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[11] V. Liu, A. Parks, V. Talla, S. Gollakota, D. Wether- [22] P. Sinha, S. Krishnamurthy, and S. Dao, “Scal-
all, and J. R. Smith, “Ambient backscatter: wire- able unidirectional routing with zone routing proto-
less communication out of thin air,” in ACM SIG- col (ZRP) extensions for mobile ad-hoc networks.”
COMM Computer Communication Review, vol. 43, IEEE, Sept. 23-28, 2000, pp. 1329–1339.
no. 4. ACM, 2013, pp. 39–50.
[23] A. Ahmad, Y. Huang, X. Sha, A. Athalye,
[12] N. Van Huynh, D. T. Hoang, X. Lu, D. Niyato, M. Stanaćević, S. R. Das, and P. M. Djurić,
P. Wang, and D. I. Kim, “Ambient backscatter “On measuring Doppler shifts between tags in
communications: A contemporary survey,” IEEE a backscattering tag-to-tag network with applica-
Communications Surveys & Tutorials, vol. 20, tions in tracking,” in ICASSP 2020-2020 IEEE In-
no. 4, pp. 2889–2922, 2018. ternational Conference on Acoustics, Speech and
Signal Processing (ICASSP). IEEE, 2020, pp.
[13] Y. Karimi, A. Athalye, S. Das, P. M. Djurić, and 9055–9059.
M. Stanaćević, “Design of backscatter-based tag-
to-tag system,” in IEEE International Conference AUTHORS
on RFID (RFID), 2017.

[14] C. Xu, L. Yang, and P. Zhang, “Practical backscat-

ter communication systems for battery-free internet
of things: A tutorial and survey of recent research,”
IEEE Signal Processing Magazine, vol. 35, no. 5,
pp. 16–27, 2018.

[15] Z. J. Haas and Z. Zheng, “Waveform design for

RF power transfer,” U.S. Patent Application, 15-
959,917, 2018.

[16] J. Ryoo, J. Jian, A. Athalye, S. R. Das, and

M. Stanaćević, “Design and evaluation of “BTTN”: Milutin Stanaćević received the B.S. degree in Elec-
A backscattering tag-to-tag network,” IEEE Inter- trical Engineering from the University of Belgrade, Ser-
net of Things Journal, vol. 5, no. 4, pp. 2844–2855, bia and Ph.D. degree in Electrical and Computer Engi-
2018. neering from Johns Hopkins University, Baltimore, MD.
He is currently an Associate Professor in the Depart-
[17] Z. Shen, A. Athalye, and P. M. Djurić, “Phase can- ment of Electrical and Computer Engineering at Stony
cellation in backscatter-based tag-to-tag communi- Brook University, Stony Brook, NY. His research in-
cation systems,” IEEE Internet of Things Journal, terests include mixed-signal and RF circuit and system
vol. 3, no. 6, pp. 959–970, 2016. design. Dr. Stanaćević is a recipient of the National
Science Foundation CAREER award.
[18] A. N. Parks, A. Liu, S. Gollakota, and J. R. Smith,
“Turbocharging ambient backscatter communica-
tion,” in ACM SIGCOMM Computer Communi-
cation Review, vol. 44, no. 4. ACM, 2014, pp.
619–630.

[19] J. Qian, A. N. Parks, J. R. Smith, F. Gao, and

S. Jin, “Iot communications with 𝑚-psk modulated
ambient backscatter: Algorithm, analysis, and im-
plementation,” IEEE Internet of Things Journal,
vol. 6, no. 1, pp. 844–855, 2018.

[20] C. Liu and Z. J. Haas, “Multi-hop routing protocols

for rfid systems with tag-to-tag communication,” in Akshay Athalye is a Co-Founder of Scandent LLC,
Proceedings of the 36th IEEE Military Communi- New York, NY, USA, where he currently serves as a
cations Conference. IEEE, 2017, pp. 563–568. Chief Technology Officer. He is also an Adjunct Pro-
fessor with the Department of Electrical and Computer
[21] C. Liu, Z. J. Haas, and Z. Tian, “On the design Engineering, Stony Brook University, Stony Brook, NY,
of multi-hop tag-to-tag routing protocol for large- USA. He has been involved in RFID, Backscatter Com-
scale networks of passive tags,” in IEEE Open Jour- munications and related research more than a decade.
nal of the Communications Society. IEEE, 2020, He has received the IEEE Region 1 Technical Excellence
pp. 1035–1055. Award.

10 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

land, Kingston, RI, USA. He is a SUNY Distinguished

Professor and currently a Chair of the Department of
Electrical and Computer Engineering, Stony Brook Uni-
versity, Stony Brook, NY, USA. His research has been
in the area of signal and information processing with
primary interests in the theory of Monte Carlo-based
methods; Bayesian machine learning; signal modeling,
detection, and estimation; signal and information pro-
cessing over networks; RFID and the IoT. Recently, his
research has been applied to problems related to ma-
Zygmunt J. Haas has been with AT&T Bell Labora- chine learning methods for intrapartum fetal monitor-
tories from 1988 until 1995, and since 1995 with Cor- ing and brain signals. He has been invited to lecture
nell University, Ithaca, NY, USA. Since 2013, he also at many universities in the United States and overseas.
holds the title of a Professor and Distinguished Chair Prof. Djurić was a recipient of the IEEE Signal Pro-
at the University of Texas at Dallas, Richardson, TX, cessing Magazine Best Paper Award in 2007 and the
USA. He is a recipient of numerous awards and distinc- EURASIP Technical Achievement Award in 2012. In
tions, including IEEE Fellow, IET Fellow, EAI Fellow, 2008, he was the Chair of Excellence of Universidad
and Best Paper awards, the 2012 IEEE ComSoc WTC Carlos III de Madrid-Banco de Santander. From 2008
Recognition Award, and the 2016 IEEE ComSoc AHSN to 2009, he was a Distinguished Lecturer of the IEEE
Recognition Award. Signal Processing Society. He has been on numerous
committees of the IEEE Signal Processing Society and
of many professional conferences and workshops. He was
Editor-in-Chief of the IEEE Transactions on Signal and
Information Processing over Networks. Prof. Djurić is
a Fellow of IEEE and EURASIP.

Samir R. Das is Professor in the Department of Com-

puter Science at Stony Brook University. His research
interests are in wireless networking and mobile com-
puting, focusing on protocols, systems and performance
evaluation. He co-chaired the technical program com-
mittees of premier mobile networking conferences, in-
cluding ACM MobiHoc and ACM MobiCom. He served
on the editorial boards of IEEE/ACM Transactions on
Networking and IEEE Transactions on Mobile Comput-
ing.

Petar M. Djurić received the B.S. and M.S. degrees

in electrical engineering from the University of Belgrade,
Belgrade, Yugoslavia, respectively, and the Ph.D. degree
in electrical engineering from the University of Rhode Is-

© International Telecommunication Union, 2020 11

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

NON‑COHERENT MASSIVE MIMO‑OFDM FOR COMMUNICATIONS IN HIGH MOBILITY

SCENARIOS

Kun Chen‑Hu1 , Yong Liu2 , Ana Garcia Armada1

1
Deparment of Signal Theory and Communications of Universidad Carlos III of Madrid (Spain). E‑mails: {kchen,
agarcia}@tsc.uc3m.es, 2 Shanghai Research Center, Huawei Technologies (China). E‑mail: [email protected].

Abstract – Under scenarios of high mobility, the traditional coherent demodulation schemes (CDS) have a limited perfor‑
mance, due to the fact that reference signals cannot effectively track the variations of the channel with an affordable overhead.
As an alternative solution, non‑coherent demodulation schemes (NCDS) based on differential modulation have been proposed.
Even in the absence of reference signals, they are capable of outperforming the CDS with a reduced complexity. The literature
on NCDS laid the theoretical foundations for simpli ied channel and signal models, often single‑carrier and spatially uncorre‑
lated lat‑fading channels. In this work, the most recent results assuming orthogonal frequency division multiplexing (OFDM)
signaling and realistic channel models are explained, and the impact of some hardware impairments such as the phase noise
(PN) and the non‑linear high power ampli ier (HPA) are also considered. Moreover, new potential research lines are also
highlighted.
Keywords – 5G, channel estimation, MIMO, non‑coherent, OFDM.

1. INTRODUCTION state information (CSI). This overhead can be excessively

high for very fast time‑varying channels. In such cases, a
The Fifth Generation of mobile communications (5G) [1] signi icant number of reference signals is required for the
is the global standard for a uni ied wireless air interface, continuous tracking of the channel estimation [20].
which is capable of providing a great lexibility for a mul‑
titude of use cases. The three main requirements of those The works in the literature have provided the theoreti‑
services are enhanced mobile broadband (eMBB), mas‑ cal foundations to understand NCDS and point to some
sive machine type communications (mMTC) and ultra re‑ cases when they can outperform the traditional CDS, in
liable low‑latency communications (URLLC). Therefore, particular in scenarios with high mobility [20]. Also, re‑
the peak data‑rate is not the only feature to be improved, cent works show some combinations of NCDS with MIMO‑
but also an enormous number of connected devices and OFDM for the uplink (UL) and downlink (DL). In the
the latency‑sensitive services are taken into account. Also present work, convinced that NCDS is an idea whose time
there is an increasing interest in providing an adequate has come, we discuss the implementation of the NCDS in
service in high mobility scenarios [2] ‑ [4]. practical MIMO‑OFDM communication systems, assum‑
Orthogonal frequency division multiplexing (OFDM) with ing some realistic channels characterized by high mobil‑
multiple‑input multiple‑output (MIMO) [5] have been re‑ ity. We provide the details of how to integrate the differ‑
cently set as the radio techniques for the physical layer ential modulation [21] in the two‑dimensional resource
in 5G [1]. New frequency bands are proposed to be grid (time and frequency) provided by the OFDM. Addi‑
exploited to obtain more available bandwidth, such as tionally, we also show the performance of this combina‑
3.5 GHz and millimeter‑waves (mm‑Wave) [6], and thus, tion under the effects of the phase noise (PN) [22], [23] or
the existing services can be improved and new ones can high power ampli ier (HPA) [24], for both UL and DL, and
be implemented. The integration of massive MIMO is a its bene its as compared to the traditional CDS. Finally, a
must, not only to improve the average capacity of the discussion related to challenges and opportunities is pro‑
link, but also for the implementation of beam‑steering vided together with some concluding remarks, in order to
and beamforming to mitigate propagation losses in these stimulate the research on this promising topic.
new higher bands. Furthermore, the complexity of the
signal processing techniques need to be bounded to re‑ The remainder of the paper is organized as follows. Sec‑
duce the cost of the devices and the delay of the required tion 2 introduces the main differences between CDS and
operations. As an alternative to classical coherent de‑ NCDS, especially for high mobility scenarios. Section 3
modulation schemes (CDS), non‑coherent demodulation and Section 4 provide the details of how to integrate the
schemes (NCDS) [7–9] have been proposed recently to be NCDS with MIMO‑OFM for the UL and DL, respectively.
combined with massive MIMO systems [10] ‑ [19]. They Section 5 presents several numerical results to evaluate
are capable of avoiding the overhead produced by the ref‑ the proposed scheme under some realistic channel mod‑
erence signals due to the fact that the transmitted sym‑ els, providing an assessment of the achieved system per‑
bols can be recovered without the knowledge of channel formance. Finally, in Section 6, the conclusions follow.

© International Telecommunication Union, 2020 13

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

2. BENEFITS AND WEAKNESSES OF THE CO‑ signal and low‑frequency noise when a sinusoidal signal
HERENT AND NON‑COHERENT DEMOD‑ plus noise is applied to several types of detector. It is
ULATION SCHEMES shown that a considerable gain may be obtained by using
the (new at that time) coherent detector as compared to
2.1 Coherent demodulation schemes (CDS) the non‑coherent square‑law detector when the input sig‑
nal to noise ratio (SNR) is low. In [8], a complete theory
Well‑known coherent detection requires a replica of the of detection is presented for threshold reception, which
carrier at the receiver, with frequency and phase synchro‑ requires either a suitably weighted cross‑correlation of
nized, with the transmitted one, and an estimation of the the received data with the a priori known signal (CDS), or
channel attenuation and phase. Then, the received signal a suitably weighted autocorrelation of the received data
and a replica of the received version of all possible trans‑ with itself (NCDS). The Kineplex system developed by
mitted signals can be cross‑correlated to make a decision. Collins Radio Company introduced the technique of dif‑
CDS are widely used by many communication systems. In ferential phase shift keying (DPSK), as described in [9].
particular, they are used in 5G [1], where the advantages Today, NCDS have been re‑proposed as an alternative to
of MIMO‑OFDM are exploited, providing a high through‑ the traditional CDS due to the fact that they are able to
put through the use of the well‑known 𝑀 ‑ary quadrature recover the transmitted symbols without any CSI, that is,
amplitude modulation (QAM). With this modulation for‑ knowledge of the amplitude and phase of the carrier is not
mat, the information is transported in both the amplitude required. Hence, reference signals are no longer needed,
and phase of the carrier, making an ef icient use of the reducing the undesirable signalling overhead. This effect
transmission channel. However, these bene its come at is more relevant for high mobility and/or very frequency‑
the expense of transmitting some reference signals in or‑ selective scenarios. Additionally, the complexity of the
der to obtain accurate enough CSI, so that the effects pro‑ transceivers is signi icantly reduced. Typical approaches
duced by the propagation channel to the received symbols involve the detection of the signal energy of phase dif‑
can be equalized before a decision. When the channel is ferences. Despite its simplicity, non‑coherent detection
frequency‑selective, OFDM facilitates the implementation usually implies a 3 dB loss in SNR as compared to CDS.
of CDS due to the fact that each subcarrier can be consid‑ For this reason, it has just been used in a few communi‑
ered as having an independent lat‑fading channel, reduc‑ cation systems where low complexity was a primary re‑
ing the complexity of the equalization. quirement. Examples of application are Bluetooth [25],
The need to obtain accurate enough CSI is accepted in with a non‑coherent frequency shift keying (FSK) receiver
most communication systems, in particular when the or Zigbee [26], using DPSK.
channel impulse response remains quasi‑static for a cer‑ Differential modulation is one of the most frequently used
tain period of time and the number of antennas is not techniques for NCDS [12] ‑ [19]. In this case the infor‑
very large. Under these conditions, a reduced amount mation is encoded by computing the phase difference be‑
of reference signals are used in order to track the chan‑ tween the current complex data symbol and the previ‑
nel variations in time, frequency and space dimensions. ously transmitted symbol. At the receiver, a simple dif‑
On the other hand, if we would like to provide commu‑ ferential decoder is required, detecting the phase differ‑
nications in new challenging environments, such as high ence between two contiguous symbols. To apply this tech‑
speed trains, autonomous vehicles, etc. these are mainly nique, the constellation is constrained to have a constant
characterized by a signi icant Doppler spread due to the modulus, such as DPSK, and a single reference symbol is
high mobility. In these situations, the traditional CDS re‑ needed at the beginning of each stream to have an ini‑
quires an enormous amount of reference signals in order tial phase reference. This means a negligible overhead to
to continuously and accurately track the variations of the the system. It is also required that the channel response
channel, reducing considerably the overall ef iciency of of every two contiguous differential symbols should be
the system, as pointed out in [10], [11], [20]. Otherwise, if very similar, otherwise the differential decoder is not able
the CSI is not properly estimated, the performance of the to successfully recover the transmitted information data.
CDS is also seriously compromised. Moreover, when mas‑ This condition is usually met, even in fast varying chan‑
sive MIMO is considered, the procedures of channel esti‑ nels.
mation and the computation of the pre/post‑coding ma‑ Recently, the combination of NCDS with massive MIMO
trices may increase the complexity of the system. For ex‑ has been proposed in order to improve its performance
ample, the channel inversion of large dimension matrices leveraging the high number of antennas. In the context
for each subcarrier may be prohibitive for some real‑time of UL, [10], [11] showed that asymptotically NCDS can
applications when a zero‑forcing (ZF) criterion is chosen. achieve the same performance as CDS. Nevertheless, the
proposed technique that illustrated this idea required a
2.2 Non‑Coherent demodulation schemes very large number of antennas to get an acceptable per‑
(NCDS) formance. Then, [12] ‑ [16] proposed the use of DPSK to‑
gether with an averaging process performed at the base
Looking back in history, NCDS are older than CDS. In [7], station (BS) over the spatial domain after non‑coherent
a comparison is made of the output spectra comprising demodulation, in order to mitigate the effects of the chan‑

14 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

nel and noise. Moreover, they proposed the idea of multi‑

plexing the data of each user equipment (UE) in the con‑
stellation domain based on a joint‑symbol, which is a su‑
perposition of the symbols sent by several UEs. In the case
of DL, the combination of NCDS with MIMO has been un‑
til now based on block codes [27] ‑ [31]. However, their
application requires that the channel response remains
quasi‑static during the transmission of a block code, and
they also need a high SNR in order to provide an accept‑
able performance. Moreover, they have the problem that
they are not scalable and when the number of antennas
at the BS is very large, the design of these block codes be‑
comes unaffordable. Typically, only two and four trans‑
mit antennas are taken into account [27] ‑ [31]. More Fig. 1 – Time domain scheme in the OFDM resource grid when 𝐾 = 12
and 𝑁 = 14. The yellow box represents a reference symbol required by
recently, the combination of beamforming and NCDS has
the differential modulation.
been proposed in order to exploit the high number of an‑
tennas through compensating the path loss and enhanc‑
ing the quality of the link, and spatially multiplexing the
different UEs [18], [19]. In these cases a certain channel
knowledge is needed to point the beam towards the UE
through the beam‑management procedure, and the sig‑
nal is processed non‑coherently in each beam afterwards.
Even though the overhead is not completely eliminated,
the savings are considerable.

3. NCDS WITH MASSIVE MIMO FOR THE UP‑

LINK
We describe in this section how to integrate the NCDS
based on [12] ‑ [17] in a realistic communication sys‑ Fig. 2 – Frequency domain scheme in the OFDM resource grid when
tem for the particular scenario of UL. We consider one BS 𝐾 = 12, 𝑁 = 14 and ℐ𝑁 = {1, 8}. The yellow and blue boxes de‑
note the reference symbols required by the differential modulation and
equipped with 𝑉 antennas, which is simultaneously serv‑ phase difference estimation, respectively.
ing 𝑈 UEs. These UEs are constrained to have a reduced
number of antennas, typically single‑antenna devices. Let tion is performed, in this case between resources that be‑
us assume that the 𝑈 UEs are simultaneously transmitting long to the same frequency and contiguous symbols in the
𝑁 OFDM symbols. The OFDM signal has 𝐾 subcarriers, time domain. The differential encoding can be described
and the length of the cyclic pre ix (CP) is long enough to as
absorb the effects of the multi‑path channel. At the re‑ 𝑟𝑢 , 𝑛=1
𝑥𝑢𝑘,𝑛 = { 𝑢 𝑘,𝑛𝑢 , (1)
ceiver side, after removing the CP and performing a fast‑ 𝑥𝑘,𝑛−1 𝑠𝑘,𝑛−1 , 2 ≤ 𝑛 ≤ 𝑁
Fourier transform (FFT) to each block at each antenna of
the BS, we can process each subcarrier as one of a set of 1 ≤ 𝑢 ≤ 𝑈, 1 ≤ 𝑘 ≤ 𝐾,
𝐾 independent subchannels.
𝑢
where 𝑟𝑘,1 is the reference symbol transmitted at the 𝑘‑
3.1 Integration of differential encoding in th subcarrier of the irst OFDM symbol by the 𝑢‑th UE,
OFDM for high mobility scenarios 𝑠𝑢𝑘,𝑛 and 𝑥𝑢𝑘,𝑛 are the complex data and differential sym‑
bol, respectively, transmitted at the 𝑘‑th subcarrier and
Similar to CDS, the NCDS can be also implemented in an 𝑛‑th OFDM symbol by the 𝑢‑th UE. The data symbol 𝑠𝑢𝑘,𝑛
OFDM system [17], suitable for dealing with a doubly‑ belongs to a PSK constellation due to the fact that the
dispersive channel. The stream of differential symbols differential encoding can only transmit information in
produced by the differential encoding can be mapped in the phase component. However, this time‑domain im‑
the two‑dimensional resource grid provided by the OFDM plementation has the drawback of an increased latency
(time and frequency). According to [17], the way this and memory consumption, since this mapping scheme re‑
mapping is performed will signi icantly impact on the quires waiting for the reception of two complete OFDM
overall system, especially for high mobility scenarios. symbols in order to obtain 𝑠𝑢𝑘,𝑛 , due to the fact that it per‑
Typically, NCDS based on differential modulation is per‑ forms a differential decoding of two contiguous symbols
formed using the time domain scheme. This scheme is in the time domain. Also, it cannot be exploited when
represented in Fig. 1, where the red arrows indicate the the Doppler shift is very high, since any two consecutive
direction in which differential modulation and demodula‑ OFDM symbols will not face a similar channel response.

© International Telecommunication Union, 2020 15

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Alternatively, the OFDM frame enables exploiting the fre‑

quency dimension, and hence, the differential modulation
technique can be also implemented using the frequency
domain scheme (see Fig. 2). According to [17], the dif‑
ferential symbols are mapped into contiguous frequency
resources of the same OFDM symbol as

⎧ 𝑟𝑢 , 𝑘 = 1,
{ 𝑢 𝑘,𝑛 𝑢
𝑥𝑢𝑘,𝑛 = ⎨ 𝑥𝑘−1,𝑛 𝑝𝑘,𝑛 , 𝑘 = 2, 𝑛 ∈ ℐ𝑁 , (2)
{ 𝑢 𝑢
⎩𝑥𝑘−1,𝑛 𝑠𝑘−1,𝑛 , otherwise

1 ≤ 𝑢 ≤ 𝑈, 1 ≤ 𝑛 ≤ 𝑁,
𝑢 𝑢
where 𝑟1,𝑛 and 𝑝2,𝑛 are two reference symbols for differ‑
Fig. 3 – Mixed domain scheme in the OFDM resource grid when 𝐾 = 12
ent purposes, and the set ℐ𝑁 contains the indexes that and 𝑁 = 14. The yellow and blue boxes denote the reference symbols
𝑢
correspond to those OFDM symbols which carry 𝑝2,𝑛 . The required by the differential modulation and phase difference estimation,
irst kind of reference symbol is required for the differen‑ respectively.
tial demodulation as explained before. The second one function, where the dramatic reduction of reference sig‑
is required for the estimation of the phase difference be‑ nals can be observed. This policy mainly follows the fre‑
tween two subcarriers, consequence of the frequency‑ quency domain scheme, except for the edge subcarriers
domain mapping; see [17] for more details. We can see of the block, that follow a time domain scheme. This pro‑
that this scheme has a reduced latency and is robust posal cannot only signi icantly reduce the number of ref‑
against high Doppler shifts. Furthermore, it is reasonable erence symbols, but it is also capable of taking all advan‑
to assume that the channel responses of any two contigu‑ tages of a frequency domain scheme. Moreover, in the
ous subcarriers are similar due to the fact that the number case of time‑varying channels, only those complex sym‑
of subcarriers is always designed to be much larger than bols placed at both edge subcarriers may suffer from an
the number of taps of the channel. However, these bene‑ additional degradation, that can be easily mitigated by us‑
its come at the expense of an additional phase estimation ing some channel coding [16], [32] or spreading [33] tech‑
and compensation procedure. This additional phase com‑ niques.
ponent is very small and consequently can be neglected For the sake of conciseness and to ease the notation, the
for channels that are not very frequency‑selective. On the frequency domain scheme is the chosen one for the rest
other hand, this phase must be compensated for strong of the paper. Note that any of the presented techniques in
frequency‑selective channels. However, when diversity is the following sections can be straightforwardly adopted
exploited, only an additional reference pilot is required for both time and mixed domain schemes.
𝑢
for all OFDM symbols within the coherence time (𝑝2,𝑛 ),
which produces a negligible impact on overhead.
Both time and frequency domain schemes, presented in
3.2 Multi‑user multiplexing in the constella‑
[17], may introduce a signi icant overhead, if the number tion domain
of allocated resources is reduced (𝐾 ↓ and/or 𝑁 ↓). For For a single‑user case, the use of a constant modulus con‑
example, in scenarios of mMTC, the machine devices are stellation, such as DPSK [12] ‑ [15], is the only require‑
designed to send short packets of just a few bytes. The ment for the non‑coherent demodulation based on differ‑
adoption of any of the two presented schemes implies to ential detection. However, when a multi‑user scenario is
send a signi icant amount of reference symbols. Hence, considered, if we would like that all independent transmit
we propose a new mapping scheme named as mixed do‑ sources are transmitting in the same time‑frequency re‑
main scheme (see Fig. 3). Firstly, we differentially encode source (to increase the spectral ef iciency), the received
the data symbols as signals from these independent sources are summed up
and need to be conveniently separated [12]. Then, the
⎧ 𝑟𝑢 , 𝑗=1 choice of the constellation for each individual UE is crucial
{ 𝑢𝑗 𝑢
𝑥𝑢𝑗 = ⎨ 𝑥𝑗−1 𝑝𝑗 , 𝑗 = 2 , 1≤𝑢≤𝑈 (3) in order to produce joint‑symbols that belong to a joint‑
{𝑥 𝑢
𝑠 𝑢
⎩ 𝑗−1 𝑗−1 , 3 ≤ 𝑗 ≤ 𝐾𝑁 constellation from which it is possible to unambiguously
recover the transmitted data of all UEs.
where the 𝑗 denotes the resource index. Then, the differ‑ At the BS, after removing the CP and performing the DFT,
ential symbols 𝑥𝑢𝑗 are allocated to the two‑dimensional re‑ the received signal at the 𝑘‑th subcarrier, 𝑛‑th OFDM sym‑
source grid as bol and 𝑣‑th antenna can be expressed as
𝑥𝑢𝑘,𝑛 = 𝑥𝑢𝑗 ∣ (𝑘, 𝑛) = 𝑓(𝑗), 1 ≤ 𝑗 ≤ 𝐾𝑁 , (4) 𝑈
𝑣
𝑦𝑘,𝑛 = ∑ √𝛽𝑢 ℎ𝑢,𝑣 𝑢 𝑣
𝑘,𝑛 𝑥𝑘,𝑛 + 𝑤𝑘,𝑛 , (5)
where 𝑓(•) is the resource mapping policy function. Fig. 𝑢=1

3 shows a recommended example of a mapping policy 1≤𝑣 ≤𝑉, 1 ≤ 𝑘 ≤ 𝐾, 1 ≤ 𝑛 ≤ 𝑁,

16 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Individual constellation UE 1 Individual constellation UE 1

j j

Joint-constellation at the BS Joint-constellation at the BS

1.5j
-1 1 -1 1 1.707j

-j -j

Individual constellation UE 2 Individual constellation UE 2

-1.707 1.707
-1.5 1.5
0.707j
0.5j
-0.5 -1.707j
-1.5j
-0.707 0.707
0.5
-0.5j
-0.707j

Type B Equal Error Protection (EEP)

Fig. 4 – Example of two joint‑constellations, for two UEs and four symbols each. The symbols of UE1 are shown using different colours and the symbols
of UE2 are plotted using different markers.

where 𝛽𝑢 is the average power of the signal of the 𝑢‑th UE, symbol must be properly designed to enable the demodu‑
𝑣
𝑤𝑘,𝑛 denotes the additive white Gaussian noise (AWGN) lation of the transmitted information by all the UEs. Con‑
at 𝑘‑th subcarrier, 𝑛‑th OFDM symbol and 𝑣‑th antenna, sequently, the choice of the individual constellation is cru‑
distributed as 𝒞𝒩(0, 𝜎𝑤 2
); and ℎ𝑢,𝑣
𝑘,𝑛 corresponds to the cial to produce a robust joint‑constellation against inter‑
channel frequency response between the 𝑢‑th UE and the ference and noise effects. The most used constellations
𝑣‑th antenna at the 𝑘‑th subcarrier and 𝑛‑th OFDM sym‑ are the Type B [12] and equal error protection (EEP) [13].
bol, distributed as 𝒞𝒩(0, 1). For simplicity, we assume The constellation of the 𝑢‑th UE can be expressed as
here that the channel response is spatially uncorrelated,
while we will use more realistic channel models for per‑ 2𝜋
ℳ𝑢𝐵 = {√𝛽𝑢 exp (𝑖 𝑚) ∣ 0 ≤ 𝑚 ≤ 𝑀 − 1} , (8)
formance evaluation. Besides, note that the difference in 𝑀
𝛽𝑢 among different UEs may be due to the constellation 2𝜋 𝜋
design or to different propagation path loss. In the latter ℳ𝑢𝐸 = {exp (𝑖 ( 𝑚+ 𝑢)) ∣ 0 ≤ 𝑚 ≤ 𝑀 − 1} ,
𝑀 2𝑈
case, an accurate power control must be implemented to (9)
compensate this difference. for Type B and EEP, respectively, where 𝑀 denotes the
𝑣
According to [12], 𝑦𝑘,𝑛 is fed to the differential decoder number of symbols in the constellation. Fig.4 shows
and averaged over the spatial dimension as an illustrative example of these two types of joint‑
constellations, designed for the particular case of UL with
1 𝑉 𝑣
∗
𝑣 only two UEs, each of them using a 4‑DPSK. In the irst
𝑧𝑘,𝑛 = ∑ (𝑦𝑘−1,𝑛 ) 𝑦𝑘,𝑛 , (6)
𝑉 𝑣=1 case, all UEs have the same 4‑DPSK constellation and
are distinguished with a different amplitude. This pro‑
2 ≤ 𝑘 ≤ 𝐾, 1 ≤ 𝑛 ≤ 𝑁, duces the joint‑constellation also shown in the same ig‑
∗
where (•) is the complex conjugate operation and 𝑧𝑘,𝑛 ure, where we can see that all symbols are equally spaced
denotes the received joint‑symbol at the 𝑘‑th subcarrier providing a robustness against possible interference and
and 𝑛‑th OFDM symbol. When the number of antennas is noise terms. However, those UEs with a lower amplitude
large enough and making use of the Law of Large Num‑ will obtain a worse performance for the same noise condi‑
bers, 𝑧𝑘,𝑛 can be approximated as tions as compared to the stronger ones. Indeed, the aver‑
age distance of the symbols of UE 𝑢 = 1 (shown in differ‑
𝑈 ent colours) is much larger than the distance of UE 𝑢 = 2
𝑉 →∞
𝑧𝑘,𝑛 −−−→ 𝑠𝑘,𝑛 = ∑ 𝛽𝑢 𝑠𝑢𝑘,𝑛 (7) (plotted in different markers). In EEP, both UEs have the
𝑢=1 same amplitude, and then the same performance. Their
constellations differ in a rotation of 45°. However, this
2 ≤ 𝑘 ≤ 𝐾, 1 ≤ 𝑛 ≤ 𝑁,
option presents several symbols of the joint‑constellation
where 𝑠𝑘,𝑛 is the joint‑symbol at the 𝑘‑th subcarrier and (those placed in the middle) that are too close to each
𝑛‑th OFDM symbol. Note that the interference and noise other, degrading the performance.
terms are averaged out thanks to the large number of an‑ The design of optimal individual constellations for multi‑
tennas at the BS, otherwise the performance may be de‑ user NCDS that work well in realistic channel conditions
graded. More details are given in [12] ‑ [17]. is still a very challenging topic, due to the diverse effects
The performance of the overall multi‑user systems de‑ of the channel impairments and interference and the dif‑
pends on the constellation of the joint‑symbol. This joint‑ iculty to analyse them.

© International Telecommunication Union, 2020 17

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

DPSK Channel
UE1 Diff. OFDM Rx Diff.
Modu‑ OFDM Tx UE1
bits Encoding ant. 𝑣 = 1 Decoding
lation UE1 bits
Joint‑
DPSK UE2 bits
UE2 Diff. Channel OFDM Rx Diff. Constellation
Modu‑ OFDM Tx UE2 Decision
bits Encoding ant. 𝑣 = 2 Decoding UE𝑈 bits
lation

Spatial
⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
Combiner
DPSK Channel OFDM
UE𝑈 Diff. Diff.
Modu‑ OFDM Tx UE𝑈 Rx ant.
bits Encoding Decoding
lation 𝑣 = 𝑉
BS
Fig. 5 – Block diagram for UL, where different UEs are multiplexed in the constellation domain.

The block diagram of a UL system addressing a more gen‑ are obtained through a beam management procedure to
eral multi‑user case is shown in Fig. 5, from [17]. In this point the beams towards each UE. Once this is achieved,
system, several UEs map their data bits into complex sym‑ the data information is sent over a non‑coherently pro‑
bols that belong to a speci ic DPSK constellation. Then, cessed link. Reference [18] proposed a similar idea as‑
the differentially encoded symbols are transmitted into suming, however, an ideal case where the MUI is com‑
an OFDM signal through a multi‑path channel. The re‑ pletely mitigated by the beamforming. Meanwhile, the
ceived signal at each antenna of the BS after the OFDM combination of NCDS with a practical beamforming tech‑
receiver corresponds to a superposition of different sig‑ nique is proposed in [19], taking into account the residual
nals coming from each UE. Later, these differential sym‑ MUI.
bols are non‑coherently combined by using a differential In [19] the beam‑management procedure de ined in 5G
decoder and averaged over the spatial domain in order to [1] is suggested to be performed as a irst step. This pro‑
obtain the desired joint‑symbol. Exploiting the spatial di‑ cedure is responsible for accurately determining the an‑
versity in order to reduce the noise and multi‑user inter‑ gle of the spatial clusters of the propagation channel con‑
ference (MUI) is crucial to obtaining a right decision from tributing to the signal of each UE, by transmitting some
the joint‑symbol [12]. Additionally, the constellation de‑ reference signals. These reference signals are speci i‑
sign can be combined with channel coding, where the soft cally the synchronization signals (SS) and channel state
information can either improve the performance or re‑ information‑reference signals (CSI‑RS). The former are
duce the number of required antennas at the BS [16] [32]. used when a UE would like to enter the system for the irst
time, while the latter are exploited for updating the angu‑
4. NCDS WITH MASSIVE MIMO FOR THE lar position of an existing UE in the system.
Then, the BS transmits one or several differential data
DOWNLINK
streams to each UE by using beamforming [34] as
Now turning to the DL, the massive number of antennas
at the BS are used for transmission and we can only count 𝑥𝑢,𝑣 𝑢,𝑣 𝑢
𝑘,𝑛 = 𝑏𝑘,𝑛 𝑥𝑘,𝑛 , 1≤𝑣 ≤𝑉, (10)
on a few antennas at the UE receivers, just one per UE in
1 ≤ 𝑢 ≤ 𝑈, 1 ≤ 𝑘 ≤ 𝐾, 1 ≤ 𝑛 ≤ 𝑁,
many cases. The mapping schemes and multi‑user con‑
𝑢,𝑣
stellations proposed for the UL are still valid, while we where 𝑏𝑘,𝑛 is the precoding coef icient for the 𝑢‑th UE and
need a few more ingredients to make these schemes suit‑ 𝑣‑th antenna of the BS, placed at the 𝑘‑th subcarrier and
able for the DL. We explain in the following these key in‑ 𝑛‑th OFDM symbol. This precoding coef icient is obtained
gredients for a combination of NCDS with beamforming according to the estimated angular positions of each UE,
based on [19], where the BS is simultaneously transmit‑ and thus, it is in charge of focusing the energy in the ob‑
ting the 𝑈 data streams through its 𝑉 antennas, while the tained speci ic directions. In this way, the path loss is
UEs are receiving with their single‑antenna device. compensated and the MUI that results from spatially mul‑
tiplexing the UEs in different beams is avoided. Similarly,
4.1 Beamforming beamforming can be used in the UL for the BS to receive
the signal coming from these spatial directions.
The exploitation of the diversity from the transmitter Depending on the angular position of different UEs, it
without the knowledge of the CSI is still a challenge. Due may be dif icult to completely remove the MUI by exploit‑
to the fact that techniques based on block codes [27] failed ing the beamforming. Therefore, the overall performance
to exploit the large number of antennas at the transmitter, critically depends on the scheduler which is capable of
we propose the use of beamforming in order to take ad‑ properly selecting those UEs to be simultaneously served
vantage of the massive number of antennas of the BS at in the same time and frequency resources and minimizing
the expense of using some (reduced) channel knowledge. the negative impact of the mentioned MUI. Even though
Then, it is assumed that the angular positions of each UE we are making use of some reference signals to perform

18 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

the beamforming, any other additional overhead used 5.1 Simulation parameters
in the CDS, such as the demodulation‑reference signals
(DM‑RS), is avoided, increasing the spectral ef iciency. To show some illustrative results, the numerology of the
OFDM signal is chosen according to 5G [1]. The carrier
spacing is Δ𝑓 = 30 KHz, which is the most frequent value
4.2 Frequency diversity in different bands [35]. The bandwidth of the system is
Due to the usually limited number of antennas at the UE, 𝐵𝑊 = 100 MHz and the carrier frequency is 𝑓𝑐 = 3.5
averaging in any dimension other than space (e.g. in time GHz. The BS is equipped with a uniform linear array
or frequency) is proposed in [19] to provide an additional (ULA) of 𝑉 = 128 antenna elements, which are simul‑
source of diversity. This diversity is needed in order to taneously serving two UEs (𝑈 = 2) in the whole band‑
obtain the required SINR gain for a good performance of width. Their angular separation corresponds to 72° and
the NCDS [12]. It is particularly needed if we want to mul‑ the path loss is not considered, since the power control is
tiplex several UEs in the constellation domain or enable assumed to work perfectly. We adopt a geometric chan‑
services that are critical in terms of performance. The use nel model, which corresponds to a spatially correlated
of the frequency dimension is described in [17], with the channel, where the power delay pro ile corresponds to
advantage that each OFDM symbol can be independently the Type B given in [35], the delay spread is 𝐷𝑆 = 16
processed. The proposed scheme can be easily extended ns and the angular spread is 𝐴𝑆 = 5°. We assume that
to averaging either in time (processing several consecu‑ there is a Doppler shift of 𝑓𝑑 = 1.6 KHz which corre‑
tive OFDM symbols) or space (increasing the number of sponds approximately to a speed of 𝑣 = 500 km/h at the
receive antennas of the UE, when possible). mentioned carrier frequency. We assume a perfect time‑
The way to leverage frequency diversity consists in trans‑ frequency synchronization and power control at the re‑
mitting the same differential complex symbol in several ceiver. For the sake of space we do not provide any re‑
frequency resources. At the transmitter, after performing sults for a higher carrier frequency. However, the cho‑
the differential encoding, the 𝑄 differential symbols are sen delay and angular spread can be also representative
repeated as of the propagation at higher frequencies, and the same
Doppler frequency would correspond to a smaller speed.
𝑥𝑢𝑘,𝑛 = 𝑥𝑢𝑞,𝑛 ∣ 𝑞 = mod (𝑘 − 1, 𝑄) + 1, 𝐾 = 𝑄 × 𝐹, Hence, the conclusions obtained with these numerical re‑
(11) sults, in particular those including beamforming (which
1 ≤ 𝑢 ≤ 𝑈, 1 ≤ 𝑘 ≤ 𝐾, 1 ≤ 𝑛 ≤ 𝑁, would be mandatory to compensate the path loss), can
be extrapolated to other higher frequency bands, such as
where 𝐹 is the frequency repetition/averaging factor. mm‑Wave [36]. The SNR is conventionally de ined as the
At the receiver, analogously to ((6)), the frequency diver‑ ratio of the received signal power over the noise power at
sity is exploited in the non‑coherent detection, where the each antenna of the receiver.
received data in the subcarriers that carry the same trans‑ For a baseline CDS system to compare the performance,
mitted data are averaged as we adopt the pilot con iguration speci ied for the demod‑
ulation reference signals (DM‑RS) in 5G [1]. In the time
1 𝐹 −1 𝑣 ∗
𝑣
domain, due to the high mobility, we set four reference
𝑧𝑞,𝑛 = ∑ (𝑦 ) 𝑦𝑞+𝑓𝑄,𝑛 , (12) symbols for each slot, which corresponds to the maximum
𝐹 𝑓=0 𝑞−1+𝑓𝑄,𝑛
pilot density allowed by the standard. In the frequency
domain, we assume the con iguration Type‑1, where each
2 ≤ 𝑞 ≤ 𝑄, 1 ≤ 𝑛 ≤ 𝑁. half of the subcarriers are allocated to each UE: the even
With this scheme there is a trade‑off between overhead subcarriers are used for the channel estimation of the UE1
and robustness. We will see that for some particular sce‑ and the odd subcarriers are for UE2 . At the receiver, the
narios with high mobility, even with the added overhead, channel estimation is irstly obtained at the pilot symbol
this scheme will outperform the CDS in terms of through‑ resources by applying least squares (LS) [37]. Then, these
put. estimates are interpolated to the entire resource grid by
using spline interpolation.
In Fig. 6 the block diagram of the system proposed in [19]
is shown, combining the beamforming with the NCDS. At Moreover, some hardware impairments are also consid‑
the receiver, assuming single‑antenna devices, only fre‑ ered, such as the effects of PN and the non‑linear HPA. The
quency diversity is exploited in order to reduce the noise effect of the PN is due to the instability of the local oscil‑
and MUI terms. lators, that can only be reduced by making a more expen‑
sive one. The PN is typically modelled according to a clas‑
sical Wienner random walk process given in [23], where
5. PERFORMANCE DISCUSSION the system performance is related to the phase noise in‑
In this section we illustrate the performance of the com‑ crement variance 𝜎𝜂2 over the sample period 𝑇 , where
bination of NCDS with a large number of antennas by dis‑ 𝜎𝜂2 = 2𝜋𝐵𝜂 𝑇 , with 𝐵𝜂 equal to the 3‑dB bandwidth of
cussing some numerical results. A comparison with some the Lorentzian power density carrier spectrum. The neg‑
CDS counterparts is also provided. ative effect of the PN not only will degrade the received

© International Telecommunication Union, 2020 19

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

DPSK Channel
UE1 Diff. OFDM Tx Diff. Time/Freq. UE1
Modu‑ UE1 OFDM Rx Decision
bits Encoding ant. 𝑣 = 1 Decoding Combiner bits
lation
DPSK Channel
UE2 Diff. OFDM Tx Diff. Time/Freq. UE2
Modu‑ Time/Freq. UE2 OFDM Rx Decision
bits Encoding ant. 𝑣 = 2 Decoding Combiner bits
lation Diversity
and
Precoding
⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮

DPSK OFDM Channel

UE𝑈 Diff. Diff. Time/Freq. UE𝑈
Modu‑ Tx ant. UE𝑈 OFDM Rx Decision
bits Encoding Decoding Combiner bits
lation 𝑣 = 𝑉

BS
Fig. 6 – Block diagram for DL, where the BS uses a beamforming technique and all UEs are single‑antenna devices.

symbols, but it will also add a common phase error [23]. fectively reduce its SER, while it is not enough for CDS to
The variance of the PN corresponds to 𝜎𝜂2 = 10−5 rad2 . work properly. When hardware impairments are consid‑
On other hand, the realistic transfer function of the HPA ered, the performance of NCDS is degraded by both HPA
is not a linear function for all possible input values. This and PN effects. In the same way as for the UL, we can see
implies that the output might be saturated for those in‑ that NCDS is very robust to the PN effects due to the dif‑
put values that are higher than the saturation point. This ferential modulation. However, the non‑linear HPA sig‑
non‑linear effect will not only degrade the quality of the ni icantly degrades its performance. In this case the BS
received signal, but it will also enhance the out‑of‑band is simultaneously transmitting the signals of the two UEs
emissions. According to [24], we consider a solid state and, consequently, the constant envelope characteristic
power ampli ier whose output back‑off (OBO) is OBO = 8 of each of the PSK signals is lost when they are combined.
dB. It turns out that now the OBO is not enough and some of
the signal peaks are clipped. This affects equally to both
5.2 Numerical results NCS and CDS. In Fig. 9, a comparison in terms of through‑
put is provided for the DL, whose expression is given by
In Fig. 7, we show the SER comparison between CDS 𝐵𝑊
and NCDS for the UL. The constellations of the two UEs 𝑇𝑟 = log2 (𝑀 ) (1 − 𝑆𝐸𝑅) . (13)
𝐹
are QPSK for CDS and EEP for NCDS, both using two bits
We can see that even with the overhead due to a very
per symbol. The CDS performs a post‑equalizer at the BS
high frequency averaging factor (𝐹 = 16), the NCDS
based on a ZF criterion. In the absence of PN and HPA, the
still outperforms the traditional CDS. This difference is
NCDS outperforms the traditional CDS by almost two or‑
even higher when either PN or HPA effects are considered.
ders of magnitude of SER for moderate and high SNR sce‑
Therefore, the throughput reduction due to the overhead
narios. When hardware impairments are considered, the
produced by the frequency diversity is negligible as com‑
PN and HPA effects do not signi icantly degrade the per‑
pared to the small throughput achieved by the CDS due to
formance of NCDS. On the other hand, the performance of
a poor performance obtained even with a large overhead.
CDS with and without the effect of the HPA is very poor,
and it is even worse with the PN. The PN does not affect
our proposed system due to the use of a differential mod‑ 6. CONCLUSION
ulation and the fact that the phase noise does not change We have provided a detailed description of the novel com‑
between two contiguous subcarriers [23]. The negative bination of NCDS and multi‑user MIMO‑OFDM based on a
effect of the HPA is negligible in both systems because the differential modulation scheme. Both DL and UL scenar‑
OBO is enough, in view of the robustness of the PSK sig‑ ios are considered and the performance is analyzed for
nals, which are ampli ied separately at the transmitter of realistic channel conditions including the effect of the PN
each UE. and HPA.
In Fig. 8, we plot the SER comparison between CDS and It is shown that for channels with high mobility, the NCDS
NCDS for the DL. The same beamforming is considered outperforms the traditional CDS obtaining a better per‑
for CDS and NCDS to spatially multiplex the two UEs. Ad‑ formance, even more noticeable when PN and non‑linear
ditionally, a frequency averaging of factor 𝐹 = 16 is HPA effects are taken into account. NCDS does not require
performed in both schemes to leverage diversity and im‑ any additional PN estimation and equalization since it is
prove the overall performance, which would be otherwise inherently robust to these effects. Moreover, it shows a
compromised. Again, in the absence of PN and HPA, the similar degradation with the non‑linear effects of the HPA
NCDS outperforms the CDS by several orders of magni‑ to that suffered by CDS, since they share the sensitivity of
tude, showing that the frequency averaging is able to ef‑ OFDM to these effects.

20 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

100
and ideas to improve the NCDS. In particular, the multi‑
plexing of UEs in the constellation domain is not ef icient,
so an excessively high number of antennas is currently
10-1 needed to multiplex more than two UEs [12]. There is a
need to ind new constellation designs that overcome this
limitation. A possible way to obtain these optimal designs
is to use some arti icial intelligence techniques in order to
SER

10-2
automatically deal with the channel and multi‑user inter‑
ference effects. Not only the phases, but also the constant
CDS - ZF amplitudes of the constellations are possible values to se‑
10-3 CDS - ZF, PN
CDS - ZF, HPA lect, giving a more complex search space, where arti icial
NCDS
NCDS, PN
intelligence can help.
NCDS, HPA Additionally, we have seen that the performance is very
10-4
0 2 4 6 8 10 12 14 16 18 20 sensitive to the spatial separation of the UEs that are
SNR (dB) multiplexed, either in constellation or space. Therefore,
scheduling algorithms that take this into account and op‑
Fig. 7 – SER comparison for UL. timize a particular performance metric are also crucial.
100 The advantages of NCDS with respect to CDS vanish when
the channel is quasi‑static and with high SNR. Then, it is
10-1
advisable to ind hybrid schemes, such as [20] where the
best of both paradigms is used according to the commu‑
10-2
nication scenario and needs.
Finally, it is foreseen that in the future communications
10-3
will be tightly integrated with sensing, which is one of the
SER

10-4
main objectives of the the Sixth Generation (6G) of mobile
communications [38]. In these new systems, the ef icient
10-5 CDS
exploitation of CSI under a variety of scenarios will play an
CDS, PN important role, and hence, we forecast that the exploita‑
CDS, HPA
10-6 NCDS tion of non‑coherent techniques will be an interesting al‑
NCDS, PN
NCDS, HPA
ternative, in order to increase the ef iciency of the overall
10-7 system. We hope that this review of the NCDS characteris‑
0 2 4 6 8 10 12 14 16 18 20
SNR (dB) tics, feasible implementation and performance will stim‑
ulate new research and advances in this topic.
Fig. 8 – SER comparison for DL.
7 ACKNOWLEDGEMENT
6 This work has been funded by project TERESA‑ADA
(TEC2017‑90093‑C3‑2‑R) (MINECO/AEI/FEDER, UE).
5 The authors would like to thank Ignasi Piqué ‑Muntané
for his help in the elaboration of some igures.
Throughput (Msps)

REFERENCES
3
[1] NR; Physical channels and modulation (Release 16).
CDS
2 CDS, PN Technical report, 3GPP, France, 2020.
CDS, HPA

1
NCDS
NCDS, PN
[2] W. Guo, W. Zhang, P. Mu, F. Gao, and H. Lin.
NCDS, HPA High‑mobility wideband massive MIMO communi‑
0
cations: Doppler compensation, analysis and scal‑
0 2 4 6 8 10 12 14 16 18 20 ing laws. IEEE Transactions on Wireless Communica‑
SNR (dB)
tions, 18(6):3177–3191, June 2019.
Fig. 9 – Throughput comparison for DL. [3] Y. Ge, W. Zhang, F. Gao, S. Zhang, and X. Ma.
The NCDS combined with multi‑user MIMO‑OFDM is Beamforming network optimization for reducing
shown to be a feasible and very attractive technique channel time variation in high‑mobility massive
which substantially improves the performance of the co‑ MIMO. IEEE Transactions on Communications,
herent systems, especially in challenging scenarios such 67(10):6781–6795, Oct. 2019.
as systems with realistic propagation channels and high [4] M. Gao, B. Ai, Y. Niu, W. WU, P. Yang, F. Lyu, and
mobility. However, there is still need for new research X. Shen. Ef icient hybrid beamforming with anti‑

© International Telecommunication Union, 2020 21

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

blockage design for high‑speed railway communica‑ [16] V. M. Baeza and A. G. Armada. Noncoherent mas‑
tions. IEEE Transactions on Vehicular Technology, sive MIMO. In Wiley 5G Ref: The Essential 5G Refer‑
2020. ence Online, chapter 10, pages 266–290. John Wiley
& Sons, Ltd., 2019.
[5] G. L. Stuber, J. R. Barry, S. W. McLaughlin, Ye Li, M. A.
Ingram, and T. G. Pratt. Broadband MIMO‑OFDM [17] K. Chen‑Hu and A. G. Armada. Non‑coherent mul‑
wireless communications. Proceedings of the IEEE, tiuser massive MIMO‑OFDM with differential modu‑
92(2):271–294, Feb. 2004. lation. In ICC 2019 ‑ 2019 IEEE International Confer‑
ence on Communications (ICC), pages 1–6, May 2019.
[6] B. Yang, Z. Yu, J. Lan, R. Zhang, J. Zhou, and
W. Hong. Digital beamforming‑based massive [18] S. Bucher, G. Yammine, R. F. H. Fischer, and C. Wald‑
MIMO transceiver for 5G millimeter‑wave commu‑ schmidt. A noncoherent massive MIMO system
nications. IEEE Transactions on Microwave Theory employing beamspace techniques. IEEE Trans‑
and Techniques, 66(7):3403–3418, July 2018. actions on Vehicular Technology, 68(11):11052–
11063, Nov. 2019.
[7] R. A. Smith. The relative advantages of coherent and
incoherent detectors: a study of their output noise [19] K. Chen‑Hu, Y. Liu, and A. G. Armada. Non‑coherent
spectra under various conditions. Proceedings of the massive MIMO‑OFDM down‑link based on differen‑
IEE ‑ Part III: Radio and Communication Engineering, tial modulation. IEEE Transactions on Vehicular
98(55):401–406, Sep. 1951. Technology, 2020. (In press).

[8] D. Middleton. Statistical theory of signal detection. [20] M. J. Lopez‑Morales, K. Chen‑Hu, and A. Garcia‑
Transactions of the IRE Professional Group on Infor‑ Armada. Differential data‑aided channel estimation
mation Theory, 3(3):26–51, March 1954. for up‑link massive SIMO‑OFDM. IEEE Open Journal
of the Communications Society, 1:976–989, 2020.
[9] M. L. Doelz, E. T. Heald, and D. L. Martin. Binary data
[21] F. Adachi. Adaptive differential detection for M‑
transmission techniques for linear systems. Pro‑
ary DPSK. IEE Proceedings ‑ Communications,
ceedings of the IRE, 45(5):656–661, May 1957.
143(1):21–28, Feb. 1996.
[10] A. Manolakos, M. Chowdhury, and A. J. Goldsmith.
[22] R. Corvaja and A. G. Armada. Phase noise degrada‑
CSI is not needed for optimal scaling in multiuser
tion in massive MIMO downlink with zero‑forcing
massive SIMO systems. In 2014 IEEE International
and maximum ratio transmission precoding. IEEE
Symposium on Information Theory, pages 3117–
Transactions on Vehicular Technology, 65(10):8052–
3121, June 2014.
8059, Oct. 2016.
[11] M. Chowdhury, A. Manolakos, and A. Goldsmith. [23] H. Ghozlan and G. Kramer. Models and information
Scaling laws for noncoherent energy‑based commu‑ rates for wiener phase noise channels. IEEE Trans‑
nications in the simo mac. IEEE Transactions on In‑ actions on Information Theory, 63(4):2376–2393,
formation Theory, 62(4):1980–1992, April 2016. April 2017.
[12] A. G. Armada and L. Hanzo. A non‑coherent [24] E. Costa, M. Midrio, and S. Pupolin. Impact of
multi‑user large scale SIMO system relaying on M‑ ampli ier nonlinearities on ofdm transmission sys‑
ary DPSK. In 2015 IEEE International Conference tem performance. IEEE Communications Letters,
on Communications (ICC), pages 2517–2522, June 3(2):37–39, Feb. 1999.
2015.
[25] L. Lampe, R. Schober, and M. Jain. Noncoherent
[13] V. M. Baeza, A. G. Armada, M. El‑Hajjar, and L. Hanzo. sequence detection receiver for Bluetooth systems.
Performance of a non‑coherent massive SIMO M‑ IEEE Journal on Selected Areas in Communications,
DPSK system. In 2017 IEEE 86th Vehicular Technol‑ 23(9):1718–1727, Sep. 2005.
ogy Conference (VTC‑Fall), pages 1–5, Sep. 2017.
[26] C. Wang, C. Huang, J. Huang, C. Chang, and C. Li.
[14] V. M. Baeza, A. G. Armada, W. Zhang, M. El‑Hajjar, Zigbee 868/915‑mhz modulator/demodulator for
and L. Hanzo. A non‑coherent multiuser large‑ wireless personal area network. IEEE Transac‑
scale SIMO system relying on M‑ary DPSK and BICM‑ tions on Very Large Scale Integration (VLSI) Systems,
ID. IEEE Transactions on Vehicular Technology, 16(7):936–939, July 2008.
67(2):1809–1814, Feb. 2018.
[27] J. Cabrejas, S. Roger, D. Calabuig, Y. M. M. Fouad,
[15] V. M. Baeza and A. G. Armada. Non‑coherent mas‑ R. H. Gohary, J. F. Monserrat, and H. Yanikomeroglu.
sive SIMO system based on M‑DPSK for Rician chan‑ Non‑coherent open‑loop MIMO communications
nels. IEEE Transactions on Vehicular Technology, over temporally‑correlated channels. IEEE Access,
68(3):2413–2426, March 2019. 4:6161–6170, 2016.

22 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[28] B. M. Hochwald and W. Sweldens. Differential uni‑ AUTHORS

tary space‑time modulation. IEEE Transactions on
Communications, 48(12):2041–2052, Dec. 2000.
[29] V. Tarokh and H. Jafarkhani. A differential detection
scheme for transmit diversity. IEEE Journal on Se‑
lected Areas in Communications, 18(7):1169–1174,
July 2000.
[30] M. Beko, J. Xavier, and V. A. N. Barroso. Noncoherent
communication in multiple‑antenna systems: Re‑
ceiver design and codebook construction. IEEE
Transactions on Signal Processing, 55(12):5703–
5715, Dec. 2007.
[31] R. H. Gohary and T. N. Davidson. Noncoherent MIMO
communication: Grassmannian constellations and Kun Chen‑Hu received his Ph.D. degree in Multimedia
ef icient detection. IEEE Transactions on Informa‑ and Communications in 2019 from Universidad Carlos
tion Theory, 55(3):1176–1205, March 2009. III de Madrid (Spain). Currently, he is a post‑doctoral
researcher in the same institution. He was awarded
[32] F. Adachi. Reduced‑state Viterbi differential detec‑ by UC3M in 2019 in recognition of his outstanding pro‑
tion using a recursively estimated phase reference fessional career after graduation. He visited Eurecom
for M‑ary DPSK. IEE Proceedings ‑ Communications, (France) and Vodafone Chair TU Dresden (Germany),
142(4):263–270, Aug. 1995. both as guest researcher. He also participated in differ‑
ent research projects in collaboration with several top
[33] N. Prasad, S. Wang, and X. Wang. Ef icient re‑ companies in the area of mobile communications. His
ceiver algorithms for DFT‑spread OFDM systems. research interests are related to signal processing tech‑
IEEE Transactions on Wireless Communications, niques, such as waveforms design, non‑coherent massive
8(6):3216–3225, June 2009. MIMO and channel estimation.
[34] J. Lota, S. Sun, T. S. Rappaport, and A. Demos‑
thenous. 5G uniform linear arrays with beamform‑
ing and spatial multiplexing at 28, 37, 64, and 71 ghz
for outdoor urban communication: A two‑level ap‑
proach. IEEE Transactions on Vehicular Technology,
66(11):9972–9985, Nov. 2017.
[35] Study on channel model for frequencies from 0.5
to 100 GHz (Release 16). Technical report, 3GPP,
France, 2019.
[36] T. S. Rappaport, R. W. Heath, R. C. Daniels, and J. N.
Murdock. Millimeter wave wireless communications.
Prentice Hall, 2015. Yong Liu received a Ph.D in electronic engineering from
the Department of Electric Engineering, Shanghai Jiao
[37] J. Lin. Least‑squares channel estimation for mobile Tong University, Shanghai, China, in 2012. He is now
ofdm communication on time‑varying frequency‑ with the wireless network RAN research department in
selective fading channels. IEEE Transactions on Ve‑ Huawei Technologies Co.,Ltd., Shanghai, China. His cur‑
hicular Technology, 57(6):3538–3550, Nov. 2008. rent research interests lie in the area of 5G and 5G+ MIMO
communication and AI assisted wireless networks..
[38] K. B. Letaief, W. Chen, Y. Shi, J. Zhang, and Y. A. Zhang.
The roadmap to 6G: AI empowered wireless net‑
works. IEEE Communications Magazine, 57(8):84–
90, Aug. 2019.

© International Telecommunication Union, 2020 23

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

nical committee in 2019. Her main interests are multi‑

carrier and multi‑antenna techniques and signal process‑
ing applied to wireless communications.

Ana García Armada (S’96‑A’98‑M’00‑SM’08) received a

Ph.D. degree in electrical engineering from the Polytech‑
nical University of Madrid in February 1998. She is cur‑
rently a Professor at University Carlos III of Madrid, Spain.
She is leading the Communications Research Group at this
university. She has been a visiting scholar at Stanford
University, Bell Labs and University of Southampton. She
has participated (and coordinated most of them) in more
than 30 national and 10 international research projects,
as well as 20 contracts with the industry, all of them re‑
lated to wireless communications. She is the co‑author
of eight book chapters on wireless communications and
signal processing. She has published around 150 papers
in international journals and conference proceedings and
she holds four patents. She has contributed to interna‑
tional standards organizations, such as ITU and ETSI, and
is a member of the expert group of the European 5G PPP
and member of the advisory committee 5JAC of the ESA as
an expert appointed by Spain on 5G. She has served on the
editorial boards of Physical Communication (2008‑2017),
IET Communications (2014‑2017). She has been serving
on the editorial board of IEEE Communications Letters
since 2016 (Editor until Feb 2019, Senior Editor from Mar
2019, Exemplary Editor Award 2017 and 2018) and IEEE
Transactions on Communications since 2019. She has
served on the TPC of more than 40 conferences and she
has been a member of the organizing committee of IEEE
Globecom 2019, IEEE Vehicular Technology Conference
(VTC) Fall 2018, Spring 2018 and 2019 and IEEE 5G Sum‑
mit 2017, among others. She will be the General Chair
of Globecom 2021. She was the Newsletter Editor of the
IEEE ComSoc Signal Processing and Consumer Electron‑
ics Committee (2017‑2018) and is now the Secretary of
this committee (since 2019). She has been the Secretary
of the IEEE ComSoc Women in Communications Engineer‑
ing Standing Committee (2016‑2017) and the Chair of this
committee (2018‑2019). She has received the Young Re‑
searchers Excellence Award, the Award to Outstanding
achievement in research, teaching and management and
the Award to Best Practices in Teaching, all from Univer‑
sity Carlos III of Madrid, She was awarded the third place
Bell Labs Prize 2014 for shaping the future of informa‑
tion and communications technology. She received the
Outstanding service award from the IEEE ComSoc Signal
Processing and Communications Electronics (SPCE) tech‑

24 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

MSICA: MULTI-SCALE SIGNAL DECOMPOSITION BASED ON INDEPENDENT

COMPONENT ANALYSIS WITH APPLICATION TO DENOISING AND
RELIABLE MULTI-CHANNEL TRANSMISSION

Abolfazl Hajisami and Dario Pompili

Dept. of Electrical and Computer Engineering, Rutgers University–New Brunswick, NJ, USA

Abstract – Multi-scale decomposition is a signal description method in which the signal is decomposed into multiple
scales, which has been shown to be a valuable method in information preservation. Much focus on multi-scale decom-
position has been based on scale-space theory and wavelet transform. In this article, a new powerful method to perform
multi-scale decomposition exploiting Independent Component Analysis (ICA), called MSICA, is proposed to translate
an original signal into multiple statistically independent scales. It is proven that extracting the independent components
of the even and odd samples of a digital signal results in the decomposition of the same into approximation and detail.
It is also proven that the whitening procedure in ICA is equivalent to a filter bank structure. Performance results of
MSICA in signal denoising are presented; also, the statistical independency of the approximation and detail is exploited
to propose a novel signal-denoising strategy for multi-channel noisy transmissions aimed at improving communication
reliability by exploiting channel diversity.
Keywords – Channel Diversity, Independent Component Analysis, Multi-scale Decomposition, Wavelet Transform.

1. INTRODUCTION passing the signal through a low-pass filter and high-

pass filter, respectively, followed by a downsampling by
Overview: Multi-scale decomposition is an invaluable
a factor of 2. This results in a decomposition of the
tool in digital signal processing with applications such
signal into different scales, which can be considered as
as those in [1, 2, 3, 4, 5], where an original signal is de-
low and high frequency bands. Multi-scale decomposi-
composed into a set of signals, each of which provides
tion by wavelet transforms has a number of advantages
information about the original signal at a different scale.
over the scale-space decomposition and empirical mode
A major signal-processing task where multi-scale decom-
decomposition [20]. First, since the signal information
position has been shown to be very useful is denoising,
at one scale is not contained in another scale, signal in-
based on the intuition that information pertaining to the
formation at different scales are better separated in the
noise would be accurately characterized in certain scales
wavelet domain. Second, scale selection when perform-
that are separate from the scales of the signal. The
ing noise suppression using wavelet decomposition is less
main literature works in multi-scale decomposition have
critical than that for noise suppression using scale-space
focused on scale-space decomposition [6, 7, 8, 9, 10],
decomposition as all the scales are considered in noise
empirical mode decomposition [11, 12, 13], and wavelet
suppression using wavelet decomposition as opposed to a
transform [14, 15, 16, 17, 18].
single scale as done in scale-space decomposition. How-
In scale-space theory [19], a signal is decomposed into a
ever, there are a number of limitations pertaining to
single-parameter family of 𝑛 signals with a progressive
noise suppression using wavelet transform [20]. For in-
decrease in fine scale signal information between suc-
stance, signals processed using wavelet transforms can
cessive scales. This allows analyzing signals at coarser
exhibit oscillation artifacts related to wavelet basis func-
scales without the influence of finer scales such as those
tions used in the wavelet transform, which is particularly
pertaining to noise. Knowing this, one can employ scale-
noticeable in low Signal-to-Noise Ratio (SNR) regimes.
space theory to suppress noise by performing scale-space
Moreover, in DWT the approximation and detail are
decomposition on the signal and then treating one of
not statistically independent, which may cause a poor
signals at a coarser scale as the noise-suppressed sig-
performance in signal denoising.
nal. However, selecting the scale that represents the
noise-suppressed signal can be challenging. Moreover, Motivation and Approach: Given these limitations
noise suppression using scale-space theory does not fa- of both space-scale and wavelet decomposition in terms
cilitate the fine-grained noise suppression at the individ- of signal denoising, we were motivated to explore al-
ual scales, which limits its overall flexibility in striking a ternative approaches. We investigate the problem of
balance between noise suppression and signal structural decomposing a signal into multiple scales from a differ-
preservation [20]. ent point of view, i.e., we propose a new approach that
In Discrete Wavelet Transform (DWT), the original sig- takes a statistical perspective on multi-scale decomposi-
nal is decomposed into approximation and detail by tion according to which a signal is considered as a mix-

© International Telecommunication Union, 2020 25

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

ture of statistically independent signals, each character- 2. BLIND SOURCE SEPARATION

izing signal information at a different scale. Having this
In BSS, a set of mixtures of different source signals is
perspective in multi-scale decomposition can be benefi-
available and the goal is to separate the source signals
cial in signal denoising for two reasons. First, since most
when we have no information about the mixing system
of the signal information in one scale is not included in
or the source signals (hence the name “blind”) [26, 27].
the other scales, such decomposition provides the ad-
vantage of noise suppression at the individual scales in 𝑠1 - - 𝑥1 - - 𝑦1
order to trade off noise suppression for signal-quality 𝑠2 - - 𝑥2 - - 𝑦2
preservation. Second, since the noise signal is statisti-
�
A � �
B �
cally independent from the original signal, by decompos-
ing the noisy signal into statistically independent scales, 𝑠𝑁 - -𝑥𝑁 - - 𝑦𝑁
the noise is expected to be separated in finer scales. Mixing System Separating System
Our Contribution: Given this motivation and per- Fig. 1 – Mixing and separating systems in Blind Source Separa-
spective, we propose a new method for Multi-Scale de- tion (BSS).
composition exploiting Independent Component Anal- As in Fig. 1, the mixing and separating systems can be
ysis (ICA), called MSICA, in which the original digital represented as,
signal is decomposed into approximation and detail with
statistically independent components. Specifically, we x(𝑛) = As(𝑛),
extract two correlated signals from the original signal (1)
y(𝑛) = Bx(𝑛),
and apply a linear transformation to the extracted sig-
nals so as to decompose the original signal into multiple where s(𝑛) = [𝑠1 (𝑛), … , 𝑠𝑁 (𝑛)]𝑇 is the vector of sources
scales. Since we need a suitable transform to decompose that are mixed by the mixing matrix A and create the
the original signal into statistically independent compo- observations vector x(𝑛) = [𝑥1 (𝑛), … , 𝑥𝑁 (𝑛)]𝑇 . Let A
nents, we consider our problem as a Blind Source Sepa- be a square matrix (𝑁 × 𝑁 ) of full column rank, which
ration (BSS) problem in which the extracted signals are means that the number of sources is equal to the num-
considered as the observations of the source separation ber of observations and that the sources are linearly in-
problem. To relate this problem with the concept of dependent. The goal of BSS is to find the separating
multi-scale decomposition, we introduce an equivalent matrix B such that y(𝑛) = [𝑦1 (𝑛), … , 𝑦𝑁 (𝑛)]𝑇 is an es-
filter-bank structure for the proposed method, which is timation of the sources.
similar to the structure introduced in [14] for the DWT A method to solve the BSS problem is via ICA, which
implementation. We also propose a method for multi- exploits the assumption of source independence and es-
channel transmission in which MSICA outperforms com- timates B such that the outputs 𝑦𝑖 ’s be statistically in-
mon wavelet transforms in denoising of the received sig- dependent [28]. As studied in [28, 29], the ICA can be
nal. We show that if the even and odd samples of performed by two steps: 1) whitening (or decorrelating)
the original signal are transmitted through two Addi- and 2) rotation. To illustrate the ICA model, we con-
tive White Gaussian Noise (AWGN) channels, MSICA sider two independent components, 𝑠𝑖 , 𝑖 = 1, 2, with a
is able to extract and filter out the noise from the noisier uniform distribution,
channel. This key property of MSICA—which exploits
channel diversity and generalizes to the case in which 1 if |𝑠𝑖 | ≤ 0.5,
𝑝(𝑠𝑖 ) = { (2)
more than two channels are available—can be used to 0 otherwise,
increase the transmission efficiency in noisy communi-
where the joint density of 𝑠1 and 𝑠2 is uniform on a
cation channels, although the marginal return dimin-
square, as illustrated in Fig. 2(a). This follows from
ishes as the number of channels increases. It should be
the definition that the joint density of two independent
noted that, although single-channel ICA has been used
variables is the product of their marginal densities. Let
in previous works [21, 22, 23, 24, 25] (including the spa-
us now mix these two independent components by the
tial case of using even and odd samples), in this work
following mixing matrix A,
single-channel ICA has been studied as a technique for
signal decomposition into statistically independent ap- 𝑥1 1 2 𝑠
proximation and detail and its performance in denoising [ ]= [ ][ 1 ], (3)
𝑥2 ⏟⏟⏟⏟⏟ 𝑠2
2 1
has been compared with other wavelet transforms.
A
Article Organization: In Section 2, we provide some
background on BSS and ICA. In Section 3, we propose where the mixed variables 𝑥1 and 𝑥2 have a uniform
our ICA-based transform for multi-scale decomposition. distribution on a parallelogram, as shown in Fig. 2(b).
In Section 4, we examine the performance of MSICA in Note that 𝑥1 and 𝑥2 are not independent anymore. To
signal denoising and show how to increase transmission show this consider whether it is possible to predict the
efficiency when multiple (noisy) channels are available. value of one of them, say 𝑥2 , from the value of the other;
Finally, in Section 5, we draw the main conclusions and it is clear that if 𝑥1 attains one of its maximum or min-
wrap up the article by discussing future work. imum values, then this completely determines the value

26 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

0.5 1.5 0.8 0.5

y2
0
s2

x2

z2
0 0 0

-0.5 -1.5 -0.8 -0.5

-0.5 0 0.5 -1.5 0 1.5 -2.5 0 2.5 -1 0 1
s1 x1 z1 y1

(a) (b) (c) (d)

Fig. 2 – Illustration of Independent Component Analysis (ICA) algorithm. (a) Joint distribution of the independent components 𝑠1 and
𝑠2 with uniform distribution; (b) Joint distribution of the observed mixtures 𝑥1 and 𝑥2 ; (c) Joint distribution of the whitened mixtures,
𝑧1 and 𝑧2 ; (d) Joint distribution of the independent output components, 𝑦1 and 𝑦2 , as determined by the ICA.

of 𝑥2 . However, the situation for variables 𝑠1 and 𝑠2 that decomposing the neighboring samples of the sig-
is different: from Fig. 2(a) it is clear that knowing the nal into their independent components would decom-
value of 𝑠1 does not help predict the value of 𝑠2 . pose the signal into its approximation and detail. Given
The problem of source separation is now to estimate the this motivation, we propose an ICA-based method for
mixing matrix A and multiply its inverse (B = A−1 ) to multi-scale decomposition in which the approximation
the vector of the mixtures to retrieve the sources (𝑠1 and and details are statistically independent. Our algorithm
𝑠2 ). As studied in [28, 29], the ICA can be considered as consists of two steps: 1) extracting the observation sig-
a two-step procedure where, in the first step, the mixed nals from the original signal and 2) decomposing the
data gets whitened (uncorrelated) by multiplying the original signal into approximation and detail by apply-
whitening matrix by the vector of mixtures, i.e., ing a linear transformation to the observation signals.
Suppose that 𝑥(𝑛) is a Wide Sense Stationary (WSS)
[
𝑧1 𝑥
] = W[ 1 ]; (4) signal. We consider even and odd samples of 𝑥(𝑛) as
𝑧2 𝑥2 𝑥1 (𝑛) and 𝑥2 (𝑛), respectively, i.e.,

and then, in the second step, the independent compo- 𝑥1 (𝑛) = 𝑥(2𝑛), 𝑥2 (𝑛) = 𝑥(2𝑛 − 1). (7)
nents are separated by applying a rotation matrix, i.e.,
We prove that, if 𝑥1 (𝑛) and 𝑥2 (𝑛) are the observations
𝑦 𝑧 signals (mixtures) of the ICA, the outputs of the ICA
[ 1 ] = R[ 1 ]. (5)
𝑦2 𝑧2 will be the approximation and detail of 𝑥(𝑛), which
are statistically independent. We define the observation
Fig. 2(c) illustrates the effect of whitening in which the vector x as,
data in Fig. 2(b) has been whitened. Also, Fig. 2(d)
shows how rotating the whitened data can return the x=[
𝑥1 (𝑛)
]=[
𝑥(2𝑛)
], (8)
statistical independency to the outputs and recover the 𝑥2 (𝑛) 𝑥(2𝑛 − 1)
independent components. Therefore, the separation ma-
trix B can be considered as the product of the whitening where the even and odd samples of the original signal
and rotation matrices, i.e., are the first and second rows of the observation vector,
respectively. If we apply the ICA to the observation
B = RW, (6) vector, the output is,

where W is the whitening matrix and R is the rotation 𝑦1 (𝑛)

y = Bx = [ ], (9)
matrix. Note that in the case whitened components are 𝑦2 (𝑛)
statistically independent, the rotation matrix R will be where B is the separation matrix estimated by the ICA,
the identity matrix and no rotation will be needed. and 𝑦1 (𝑛) and 𝑦2 (𝑛) are statistically independent.
To prove our claim, we need to show that 𝑦1 (𝑛) and
3. MSICA: MULTI-SCALE DECOM- 𝑦2 (𝑛) are the approximation and detail of the original
POSITION BY INDEPENDENT signal. As explained in Section 2, ICA involves two
COMPONENT ANALYSIS steps: 1) whitening (or decorrelating) and 2) rotation,
and the separation matrix B can be considered as the
Generally, neighboring/consecutive samples of a com- product of the whitening and rotation matrices (i.e.,
mon signal are highly correlated and differ slightly from B = RW). Now, let us consider the whitened vector z
each other. This slight difference of neighboring samples and the whitening matrix W as follows,
comes from the details of the signal. If we consider the
detail scale of the original signal to be statistically in- 𝑧1 (𝑛) 𝑤 𝑤12
dependent from the approximation scale, it is expected z=[ ] = Wx, W = [ 11 ]. (10)
𝑧2 (𝑛) 𝑤21 𝑤22

© International Telecommunication Union, 2020 27

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Since D is a diagonal matrix, we conclude that multi-

plying Q𝑇 by the observation vector x results in uncor-
related outputs. Hence, the whitening matrix W can be
considered to be equal to Q𝑇 .
Now, we need to find the elements of matrix W so as to
determine finally the 𝑔(𝑛) and ℎ(𝑛) filters. Since 𝑥(𝑛)
is a WSS signal, we have,
Fig. 3 – Whitening process in ICA as a filter-bank structure.
𝑅𝑥 (0) = 𝐸 {𝑥2 (2𝑛)} = 𝐸 {𝑥2 (2𝑛 + 1)} = 𝜎𝑥2 ,
Using (8) and (10), we can write, (19)
𝑅𝑥 (1) = 𝐸 {𝑥 (2𝑛) 𝑥 (2𝑛 − 1)} = 𝜎𝑥2 𝜌.

𝑧1 (𝑛) = 𝑤11 𝑥1 (𝑛) + 𝑤12 𝑥2 (𝑛) , Hence, with regard to (8), we can recast C𝑥 as,
(11)
𝑧2 (𝑛) = 𝑤21 𝑥1 (𝑛) + 𝑤22 𝑥2 (𝑛) .
𝜎𝑥2 𝜎𝑥2 𝜌 1 𝜌
C𝑥 = 𝐸 {xx𝑇 } = [ ] = 𝜎𝑥2 [ ], (20)
Also, 𝑥1 (𝑛) and 𝑥2 (𝑛) can be written as, 𝜎𝑥2 𝜌 𝜎𝑥2 𝜌 1

𝑥1 (𝑛) = 𝑥(2𝑛) = [𝑥(𝑛) ∗ 𝛿(𝑛)] ↓ 2, where the eigenvalues and eigenvectors of C𝑥 are,
𝑥2 (𝑛) = 𝑥(2𝑛 − 1) = [𝑥(𝑛) ∗ 𝛿(𝑛 − 1)] ↓ 2,
𝜆1 = 𝜎𝑥2 (1 + 𝜌) 𝜆2 = 𝜎𝑥2 (1 − 𝜌) ,
(12)
√1 − √1 (21)
where ∗ and ↓ 2 denote convolution and downsampling q1 = [ 2 ] q2 = [ √1 2 ] .
√1
by a factor of 2, respectively, and 𝛿(𝑛) is the unit impulse 2 2
signal. Using (11) and (12), we can rewrite 𝑧1 (𝑛) and
𝑧2 (𝑛) as, Hence, the whitening matrix W can be presented as,

√1 √1
𝑧1 (𝑛) = 𝑤11 𝑥(2𝑛) + 𝑤12 𝑥 (2𝑛 − 1) W = Q𝑇 = [ q1
𝑇
q2 ] = [ 2 2 ]. (22)
= 𝑤11 [𝑥(𝑛) ∗ 𝛿(𝑛)] ↓ 2 + 𝑤12 [𝑥(𝑛) ∗ 𝛿(𝑛 − 1)] ↓ 2 − √12 √1
2
= [𝑥(𝑛) ∗ (𝑤11 𝛿(𝑛) + 𝑤12 𝛿(𝑛 − 1))] ↓ 2,
𝑧2 (𝑛) = 𝑤21 𝑥(2𝑛) + 𝑤22 𝑥 (2𝑛 − 1) Comparing (14) and (22), 𝑔(𝑛) and ℎ(𝑛) can be written
= 𝑤21 [𝑥(𝑛) ∗ 𝛿(𝑛)] ↓ 2 + 𝑤22 [𝑥(𝑛) ∗ 𝛿(𝑛 − 1)] ↓ 2 as,
= [𝑥(𝑛) ∗ (𝑤21 𝛿(𝑛) + 𝑤22 𝛿(𝑛 − 1))] ↓ 2. 𝑔(𝑛) = √12 𝛿(𝑛) + √12 𝛿(𝑛 − 1),
(23)
(13) ℎ(𝑛) = − √12 𝛿(𝑛) + √12 𝛿(𝑛 − 1).
Note that, if we consider the low-pass filter 𝑔(𝑛) and
high-pass filter ℎ(𝑛) as, From (23), it is clear that 𝑔(𝑛) and ℎ(𝑛) are a low and
high-pass filter, respectively, as we wanted to demon-
𝑔(𝑛) = 𝑤11 𝛿(𝑛) + 𝑤12 𝛿(𝑛 − 1), strate. Hence, we can conclude that the whitening pro-
(14) cess in the ICA (presented as a filter-bank structure in
ℎ(𝑛) = 𝑤21 𝛿(𝑛) + 𝑤22 𝛿(𝑛 − 1),
Fig. 3) decomposes the observation signals into uncorre-
(13) can be rewritten in a simpler form as, lated approximation and detail. Moreover, the rotation
process in the ICA makes sure that the approximation
𝑧1 (𝑛) = [𝑥(𝑛) ∗ 𝑔(𝑛)] ↓ 2, and the detail are statistically independent1 . Hence,
(15)
𝑧2 (𝑛) = [𝑥(𝑛) ∗ ℎ(𝑛)] ↓ 2. if the even and odd samples of a one-dimensional sig-
nal are considered as the observations of the ICA, we
From (15), we observe that the whitening process can be ensure the approximation and detail to be statistically
modeled as a filter-bank structure, as shown in Fig. 3. independent. As we will see in the next section, the sta-
Now, we need to show that 𝑔(𝑛) and ℎ(𝑛) are indeed tistical independency between the approximation and
low-pass and high-pass filters. In order to do so, we detail can be very beneficial in signal denoising, espe-
consider the covariance matrix of x as follows, cially when the even and odd samples are transmitted
through different (noisy) channels.
C𝑥 = 𝐸{xx𝑇 } = QDQ𝑇 , (16) Fig. 4 showcases a signal decomposition by different
wavelet transform and MSICA, where a Piece-Regular
where Q is an orthogonal matrix of eigenvectors and D signal is decomposed into the approximation and detail.
is a diagonal matrix of eigenvalues. Interestingly, the As shown in Fig. 4(h), like all the other wavelet trans-
covariance matrix of z = Q𝑇 x can be written as, forms, MSICA is also able to decompose the original
signal into approximation and detail, where the approx-
C𝑧 = 𝐸 {zz𝑇 } = 𝐸 {Q𝑇 xx𝑇 Q} = Q𝑇 𝐸 {xx𝑇 } Q. (17) imation and detail coefficients contain the low and high-
frequency components, respectively. As said earlier, to
Given (16) and (17), we can write,
1 Based on our simulations, the separation matrix was always close
C𝑧 = Q
⏟𝑇 Q D Q𝑇 Q = D.
⏟ (18) to (22), which means that the rotation matrix was always close
to the identity matrix.
I I

28 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

2 3 3 3

1.8
2 2 2
1.6
1 1 1
1.4

1.2 0 0 0
Amplitude

0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
1 Approximation Approximation Approximation
0.8 0.4 1 0.4

0.6 0.2 0.5 0.2

0.4 0 0 0

0.2 -0.2 -0.5 -0.2

0 -0.4 -1 -0.4
0 500 1000 1500 2000 2500 3000 3500 4000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Original Signal Detail Detail Detail

(a) (b) (c) (d)

3
3 3 3

2 2 2 2

1 1 1 1

0 0 0 0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Approximation Approximation Approximation Approximation
0.4
0.4 1 1

0.2 0.5 0.2 0.5

0 0 0 0

-0.2 -0.5 -0.2 -0.5

-0.4 -1 -0.4 -1
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Detail Detail Detail Detail

(e) (f) (g) (h)

Fig. 4 – Comparison of MSICA with different wavelet transforms in decomposing a Piece-Regular signal. (a) Original Signal. Approxi-
mation and detail by (b) Daubechies 3 wavelet; (c) Haar wavelet; (d) Biorthogonal 2.2 wavelet; (e) Coiflets 4 wavelet; (f) Fejer-Korovkin
4 wavelet; (g) discrete Meyer wavelet; and (h) our proposed MSICA.

2
4. MSICA FOR SIGNAL DENOISING
1.8
We discuss now how MSICA can be beneficial in sig-
nal denoising. Specifically, we compare MSICA with
1.6

1.4
the other wavelet transforms and show how MSICA can
1.2 suppress the noise via a simple wavelet thresholding.
Amplitude

1 We also show that, in the case of multi-channel trans-

mission, MSICA outperforms the other wavelet trans-
forms and is able to extract and filter out the noise of
0.8

0.6
the noisier communication channel by exploiting chan-
0.4 nel diversity.
0.2 Let us assume that the original signal is passed through
0
an AWGN channel, the noisy output signal is then,
0 500 1000 1500 2000 2500 3000 3500 4000
Reconstructed Signal 𝑟 (𝑛) = 𝑥 (𝑛) + 𝑤 (𝑛) , 𝑛 = 1, … , 𝑁 , (26)
Fig. 5 – Reconstructed signal by MSICA using the approximation
and detail in Fig. 4(h).
where 𝑥(𝑛) is the original signal and 𝑤(𝑛) is the AWGN
with zero mean and variance of 𝜎𝑤 2
. The goal of sig-
reconstruct the original signal we need to multiply the nal denoising is to remove the noise and obtain an es-
inverse of the separation matrix (B−1 ) by the vector of timate 𝑥(𝑛)
̂ of 𝑥(𝑛) that minimizes the Mean Squared
approximation and detail y so as to obtain the obser- Error (MSE), defined as,
vation vector x and reconstruct the original signal 𝑥(𝑛)
from x, i.e., 1 𝑁
MSE (𝑥)̂ = ∑ (𝑥(𝑛)
̂ − 𝑥(𝑛))2 . (27)
𝑁 𝑛=1
𝑥1 (𝑛)
[ ] = x = B−1 y, (24) Note that the model presented in (26) is not general
𝑥2 (𝑛)
since the noise may be non-additive, and the relation
between the noisy observed signal and the original signal
𝑥(𝑛) = (𝑥1 (𝑛) ↑ 2) + (𝑥2 (𝑛) ↑ 2) ∗ 𝛿(𝑛 + 1), (25) may be stochastic. Nevertheless, (26) is a widely used
model in many practical situations as it serves well as
where 𝑥𝑖 (𝑛) ↑ 2, 𝑖 = 1, 2, denotes the upsampling of a motivating example to give a good sense as to what
𝑥𝑖 (𝑛) by a factor of 2. Fig. 5 shows the reconstructed sig- happens in more realistic channels.
nal using (24) and (25). This figure shows that MSICA Let us emphasize that there are many existing ap-
can successfully reconstruct the original signal from the proaches in the literature to perform signal denoising,
approximation and detail obtained in the decomposition which can be roughly divided into two main categories:
procedure. 1) denoising in the original signal domain (e.g., in time

© International Telecommunication Union, 2020 29

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

or space [30]) and 2) denoising in the transform domain

(e.g., using Fourier or a wavelet transform [31, 32]).
0.07

Since the wavelet transform provides information in 0.06

both the time and frequency domain, and the infor-
Daubechies 3
Haar
mation in one scale is not contained in another scale, 0.05
Biorthogonal 2.2

approaches in this second category can strike a balance

Coiflets 4
Fejer-Korovkin 4

between noise suppression and signal structural preser- 0.04

MSICA

vation, and have therefore been used widely in signal de-

MSE
noising. It is interesting to note that, usually, the detail 0.03

coefficients of a noiseless signal are sparse. This means

that in the wavelet transform most of the detail coeffi- 0.02

cients of a noiseless signal are very small/close to zero

(as can be seen, for instance, in Fig. 4). So, the detail 0.01

coefficients with a small magnitude can be considered as

a noise component and can be set to zero. This is the
0
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
basic idea of wavelet thresholding approaches, which are w

employed in signal denoising where the coefficients are

Fig. 6 – MSICA Mean Square Error (MSE) compared to well-
compared with a threshold to determine whether they known wavelet transforms.
contain only noise or not.
It should be noted that since the approximation coeffi- Comparing (29) and (20), it is clear that the whitening
cients contain the low-frequency/important information matrix for the vector of noisy observations r is like the
of the signal, the thresholding is usually applied only one in (22),
to the detail coefficients (high-frequency components).
√1 √1
The thresholding methods retain the significant compo- W=[ 2 2 ]. (31)
nents by setting to zero only the detail coefficients whose − √12 √1
2
absolute values are less than a certain threshold. Most
of the thresholding approaches try to find the optimal Fig. 6 shows the performance of MSICA in the first sce-
threshold value. The SureShrink method [33], proposed nario in terms of MSE against well-known wavelet trans-
by Donoho and Johnstone, is an effective wavelet thresh- forms. As expected, MSICA shows similar performance
olding method for signal denoising that fits a level- compared to the other wavelet transform in suppressing
dependent threshold to the wavelet coefficient. To show the noise level and enhancing the quality of the signal
the performance of MSICA with respect to the other as it is able to decompose the signal into approxima-
wavelet transforms, we extract the first and second- tion and detail. However, in the following we will ex-
level detail coefficients of the noisy signal and use the plain how MSICA can have extraordinary performance
SureShrink method for signal denoising. In our exper- when the odd and even samples of the original signal
iments, we also employ the FastICA method [34] and are passed through different channels.
use the different standard test signals given in [35], i.e., Multi-channel Transmission: Let us consider now
Piece-Regular, Blocks, Doppler, and HeaviSine. the second scenario where we investigate the perfor-
We consider now two scenarios for signal transmis- mance of MSICA in a multi-channel transmission. As-
sion and compare the signal denoising performance of sume that two AWGN channels, named CH1 and CH2,
MSICA in single and multi-channel transmissions. are available to transmit the signal where, for exam-
Single-channel Transmission: In the first scenario, ple, we assume the variance of the noise in CH1 to be
i.e., in the case of single-channel transmission as de- smaller than in CH2, which means that CH1 is more re-
scribed in (26), if we divide the noisy signal into the liable and has better quality than CH2. In this case, if
even and odd samples, we obtain the vector of noisy we transmit the even and odd samples through different
observations as, channels, see Fig. 7(a), no matter through which one,
𝑟1 (𝑛) 𝑟(2𝑛) 𝑥(2𝑛) + 𝑤(2𝑛) then MSICA shows an extraordinary performance as it
r=[ ]=[ ]=[ ], is able to filter out the noise of a noisier channel. In
𝑟2 (𝑛) 𝑟(2𝑛 − 1) 𝑥(2𝑛 − 1) + 𝑤(2𝑛 − 1)
(28) such a scenario, the even and odd samples of the noisy
where 𝑤(2𝑛) and 𝑤(2𝑛 − 1) have the same variance, 𝜎𝑤
2
. signal are,
The covariance matrix of r can be written as,
𝑟1 (𝑛) = 𝑥(2𝑛) + 𝑤1 (𝑛), (32)a
1 𝜌′
C𝑟 = 𝐸{rr } = 𝑇
𝜎𝑟2 [ ′ ], (29) 𝑟2 (𝑛) = 𝑥(2𝑛 − 1) + 𝑤2 (𝑛), (32)b
𝜌 1

where, where 𝑤1 (𝑛) and 𝑤2 (𝑛) are the AWGN in CH1 and CH2,
𝜌𝜎2 respectively, and 𝜎𝑤
2
= 𝐾𝜎𝑤2
(𝐾 > 1). In this case, the
𝜎𝑟2 = 𝜎𝑥2 + 𝜎𝑤
2
, 𝜌′ = 2 𝑥 2 . (30) 2 1
covariance matrix of the vector of noisy observations r
𝜎𝑥 + 𝜎𝑤

30 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

0.04 14

13
0.035
Daubechies 3 Daubechies 3
Haar 12 Haar
0.03 Biorthogonal 2.2 Biorthogonal 2.2
Coiflets 4 11 Coiflets 4
Fejer-Korovkin 4 Fejer-Korovkin 4
0.025
MSICA 10 MSICA

SNRI
MSE
0.02 9

8
0.015
7
0.01
6
0.005
5

0 4
1 1.5 2 2.5 3 3.5 4 4.5 5 1 1.5 2 2.5 3 3.5 4 4.5 5

w
/ w w
/ w
2 1 2 1

(a) (b) (c)

Fig. 7 – (a) Proposed multi-channel transmission where the even and odd samples of the original signal are transmitted through different
channels and the receiver reconstructs the signal from the two outputs (𝑧−1 and 𝑧+1 indicates time shift by n=1 to the right and left,
respectively); (b) Mean Squared Error (MSE) vs. 𝜎𝑤2 /𝜎𝑤1 (𝜎𝑤1 = 0.1); (c) Performance of MSICA in terms of Signal-to-Noise Ratio
Improvement (SNRI).

3 3 3
2

2 2 2
1.5
Amplitude

1 1 1 1

0.5 0 0 0

0 -1 -1 -1
0 500 1000 1500 2000 2500 3000 3500 4000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Original Signal Approximation Approximation Approximation

3 1 1 1

2 0.5 0.5 0.5

Amplitude

1 0 0 0

0 -0.5 -0.5 -0.5

-1 -1 -1 -1
0 500 1000 1500 2000 2500 3000 3500 4000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Noisy Signal Detail Detail Detail

(a) (b) (c) (d)

3 3 3 3

2 2 2
2
1 1 1
1
0 0 0

-1 -1 -1 0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Approximation Approximation Approximation Approximation
1 1 1 1

0.5 0.5 0.5 0.5

0 0 0 0

-0.5 -0.5 -0.5 -0.5

-1 -1 -1 -1
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Detail Detail Detail Detail

(e) (f) (g) (h)

Fig. 8 – Comparison of MSICA with different wavelet transforms in decomposing a noisy PieceRegular signal. (a) Original Signal; (b)
Approximation and detail by Daubechies 3 wavelet; (c) Approximation and detail by Haar wavelet; (d) Approximation and detail by
Biorthogonal 2.2 wavelet; (e) Approximation and detail by Coiflets 4 wavelet; (f) Approximation and detail by Fejer-Korovkin 4 wavelet;
(g) Approximation and detail by discrete Meyer wavelet; (h) Approximation and detail by MSICA.

is, variance of CH2, which means that MSICA is able to

1 𝜌′ reject the AWGN of the CH2 from the noisy signal.
C𝑟 = 𝜎𝑟2 [ 2 ], (33)
𝜌′ 1+
(𝐾−1)𝜎𝑤 1 Moreover, to evaluate better the noise suppression per-
𝜎𝑟2
formance, we have also examined the performance of
where, MSICA in terms of Signal-to-Noise Ratio Improve-
𝜌𝜎𝑥2 ment (SNRI),
𝜎𝑟2 = 𝜎𝑥2 + 𝜎𝑤
2
, 𝜌′ = . (34)
1
𝜎𝑥2 + 𝜎𝑤2
1

As it is clear from (33), in the case that the variance of ∑ (𝑟(𝑛) − 𝑥(𝑛))2 ⎞
𝑁

the noise in CH1 and CH2 are different, the eigenvalues ⎛

⎜ ⎟
SNRI = SNR𝑜𝑢𝑡 −SNR𝑖𝑛 = 10 log ⎜ ⎜
𝑛=1 ⎟
⎟,
and eigenvectors of Cr are dependent to the parameter ⎜ 𝑁 2⎟
∑ (𝑥(𝑛)
̂ − 𝑥(𝑛))
𝐾. This means that the low-pass and high-pass filters in ⎝ 𝑛=1 ⎠
the whitening process will be adaptive to the parameter (35)
𝐾. In the following we will show that this adaptive filter where SNR𝑜𝑢𝑡 and SNR𝑖𝑛 are the SNR of the denoised
is able to reject the effect of CH2 almost entirely. Fig- signal (output) and of the noisy signal (input), respec-
ure 7(b) shows the performance of MSICA with respect tively. As shown in Fig. 7(c), the wavelet transforms
to the other wavelet transforms when the original signal have almost a fixed SNRI, whereas MSICA shows higher
is passed through two different channels. As shown in SNRI when the CH2 becomes noisier. This is because
Fig. 7(b), MSICA performance does not depend on the in MSICA the approximation and detail are statisti-

© International Telecommunication Union, 2020 31

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

4 3
2 4

2
1.5 2
Amplitude

2
1
1
0 0
0
0.5

-2 -2 -1
0
0 500 1000 1500 2000 2500 3000 3500 4000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Original Signal Approximation Approximation Approximation

4 1 2 2

0.5 1 1
Amplitude

2
0 0 0
0
-0.5 -1 -1

-2 -1 -2 -2
0 500 1000 1500 2000 2500 3000 3500 4000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Noisy Signal Detail Detail Detail

(a) (b) (c) (d)

3 4 3 3

2 2
2 2
1 1
0 1
0 0

-1 -2 -1 0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Approximation Approximation Approximation Approximation
2 2 2 2

1 1 1 1

0 0 0 0

-1 -1 -1 -1

-2 -2 -2 -2
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Detail Detail Detail Detail

(e) (f) (g) (h)

Fig. 9 – Comparison of MSICA with different wavelet transforms in decomposing a PieceRegular signal corrupted by impulse noise. (a)
Original and Noisy Signal; (b) Approximation and detail by Daubechies 3 wavelet; (c) Approximation and detail by Haar wavelet; (d)
Approximation and detail by Biorthogonal 2.2 wavelet; (e) Approximation and detail by Coiflets 4 wavelet; (f) Approximation and detail
by Fejer-Korovkin 4 wavelet; (g) Approximation and detail by discrete Meyer wavelet; (h) Approximation and detail by MSICA.

cally independent; hence, MSICA is able to extract the 0.14

noise signal from CH2 (via channel diversity), while the
wavelet transforms are not able to do so. 0.12 Daubechies 3

Fig. 8 shows a signal decomposition where 𝜎𝑤2 = 0.2,

Haar
Biorthogonal 2.2
0.1
𝜎𝑤1 = 0.05. As it can be seen, the approximation ob-
Coiflets 4
Fejer-Korovkin 4

tained using MSICA is less noisy than the one obtained

MSICA
0.08
MSE

using the other wavelet transforms (Daubechies 3, Haar, 0.06

Biorthogonal 2.2, Coiflets 4, Fejer-Korovkin 4, and

Meyer). This result confirms our statement and shows 0.04

that, because of the statistical independenc between the 0.02

approximation and detail, MSICA is able to extract the

AWGN from the noisier channel. 0
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

In the other experiment, in order to show visibly that Pa ×10 -3

MSICA is able to extract the noise of CH2, we have ex- Fig. 10 – Impulse noise rejection in terms of Minimum Square Er-
plored its performance when the odd samples, passed ror (MSE); MSICA performance does not depend on 𝑃𝑎 , whereas
through CH2, are corrupted by impulse noise. The the performance of the other transforms decreases when 𝑃𝑎 in-
Probability Density Function (PDF) of the impulse creases.
noise is given as, tract accurately the impulse components from the noisy
signal. However, as it is shown in Fig. 9(h), MSICA is
⎧ 𝑃𝑎 𝑤 = 𝑎,
{ successful as the detail contains all the impulse compo-
𝑃 (𝑤) = ⎨ 𝑃𝑎 𝑤 = −𝑎, (36) nents. This is because in MSICA the approximation and
{
⎩ 1 − 2𝑃𝑎 𝑤 = 0,
detail are statistically independent and, since the im-
pulse noise is statistically independent from the original
where 2𝑃𝑎 is the probability of existence of impulse noise
signal, MSICA can extract it in the detail coefficients.
in the received samples. In Fig. 9(a), the noisy signal
is obtained by passing the even samples of the original Fig. 10 shows the performance of MSICA compared with
signal through CH1 with AWGN with zero mean and different wavelet transforms when various values of 𝑃𝑎 ,
𝜎𝑤2
= 0.004, while the odd samples were passed through as in (36), are considered. Here, the detail coefficients
obtained by different methods have been set to zero to
2
CH2 with impulse noise (𝑃𝑎 = 0.01 and 𝑎 = 1.5).
Fig. 9(b)-(h) show the performance of MSICA com- denoise the noisy signal. Since MSICA is able to extract
pared to a number of well-known wavelet transforms. the impulse noise, its performance does not depend on
As it is clear from Fig. 9(b)-(g), the traditional wavelet 𝑃𝑎 , whereas the performance of the other transforms
transforms (i.e., Daubechies 3, Haar, Biorthogonal 2.2, decreases when 𝑃𝑎 increases.
Coiflets 4, Fejer-Korovkin 4, Meyer) are not able to ex- To show that MSICA works on real signals too, we

32 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

0.8 1 1 1

0.5 0.5 0.5

0.6
0 0 0
0.4
-0.5 -0.5 -0.5
0.2
-1 -1 -1
Amplitude

0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
0 Approximation Approximation Approximation
0.2 0.2 0.1
-0.2
0.1 0.05
-0.4 0.1
0 0
-0.6 0
-0.1 -0.05

-0.8 -0.1 -0.2 -0.1

0 100 200 300 400 500 600 0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
Audio Signal Detail Detail Detail

(a) (b) (c) (d)

1 1 1 1

0.5 0.5 0.5 0.5

0 0 0 0

-0.5 -0.5 -0.5 -0.5

-1 -1 -1 -1
0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 20 40 60 80 100 120 140 160 180 200
Approximation Approximation Approximation Approximation

0.1 0.2 0.1 0.2

0.05 0.1 0.05 0.1

0 0 0 0

-0.05 -0.1 -0.05 -0.1

-0.1 -0.2 -0.1 -0.2

0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 20 40 60 80 100 120 140 160 180 200
Detail Detail Detail Detail

(e) (f) (g) (h)

Fig. 11 – Approximation and detail coefficients obtained from an audio signal by some well-known wavelet transforms and MSICA.
(a) Original audio Signal. Approximation and detail by (b) Daubechies 3; (c) Haar; (d) Biorthogonal 2.2; (e) Coiflets 4; (f) Fejer-Korovkin 4;
(g) discrete Meyer; (h) MSICA.

have examined its performance on the signal depicted 2011.

in Fig. 11(a), which is an audio signal with a sampling [2] Y. Kopsinis and S. McLaughlin, “Development of emd-
frequency equal to 16 KHz. Fig. 11(b)-(g) show the based denoising methods inspired by wavelet threshold-
performance of the considered wavelet transforms in de- ing,” IEEE Transactions on Signal Processing, vol. 57,
composing this signal into approximation and detail. As no. 4, pp. 1351–1362, 2009.
it is clear from Fig. 11(h), like the other transforms, [3] A. F. Laine, S. Schuler, J. Fan, and W. Huda, “Mam-
MSICA is also able to decompose this audio signal into mographic feature enhancement by multiscale analysis,”
approximation and detail. IEEE Transactions on Medical Imaging, vol. 13, no. 4,
pp. 725–740, 1994.
[4] F. M. Bayer, A. J. Kozakevicius, and R. J. Cintra, “An
5. CONCLUSIONS iterative wavelet threshold for signal denoising,” Signal
We presented MSICA, a new method for Multi-Scale Processing, vol. 162, pp. 10–20, 2019.
decomposition based on Independent Component Anal- [5] P. G. Bascoy, P. Quesada-Barriuso, D. B. Heras, and
ysis (ICA), where the approximation and detail are sta- F. Argüello, “Wavelet-based multicomponent denoising
profile for the classification of hyperspectral images,”
tistically independent. First, we extracted two corre-
IEEE Journal of Selected Topics in Applied Earth Ob-
lated signals from the original digital signal by separat- servations and Remote Sensing, vol. 12, no. 2, pp.
ing their even and odd samples; then, we proved that ex- 722–733, 2019.
tracting the independent components of the correlated
[6] A. Wong and A. K. Mishra, “Generalized probabilistic
signals leads to the decomposition of the original sig- scale space for image restoration,” IEEE Transactions
nal into the approximation and detail. We showed that on Image Processing, vol. 19, no. 10, pp. 2774–2780,
MSICA outperforms well-known wavelet transforms in 2010.
signal denoising when transmitting a signal over mul- [7] G. Gilboa, “Nonlinear scale space with spatially varying
tiple (noisy) channels as it exploits channel diversity. stopping time,” IEEE Transactions on Pattern Analysis
This property makes MSICA useful in many critical and Machine Intelligence, vol. 30, no. 12, pp. 2175–2187,
scenarios such as transmitting multimedia content reli- 2008.
ably through underwater acoustic channels. Since these [8] A. Mishra, A. Wong, D. A. Clausi, and P. W. Fieguth,
channels may vary quickly over time, it is difficult to “Quasi-random nonlinear scale space,” Pattern Recog-
estimate them, which makes transmitting multimedia nition Letters, vol. 31, no. 13, pp. 1850–1859, 2010.
content underwater a very challenging task. [9] T. Lindeberg, “Scale-space theory: A basic tool for ana-
lyzing structures at different scales,” Journal of applied
statistics, vol. 21, no. 1-2, pp. 225–270, 1994.
REFERENCES
[10] F. Jager, I. Koren, and L. Gyergyek, “Multiresolution
[1] B. Ophir, M. Lustig, and M. Elad, “Multi-scale dictio- representation and analysis of ecg waveforms,” in Pro-
nary learning using wavelets,” IEEE Journal of Selected ceedings Computers in Cardiology. IEEE, 1990, pp.
Topics in Signal Processing, vol. 5, no. 5, pp. 1014–1024, 547–550.

© International Telecommunication Union, 2020 33

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[11] P. Flandrin, G. Rilling, and P. Goncalves, “Empirical [26] A. Belouchrani, K. Abed-Meraim, J.-F. Cardoso, and
mode decomposition as a filter bank,” IEEE Signal Pro- E. Moulines, “A blind source separation technique using
cessing Letters, vol. 11, no. 2, pp. 112–114, 2004. second-order statistics,” IEEE Transactions on Signal
[12] B. Weng, M. Blanco-Velasco, and K. E. Barner, “ECG Processing, vol. 45, no. 2, pp. 434–444, 1997.
denoising based on the empirical mode decomposi- [27] H. Sawada, N. Ono, H. Kameoka, D. Kitamura, and
tion,” in International Conference of the Engineering H. Saruwatari, “A review of blind source separation
in Medicine and Biology Society. IEEE, 2006, pp. 1–4. methods: two converging routes to ilrma originating
[13] M. Blanco-Velasco, B. Weng, and K. E. Barner, “ECG from ica and nmf,” APSIPA Transactions on Signal and
signal denoising and baseline wander correction based Information Processing, vol. 8, 2019.
on the empirical mode decomposition,” Computers in [28] A. Hyvärinen and E. Oja, “Independent component
biology and medicine, vol. 38, no. 1, pp. 1–13, 2008. analysis: algorithms and applications,” Neural Net-
[14] S. Mallat, “A theory for multiresolution signal decom- works, vol. 13, no. 4, pp. 411–430, 2000.
position: The wavelet representation,” IEEE Transac- [29] P. Comon, “Independent component analysis, a new
tions on Pattern Analysis and Machine Intelligence, concept?” Transactions on Signal Processing, vol. 36,
vol. 11, no. 7, pp. 674–693, 1989. no. 3, pp. 287–314, 1994.
[15] S. Sardy, P. Tseng, and A. Bruce, “Robust wavelet [30] I. Kauppinen and K. Roth, “Improved noise reduction
denoising,” IEEE Transactions on Signal Processing, in audio signals using spectral resolution enhancement
vol. 49, no. 6, pp. 1146–1152, 2001. with time-domain signal extrapolation,” IEEE Trans-
[16] D. L. Donoho, “De-noising by soft-thresholding,” IEEE actions on Speech and Audio Processing, vol. 13, no. 6,
Transactions on Information Theory, vol. 41, no. 3, pp. pp. 1210–1216, 2005.
613–627, 1995. [31] S. Durand and J. Froment, “Artifact free signal denois-
ing with wavelets,” in IEEE International Conference
[17] D. L. Donoho and J. M. Johnstone, “Ideal spatial adap-
on Acoustics, Speech, and Signal Processing (ICASSP),
tation by wavelet shrinkage,” biometrika, vol. 81, no. 3,
vol. 6, 2001, pp. 3685–3688.
pp. 425–455, 1994.
[32] Z. A. A. Alyasseri, A. T. Khader, M. A. Al-Betar, A. K.
[18] G. Chen, T. D. Bui, and A. Krzyżak, “Image denois-
Abasi, and S. N. Makhadmeh, “Eeg signals denoising
ing with neighbour dependency and customized wavelet
using optimal wavelet transform hybridized with effi-
and threshold,” Pattern Recognition, vol. 38, no. 1, pp.
cient metaheuristic methods,” IEEE Access, vol. 8, pp.
115–124, 2005.
10 584–10 605, 2019.
[19] A. Witkin, “Scale-space filtering: A new approach to
[33] D. L. Donoho and I. M. Johnstone, “Adapting to un-
multi-scale description,” in IEEE International Con-
known smoothness via wavelet shrinkage,” Journal of
ference on Acoustics, Speech, and Signal Processing
the American Statistical Association, vol. 90, no. 432,
(ICASSP), vol. 9. IEEE, 1984, pp. 150–153.
pp. 1200–1224, 1995.
[20] A. Wong and X. Y. Wang, “A bayesian residual trans-
[34] A. Hyvarinen, “Fast and robust fixed-point algorithms
form for signal processing,” IEEE Access, vol. 3, pp.
for independent component analysis,” IEEE Transac-
709–717, 2015.
tions on Neural Networks, vol. 10, no. 3, pp. 626–634,
[21] B. Mijovic, M. De Vos, I. Gligorijevic, J. Taelman, and 1999.
S. Van Huffel, “Source separation from single-channel
[35] R. R. Coifman and D. L. Donoho, Translation invariant
recordings by combining empirical-mode decomposition
denoising. Springer, 1995.
and independent component analysis,” IEEE Transac-
tions on Biomedical Engineering, vol. 57, no. 9, pp.
2188–2196, 2010.
[22] S. Wang and C. J. James, “On the independent com-
ponent analysis of evoked potentials through single or
few recording channels,” in 29th Annual International
Conference of the IEEE Engineering in Medicine and
Biology Society. IEEE, 2007, pp. 5433–5436.
[23] M. E. Davies and C. J. James, “Source separation using
single channel ica,” Signal Processing, vol. 87, no. 8, pp.
1819–1832, 2007.
[24] G.-J. Jang, T.-W. Lee, and Y.-H. Oh, “Single-channel
signal separation using time-domain basis functions,”
IEEE Signal Processing Letters, vol. 10, no. 6, pp.
168–171, 2003.
[25] J. Miettinen, M. Matilainen, K. Nordhausen, and
S. Taskinen, “Extracting conditionally heteroskedastic
components using independent component analysis,”
Journal of Time Series Analysis, vol. 41, no. 2, pp.
293–311, 2020.

34 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

AUTHORS

Abolfazl Hajisami received his PhD degree in Electrical

and Computer Engineering (ECE) from Rutgers University,
NJ, USA, in 2018. He received his MSc and BSc degrees from
Sharif University of Technology and from Shahid Beheshti
University (Tehran, Iran), in 2010 and 2008, respectively.
His research interest lies in wireless communication, vehicle-
to-everything (V2X) communication, and signal processing.
His PhD thesis dealt with spectral and energy efficiency in
cloud radio access networks. He is now a Staff Research En-
gineer at Qualcomm, where he works on V2X technologies.

Dario Pompili is an Assoc. Prof. with the Dept. of

ECE at Rutgers, where he directs the Cyber-Physical Sys-
tems Laboratory (CPS Lab), which focuses on mobile edge
computing, wireless networking, acoustic communications,
and sensor networks. He received his PhD in ECE from the
Georgia Institute of Technology in 2007. He had previously
received his ‘Laurea’ (combined BS and MS) and Doctor-
ate degrees in Telecommunications and System Engineering
from the U. of Rome “La Sapienza,” Italy, in 2001 and 2004,
respectively. He has received a number of awards including
the NSF CAREER’11, ONR Young Investigator Program’12,
and DARPA Young Faculty’12 awards. In 2015, he was nom-
inated Rutgers-New Brunswick Chancellor’s Scholar. He has
served in various roles on many international conferences.
He published about 150 refereed scholar publications, some
of which selected to receive best paper awards: with about
11,600 citations, Dr. Pompili has an h-index of 42 and an
i10-index of 101 (Google Scholar, Oct’20). He is a Senior
Member of the IEEE Communications Society (2014) and a
Distinguished Member of the ACM (2019). He is an Area
Editor for IEEE Transactions on Mobile Computing (TMC).

© International Telecommunication Union, 2020 35

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

a.L@"a1. aP*AP*ah :_PlS *PJJlLA*hAPLa AL h>1 ALh1_L1h P6

h>AL:a

GmB;B ixQ`BR-8 - *Hm/B *KTQHQk-8 - MiQMBQ A2`j-8 - :Bmb2TT2 JbbBKBHBMQ JBHQiik-8 - :B+QKQ JQ`#BiQ9-8 - aHpiQ`2
Zmii`QTMB8
R
.A11- lMBp2`bBiv Q7 *;HB`B- AiHv- .AA1a- lMBp2`bBiv J2/Bi2``M2 Q7 _2;;BQ *H#`B- AiHv- j .AJ1a- lMBp2`bBiv Q7
k

*H#`B- AiHv- 9 .A11A- lMBp2`bBiv Q7 *iMB- AiHv- 8 *LAh @ LiBQMH AMi2`@lMBp2`bBiv *QMbQ`iBmK 7Q`
h2H2+QKKmMB+iBQMb- AiHv-
LPh1, *Q``2bTQM/BM; mi?Q`, :B+QKQ JQ`#BiQ- ;B+QKQXKQ`#BiQ!mMB+iXBi

#bi`+i Ĝ h?2 M2r TTHB+iBQMb TQTmHiBM; i?2 6mim`2 AMi2`M2i rBHH BM+`2bBM;Hv `2Hv QM i?2 2t+?M;2 Q7 /i #2ir22M
;`QmTb Q7 /2pB+2b- /vMKB+HHv 2bi#HBb?2/ ++Q`/BM; iQ i?2B` T`Q}H2 M/ ?#Bib U2X;X-  +QKKQM BMi2`2bi BM i?2 bK2 bQ7ir`2
mT/i2b M/ b2`pB+2bVX h?Bb rBHH /2}MBi2Hv +?HH2M;2 i`/BiBQMH ;`QmT +QKKmMB+iBQM bQHmiBQMb i?i H+F i?2 M2+2bb`v
~2tB#BHBiv BM ;`QmT KM;2K2Mi M/ /Q MQi bmTTQ`i 2z2+iBp2 +QMi`QH TQHB+B2b QM BMpQHp2/ 2M/TQBMib UBX2X- mi?Q`Bx2/ b2M/2`b
M/ BMi2M/2/ `2+2Bp2`bVX hQ //`2bb i?2 +Bi2/ Bbbm2b- i?2 B/2 Q7 BMi`Q/m+BM; M2r /Bb`mTiBp2 M2irQ`F@Hv2` bQHmiBQMb ?b
2K2`;2/ 7`QK `2+2Mi HBi2`im`2X KQM; i?2K- aQ+BQ+bi ?b #22M i?2Q`Bx2/ b M 2M#H2` Q7 ~2tB#H2 BMi2`+iBQMb #2ir22M
;`QmTb Q7 /2pB+2b iB2/ #v bQ+BH `2HiBQMb?BTbX AM i?Bb TT2` r2 bi`i 7`QK i?2 +QM+2Ti Q7 aQ+BQ+bi M/ T`QTQb2  bQHmiBQM
#b2/ QM aQ7ir`2 .2}M2/ L2irQ`FBM; Ua.LV 7Q` Bib BKTH2K2MiiBQM i i?2 M2irQ`F Hv2` BM i?2 AMi2`M2i Q7 h?BM;bX h?2
T2`7Q`KM+2 Q7 aQ+BQ+bi Bb bim/B2/ M/ +QKT`2/ iQ K2i?Q/b `mMMBM; i i?2 TTHB+iBQM Hv2` i?i T`QpB/2 bBKBH` 72im`2bX
1tT2`BK2MiH `2bmHib- +?B2p2/ i?`Qm;? M 2KmHiBQM@#b2/ THv;`QmM/- +QM}`K i?i i?2 aQ+BQ+bi TT`Q+? HHQrb 7Q` 
bB;MB}+Mi `2/m+iBQM Q7 bB;MHBM; M/ /i T+F2ib +B`+mHiBM; BM i?2 M2irQ`F rBi? `2bT2+i iQ i`/BiBQMH TT`Q+?2bX
E2vrQ`/b Ĝ *QKKmMB+iBQM S`BKBiBp2b- AMi2`M2i Q7 h?BM;b- aQ+BH AMi2`M2i Q7 h?BM;b- aQ7ir`2 .2}M2/ L2irQ`Fb

RX ALh_P.l*hAPL U2X;X- p2?B+H2b QM  ;Bp2M `Q/ b2;K2MiVX

amTTQ`i Q7 BMi2`+iBQMb #2ir22M /2pB+2b `Bb2b QmibiM/@
h?2 AMi2`M2i Bb 2tT2`B2M+BM;  `TB/ i`Mb7Q`KiBQM BM; +?HH2M;2b 7Q` M2irQ`F QT2`iQ`bX 6B`bi- AQh TTHB@
Tmb?2/ #v i?2 ;`QrBM; M22/ iQ Qp2`+QK2 Bib BMi`BMbB+ HBK@ +iBQMb `2[mB`2 i?2 /vMKB+ M/ ~2tB#H2 KM;2K2Mi Q7
BiiBQMb M/ QbbB}+iBQM- r?B+? +?HH2M;2 M2irQ`F T`+@ ;`QmT@#b2/ BMi2`+iBQMb- r?Qb2 b+QT2 Bb /2+B/2/ ++Q`/@
iBiBQM2`b M/ `2b2`+?2`bX h?2 T`2bbBM; M22/ iQ +QK2 BM; iQ  ;Bp2M iQTB+ M/ iQ i?2 iB2b 2tBbiBM; #2ir22M BM@
iQ i?2 /2}MBiBQM Q7  M2r AMi2`M2i Q7 i?2 7mim`2 Bb HbQ pQHp2/ 2M/TQBMib U2X;X- +Q@HQ+HBiv- bBKBH`Biv Q7 /2pB+2b-
KQiBpi2/ #v i?2 KmHiBim/2 Q7 AMi2`M2i Q7 h?BM;b UAQhV 2i+XVX
TTHB+iBQMb i?i `2 `2+2MiHv 2K2`;BM; BM p`BQmb p2`@ a2+QM/- i?2 +QKKmMB+iBQM 2M/TQBMib b?QmH/ ?p2 i?2
iB+H K`F2ib (R)X am+? TTHB+iBQMb `2 BM+`2bBM;Hv TQr2` iQ +QMi`QH /i 2t+?M;2bX AM/22/-  +QMi`QH Q7 i?2
+?`+i2`Bx2/ #v ;`QmT@#b2/ UBX2X- QM2@iQ@KMv- KMv@ 2M#H2/ /i `2+2Bp2`b Bb bi`QM;Hv /2bB`2/ #v i?2 bQm`+2
iQ@KMvV +QKKmMB+iBQMb 2bi#HBb?2/ #2ir22M H`;2 b2ib /2pB+2- /m2 iQ i?2 TQi2MiBHHv +QM}/2MiBH Mim`2 Q7 2t@
Q7 /2pB+2b BM M22/ Q7 bBKmHiM2QmbHv 2t+?M;BM; /i- +?M;2/ /iX
2X;X- BM i?2 +b2 Q7 b2MbQ`bǶ bQ7ir`2 mT/i2b- b2`pB+2 /@ JQ`2Qp2`- i?2 KbbBp2 T`2b2M+2 Q7 ;`QmT@#b2/ +QKKm@
p2`iBb2K2Mib- /2pB+2 +QM};m`iBQMbX MB+iBQMb 2bi#HBb?2/ #v #BHHBQMb Q7 AQh /2pB+2b- 2tT2+i2/
AM ?mKM@+2Mi`B+ +QKKmMB+iBQMb- 7`2[m2Mi BMbiMi K2b@ iQ BM+`2b2 2p2M i  ?B;?2` T+2 BM i?2 M2` 7mim`2- +M
b;BM; Q++m`b rBi?BM +QKKmMBiB2b Q7 mb2`b b?`BM; bBK@ +mb2 M2irQ`F +QM;2biBQM M/ rbi2 /2pB+2 M/ M2irQ`F
BH` BMi2`2bib M/ T2QTH2 H`;2Hv BMi2`+iBM; rBi? i?2B` `2bQm`+2b- mMH2bb T`QT2` +QmMi2`K2bm`2b `2 iF2MX
7`B2M/b- M/ 7`B2M/b Q7 i?2B` 7`B2M/b- Qp2` bQ+BH M2irQ`FbX  bQHmiBQM Bb `2[mB`2/ iQ HHQr MQ/2b iQ ~2tB#Hv bT2+B7v
aBKBH`Hv- ;`QmTb Q7 AQh /2pB+2b `2 HBF2Hv iQ BMi2`+i ?Qr iQ T`BQ`BiBx2 U}Hi2`V i?2 MQ/2b 7`QK r?B+? i?2v rMi
rBi? 2+? Qi?2`- 2bT2+BHHv B7 i?2v `2 HQ+i2/ BM i?2 UQ` i?2v /Q MQi rMiV iQ `2+2Bp2 /i- M/ i?2 M2irQ`F
bK2 TH+2 U2X;X- b2MbQ`bf+imiQ`b BM i?2 bK2 #mBH/@ iQ `2+i ++Q`/BM;Hv- bQ iQ T`2p2Mi i?2 i?`2ib Q7 .2MBH
BM;V- `2 QrM2/ #v i?2 bK2 mb2` U2X;X- +QMbmK2` /2@ Q7 a2`pB+2 U.QaV ii+FbX
pB+2b M/ ?QK2 TTHBM+2bV- b?`2 bBKBH` T`Q}H2b U2X;X- *QMp2MiBQMH KmHiB+bi@#b2/ TT`Q+?2b (k)- #2BM;
i?2 bK2 #`M/ M/ ivT2V- Q` 7`2[m2MiHv K22i 2+? Qi?2` KBMHv /2bB;M2/ iQ bBKmHiM2QmbHv i`MbKBi /i 7`QK

© International Telecommunication Union, 2020 37

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

QM2 Q` KmHiBTH2 b2M/2`b iQ  ;`QmT Q7 UmMFMQrMV `2@ +iBQM Hv2`X _2bmHib b?Qr i?i i?2 aQ+BQ+bi T@
+2Bp2`b- 7BH BM MiBp2Hv +?B2pBM; bm+? Q#D2+iBp2b M/ BM T`Q+? HHQrb 7Q`  `2/m+iBQM Q7 bB;MHHBM; M/ /i
2Mbm`BM; i?2 `2[mB`2/ ~2tB#BHBiv BM ;`QmT 2bi#HBb?K2Mi T+F2ib #v  7+iQ` Q7 Ry M/ 8- `2bT2+iBp2Hv- BM i?2
M/ KM;2K2MiX *HmKbv Ti+?2b iQ 2tBbiBM; KmHiB+bi b+2M`BQ r?2`2 i?2 MmK#2` Q7 `2+BTB2Mib Bb ?B;? M/
bQHmiBQMb Kv 7m`i?2` +QKTHB+i2 i?2B` /2bB;M M/ ?BM/2` `2 +HQb2 iQ 2+? Qi?2`X
i?2B` UH`2/v HBKBi2/V /2THQvK2MiX h?2 `2KBM/2` Q7 i?Bb TT2` Bb Q`;MBx2/ b 7QHHQrbX AM
h?Bb Bb r?v BM (j) mi?Q`b `;m2 BM 7pQ` Q7  MQp2H a2+iBQM k r2 bm`p2v i?2 `2Hi2/ HBi2`im`2 BM i?2 }2H/
M/ 7mim`2@T`QQ7 +QKT`2?2MbBp2 bQHmiBQM- MK2/ aQ+BQ@ Q7 ;`QmT@#b2/ +QKKmMB+iBQMbX AM a2+iBQM j r2 BMi`Q@
+bi- 2M+QKTbbBM; #Qi?  +QKKmMB+iBQM K2i?Q/ M/ /m+2 i?2 KDQ` aQ+BQ+bi +QM+2Tib M/ /Bb+mbb i?2 /2bB;M
 /i /2HBp2`v b+?2K2- ;QBM; r2HH #2vQM/ AMi2`M2i S`Q@ ;mB/2HBM2b r2 ?p2 +QMbB/2`2/X a2+iBQM 9 /Bb+mbb2b i?2
iQ+QH UASV@#b2/ KmHiB+biX aQ+BQ+bi Bb i?2Q`Bx2/ b  +QM+2Bp2/ `+?Bi2+im`H 7`K2rQ`F HQM; rBi? i?2 2MpB@
K2Mb 7Q` B/2MiB7vBM;- BM  ~2tB#H2 KMM2`- i?2 BMi2M/2/ bBQM2/ 2MiBiB2b M/ i?2B` KBM `QH2 M/ 7mM+iBQMHBiB2bX
2M/TQBMib Ub2M/2`bf`2+2Bp2`bV Q7 /i 2t+?M;2 b2bbBQMbX AM a2+iBQM 8 r2 /2b+`B#2 i?2 KBM T`Q+2/m`2b iQ 2M#H2
:`QmTb `2 /vMKB+HHv +`2i2/ ++Q`/BM; iQ i?2 Km@ i?2 i`2iK2Mi UBX2X- 7Q`r`/BM;- /`QTTBM;- KQ/B7vBM;V Q7
imH TQbBiBQM Q7 2M/TQBMib BM  bQ+BH M2irQ`F Q7 /2pB+2b aQ+BQ+bi T+F2ibX h?2M- BM a2+iBQM e r2 /2b+`B#2 i?2
M/ i?2 ivT2 Q7 `2HiBQMb?BTb KQM; i?2K- #v K2Mb Q7 THv;`QmM/ 7Q` i?2 2pHmiBQM- #27Q`2 /Bb+mbbBM; +?B2p2/
T`QT2`Hv /2}M2/ }Hi2`BM; `mH2b M/ TQHB+B2bX `2bmHib BM a2+iBQM dX 6BMHHv- BM a2+iBQM 3 r2 /`r bQK2
h?Bb rQ`F i`2bm`2b i?2 i?2Q`2iB+H MHvbBb BM i?2 +Bi2/ +QM+Hm/BM; `2K`FbX
pBbBQM TT2` M/ iF2b  bB;MB}+Mi bi2T 7Q`r`/ #Qi?
BM i2`Kb Q7 T`+iB+H /2bB;M M/ 2tT2`BK2MiH 2pHm@
iBQMX >2`2BM- r2 `;m2 #Qmi i?2 +imH TQbbB#BHBiv
kX "*E:_PlL. L. JPhAo@
Q7 BKTH2K2MiBM; i?2 +QM+2Bp2/ aQ+BQ+bi T`BKBiBp2 b  hAPLa
M2irQ`F@Hv2` bQHmiBQM BM AQh /QKBMb- r?2`2BM brBi+?2b AM i?Bb b2+iBQM r2 rBHH }`bi Qp2`pB2r ?Qr ;`QmT +QK@
M/ `Qmi2`b `2 `2bTQMbB#H2 7Q` i?2 2{+B2Mi /2HBp2`v Q7 KmMB+iBQMb `2 i`/BiBQMHHv bmTTQ`i2/ BM i?2 AMi2`M2i
T+F2ib Bbbm2/ #v AQh /2pB+2bX AM T`iB+mH`- i?2 `272`@ Ub22 a2+iBQM kXRVc i?2M- r2 rBHH /Bb+mbb i?2 /`r#+Fb Q7
2M+2 M2irQ`F BM7`bi`m+im`2 Bb /2THQv2/ ++Q`/BM; iQ i?2 bm+? bQHmiBQMb Ub22 a2+iBQM kXkVc }MHHv- i?2 /pMi;2b
aQ7ir`[email protected]}M2/ L2irQ`FBM; Ua.LV i2+?MQHQ;v (9)X Q7 2tTHQBiBM; bQ+BH `2HiBQMb?BTb #2ir22M AQh MQ/2b `2
a.L ?b #22M BMi`Q/m+2/ iQ //`2bb  ivTB+H Bbbm2 BM bmKK`Bx2/ Ub22 a2+iBQM kXjVX
i`/BiBQMH AS M2irQ`Fb, i?2 H+F Q7 T`Q;`KK#BHBiv BM
M2irQ`F KM;2K2Mi M/ +QM};m`iBQMX h?MFb iQ Bib kXR JmHiB+bi TT`Q+?2b BM i?2 HBi2`im`2
T2+mHB`BiB2b- Bi +M THv  +`m+BH `QH2 iQ #`BM; i?2 bQ@
+BH /BK2MbBQM BMiQ i?2 ;`QmT /i /2HBp2`v T`Q+2/m`2b  H`;2 MmK#2` Q7 /Bz2`2Mi TTHB+iBQMb `2Hv QM QM2@iQ@
2M7Q`+2/ i i?2 M2irQ`F Hv2`X KMv M/ KMv@iQ@KMv /i i`{+ 2t+?M;2X h?2v
h?2 KBM +QMi`B#miBQMb Q7 i?2 rQ`F +M #2 bmKK`Bx2/ `M;2 7`QK HBp2 pB/2Q bi`2KBM;- m/BQfpB/2Q +QM72`2M+@
b 7QHHQrbX BM; (d) M/ KmHiBTHv2` ;K2b (3) iQ +QKKmMB+iBQMb #2@
ir22M ;`QmTb Q7 b2`p2`b rBi?BM /i +2Mi2`b (N) M/ rB/2@
Ç h?2 /2bB;M Q7 M `+?Bi2+im`H 7`K2rQ`F 2M+QK@ `2 +QMi`QH BM bK`i ;`B/b (Ry)X JmHiB+biBM; 7mM+iBQM@
TbbBM; HH i?2 2MiBiB2b M/ 7mM+iBQMHBiB2b bmTTQ`i@ HBiv Bb ivTB+HHv H2p2`;2/ BM bm+? +QMi2tib- r?B+? +M
BM; aQ+BQ+bi- ++Q`/BM; iQ  bQ7ir`2@/2}M2/ M2i@ #2 T2`7Q`K2/ 2Bi?2` i i?2 M2irQ`F UASV Hv2` Q` i i?2
rQ`F TT`Q+?X TTHB+iBQM Hv2` (k)- (RR) M/ HbQ rBi? i?2 bmTTQ`i Q7
Ç h?2 /2}MBiBQM Q7 i?2 KBM T`Q+2/m`2b 7Q` i?2 +`2@ a.L (Rk)- (Rj)X
iBQM Q7 i?2 aQ+BQ+bi T+F2ib- i?2B` 7Q`r`/BM; M/ AS@#b2/ KmHiB+biBM;X h`/BiBQMH KmHiB+bi `Qmi@
}Hi2`BM;- M/ i?2 bm#b+`BTiBQM Q7 /2pB+2b iQ aQ+BQ@ BM; M/ KM;2K2Mi T`QiQ+QHb- bm+? b S`QiQ+QH@
+bi ;`QmTbX AM/2T2M/2Mi JmHiB+bi USAJV (R9) M/ AMi2`M2i :`QmT
JM;2K2Mi S`QiQ+QH UA:JSV (R8)- 2z2+iBp2Hv 2bi#HBb?
Ç h?2 T2`7Q`KM+2 bb2bbK2Mi i?`Qm;? i?2 rB/2Hv M/ KBMiBM KmHiB+bi +QKKmMB+iBQM Ti?b #2ir22M
FMQrM JBMBM2i M2irQ`F 2KmHiQ` (8)- r?2M /2HBM; bQm`+2b M/ `2+2Bp2`b iQ 2M#H2 i?2 7Q`r`/BM; Q7 T+F@
rBi? Tmb?@#b2/ /i /Bbb2KBMiBQM M/ /2THQvBM; 2ib iQ  KmHiB+bi ;`QmTX 1+? ;`QmT Bb bbB;M2/  mMB[m2
i?2 aQ+BQ+bi M2irQ`F TTHB+iBQM BMiQ i?2 PLPa +Hbb . AS //`2bbX  ?Qbi +M b2M/ /i iQ  KmHiB+bi
a.L +QMi`QHH2` (e)X h?2 BKT+i Q7 /Bz2`2Mi 2M/@ ;`QmT #v mbBM; i?2 HQ+H M2irQ`F KmHiB+bi +T#BHBiv iQ
TQBMi /Bbi`B#miBQM Tii2`Mb M/ /Bz2`2Mi BMpQHp2/ i`MbKBi i?2 T+F2iX  KmHiB+bi `Qmi2`- mTQM `2+2T@
bQ+BH `2HiBQMb?BTb QM i?2 T2`7Q`KM+2 Bb 2pHmi2/ iBQM Q7  T+F2i- HQQFb mT Bib `QmiBM; i#H2 M/ 7Q`r`/b
#v +QKT`BM; Qm` T`QTQbH iQ M Hi2`MiBp2 T@ i?2 T+F2i iQ i?2 TT`QT`Bi2 Qmi;QBM; BMi2`7+2X :`QmT
T`Q+? r?2`2 i?2 ;`QmTb `2 +`2i2/ i i?2 TTHB@ K2K#2`b?BT Bb KM;2/ i i?2 M2irQ`F H2p2H i?`Qm;?

38 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

`Qmi2`bX q?2M  ?Qbi /2+B/2b iQ DQBMfH2p2  T`iB+mH` BM i?2 +b2 Q7  KmHiB@T`iv pB/2Q@+QM72`2M+BM; b2`pB+2
KmHiB+bi ;`QmT- Bi b2M/b i?2 `2[m2bi iQ i?2 HQ+H KmHiB@ (RN)X
+bi `Qmi2`- i?`Qm;? A:JS (R8)X
AS KmHiB+bi HHQrb /i iQ #2 /Bbi`B#mi2/ BM bm+?  rv kXk q2FM2bb2b M/ QT2M Bbbm2b
i?i i?2 H2bi KQmMi Q7 `2THB+b Q7 i?2 bK2 T+F2i Bb
h?2 mb2 Q7 i?2 i`/BiBQMH AS KmHiB+bi Bb T`QM2 iQ KmH@
TH+2/ BMiQ i?2 M2irQ`FX
iBTH2 Bbbm2b,
AM Bib `2+2Mi p2`bBQM- pj- A:JS HHQrb iQ bT2+B7v i?2 b2i
Q7 b2M/2`b 7`QK r?B+?  MQ/2 rMib iQ `2+2Bp2- BM ;`22@ Ç qBi?Qmi i?2 2tTHB+Bi DQBM iQ i?2 KmHiB+bi ;`QmT- 
K2Mi rBi? i?2 aQm`+2@aT2+B}+ JmHiB+bi UaaJV T`QiQ+QH `Qmi2` rBHH MQi 7Q`r`/ KmHiB+bi AS T+F2ib /2b@
(Re)X AM Qi?2` rQ`/b- i?2 QMHv T+F2ib i?i `2 /2HBp2`2/ iBM2/ iQ 2M/ ?QbibX h?Bb T`Q+2bb BKTHB2b i?2 /Bbi`B@
iQ  `2+2Bp2` `2 i?Qb2 Q`B;BMiBM; 7`QK  bT2+B}+ bQm`+2 #miBQM Q7 i?2 +QMb2Mi iQ DQBM i?2 KmHiB+bi ;`QmT
//`2bb `2[m2bi2/ #v i?2 bK2 `2+2Bp2`X >2M+2- aaJ KQM; /2pB+2b- BM+`2bBM; i?2 bB;MHBM; Qp2`?2/X
Bb T`iB+mH`Hv r2HH@bmBi2/ iQ /Bbb2KBMiBQM@bivH2 TTHB@
+iBQMb rBi? QM2 Q` KQ`2 b2M/2`b r?Qb2 B/2MiBiB2b `2 Ç h?2`2 Bb MQ rv 7Q` i?2 b2M/2` iQ +QMi`QH r?Q bm#@
FMQrM #27Q`2 i?2 TTHB+iBQM #2;BMbX b+`B#2b iQ  KmHiB+bi ;`QmTX
LQM@AS KmHiB+biBM;X h?2 /2bB;M Q7 KmHiB+bi Ç Ai T`2p2Mib i?2 +`2iBQM Q7 /Bb+`BKBMiBQM TQHB+B2b
bQHmiBQMb ?b HbQ #22M BMp2biB;i2/ #2vQM/ ASX AM #b2/ QM i?2 /2biBMiBQMb Q7 i?2 BM7Q`KiBQM rBi?BM
TTHB+iBQM@Hv2` bQHmiBQMb- ;`QmT K2K#2`b?BT- KmHiB@ i?2 bK2 KmHiB+bi ;`QmTX h?2`27Q`2- r?2M  HBK@
+bi /2HBp2`v bi`m+im`2 +QMbi`m+iBQM- M/ /i 7Q`r`/@ BiiBQM iQ i?2 /Bbi`B#miBQM Q7 T+F2ib iQ bQK2 2MiB@
BM; `2 2t+HmbBp2Hv +QMi`QHH2/ #v T`iB+BTiBM; 2M/ ?Qbib- iB2b Q7 i?2 bK2 KmHiB+bi ;`QmT Bb M22/2/- MQi?2`
i?mb- i?2 bmTTQ`i Q7 M2irQ`F MQ/2b Bb MQi M22/2/ (RR)X KmHiB+bi ;`QmT Kmbi #2 +`2i2/- rBi?  +QMb2[m2Mi
AM i?2 +H2M@bHi2 7mim`2 AMi2`M2i JQ#BHBiv6B`bi `+?Bi2+@ BM+`2b2 BM i?2 MmK#2` Q7 bB;MHBM; T+F2ib BM i?2
im`2 (Rd)-  +QMi2ti@r`2 /2HBp2`v T`BKBiBp2 Bb T`QTQb2/- M2irQ`FX
r?B+? ;2M2`HBx2b KmHiB+bi iQ ;`QmTb 2bi#HBb?2/ QM i?2
Ç HH `Qmi2`b Kmbi #2 `2TH+2/ rBi? KmHiB+bi@2M#H2/
#bBb Q7 ii`B#mi2@#b2/ /2b+`BTiQ`bX h?2 MK2 b2`pB+2-
`Qmi2`b- r?B+? +QmH/ #2 2tT2MbBp2 M/ ?`/Hv pB#H2
BM +?`;2 Q7 `2bQHmiBQM T`Q+2/m`2b #2ir22M ;HQ#H mMB[m2
7Q` i?2 M2irQ`F QT2`iQ`- `BbBM; BMi2`QT2`#BHBiv Bb@
B/2MiB}2` U:lA.V M/ M2irQ`F //`2bb2b- KBMiBMb 
bm2bX
K2K#2`b?BT b2i i?i +QMbBbib Q7 HH :lA.b Q7 /2pB+2b
i?i `2 bm#b+`B#2/ iQ i?2 KmHiB+bi ;`QmTX h?2 b2M/2` h?2 TQQ` ~2tB#BHBiv Q7 i?2 AS@#b2/ KmHiB+bi /Bb+Qm`@
Bb `2bTQMbB#H2 7Q` b2M/BM; /i iQ 2+? Q7 i?2 `2im`M2/ ;2b i?2 Tm`bmBi Q7 bm+? M TT`Q+? 7Q` i?2 rB/2 p`B@
//`2bb2bX 2iv Q7 b2M/2`@BMBiBi2/ /vMKB+ ;`QmT@#b2/ +QKKmMB@
a.L@#b2/ KmHiB+biBM;X a.L +M bBKTHB7v KmHiB@ +iBQMb- b /2KM/2/ #v 7mim`2 AQh /2THQvK2MibX
+bi i`{+ 2M;BM22`BM; i?MFb iQ i?2 +2Mi`HBx2/ Mim`2 PM i?2 Qi?2` ?M/- TTHB+iBQM@Hv2` bQHmiBQMb ?p2 i?2
Q7 i?2 M2irQ`F +QMi`QH THM2X *m``2Mi KmHiB+bi bQHm@ /`r#+F Q7  /2}MBi2Hv rQ`b2 T2`7Q`KM+2 BM i2`Kb Q7
iBQMb 2KTHQv  b?Q`i2bi@Ti? i`22 iQ +QMM2+i i?2 bQm`+2 2M/@iQ@2M/ Hi2M+v M/ 2{+B2M+v +QKT`2/ iQ AS KmHiB@
iQ i?2 `2+2Bp2`b r?B+? Bb #mBHi ++Q`/BM; iQ HQ+H BM7Q`@ +biX h?Bb Bb #2+mb2 2M/ ?Qbib ?p2 HBiiH2 Q` MQ FMQrH@
KiBQMX h`{+ 2M;BM22`BM; Bb /B{+mHi iQ #2 bmTTQ`i2/ 2/;2 Q7 i?2 mM/2`HvBM; M2irQ`F iQTQHQ;vX
BM  b?Q`i2bi@Ti? i`22X "v miBHBxBM; i?2 ;HQ#H pB2r Q7 h?MFb iQ Bib T`Q;`KK#BHBiv M/ ;HQ#H FMQrH2/;2 Q7
i?2 a.L +QMi`QHH2`- BM (R3) HH i?2 TQbbB#H2 `Qmi2b #2@ i?2 iQTQHQ;v- a.L +M KF2 KQ`2 2{+B2Mi i?2 +`2iBQM
ir22M i?2 bQm`+2b M/ 2+? ?Qbi Q7 i?2 KmHiB+bi ;`QmT Q7 i?2 KmHiB+bi i`22- BKT`QpBM; 7Q`r`/BM; T`Q+2/m`2bX
`2 +H+mHi2/ BM /pM+2X AM +QMi`bi rBi? AS KmHiB@ >Qr2p2`- iQ i?2 #2bi Q7 Qm` FMQrH2/;2- i?2 ~2tB#BHBiv
+bi- i?2`2 `2 MQ /2 7+iQ biM/`/b 7Q` a.L KmHiB+bi Q7 a.L ?b MQi #22M BMp2biB;i2/ iQ KM;2 /vMKB+
`QmiBM;X .Bz2`2Mi TT`Q+?2b i`;2iBM; /Bz2`2Mi QTiB@ ;`QmT 7Q`KiBQMX
KBxiBQM Q#D2+iBp2b +M #2 i`;2i2/ BM  ~2tB#H2 KMM2` h?2b2 Bbbm2b ?p2 KQiBpi2/ i?2 i?2Q`BxBM; Q7  M2r +QK@
M/ Bi Bb mMHBF2Hv i?i  ;Bp2M TT`Q+? Bb ;QBM; iQ #2 KmMB+iBQM K2i?Q/ M/ /i /2HBp2`v b+?2K2 (j)- #H2 iQ
/QKBMMiX a.L KmHiB+bi Bb 2M#H2/ #v r`BiBM; M T@ #2ii2` }i i?2 Mim`2 Q7 mT+QKBM; ;`QmT@#b2/ +QKKm@
THB+iBQM 7Q` i?2 a.L +QMi`QHH2` i?i QTiBKBx2b i?2 i`7@ MB+iBQMb, aQ+BQ+biX
}+ ~Qrb iQ K22i i?2 T`iB+mH` M22/b Q7 i?2 2M/ mb2` h?Bb Bb BMi`Q/m+2/ b  MQp2H M/ ~2tB#H2 bQHmiBQM i?i
(Rk)X h?2 a.L +QMi`QHH2` +M #mBH/ i?2 KmHiB+bi i`22 HHQrb ;`QmT@#b2/ +QKKmMB+iBQMb BM i?2 AQh 2M?M+2/
iQ K22i HBMF +QMbi`BMib U#M/rB/i? +QMbmKTiBQMV Q` rBi? i?2 MQiBQM Q7 bQ+BH iB2bX Ai BM?2`Bib i?2 bi`2M;i?b Q7
Ti? +QMbi`BMib U2M/@iQ@2M/ /2HvV (Rj)X >2M+2- Bi Bb AS KmHiB+bi- BM i?i Bi H2ib M2irQ`F MQ/2b /Bbb2KBMi2
 pHm#H2 bQHmiBQM r?2M ZmHBiv Q7 a2`pB+2 UZQaV `2@ T+F2ib BM M 2{+B2Mi KMM2`, bQ+BQ+bi T+F2ib `2 b@
[mB`2K2Mib M22/ iQ #2 2Mbm`2/ iQ  KmHiB+bi ~Qr- 2X;X- bB;M2/ M AS //`2bb iQ 7+BHBii2 i?2B` 7Q`r`/BM;X AM

© International Telecommunication Union, 2020 39

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

//BiBQM- i?2 T`QTQbH BM (j) 2M#H2b  KmimH +QMi`QH Q7 +BQ+bi i?Bb i`MbHi2b BMiQ i?2 TQbbB#BHBiv Q7 2{+B2Mi M/
i?2 2M/TQBMibX LQi QMHv i?2 `2+2Bp2` +M }Hi2` /Bz2`2Mi ~2tB#H2 ;`QmT 2M/@TQBMib /Bb+Qp2`v- M BMi`BMbB+ TQbbB#BH@
b2M/2`b- b BM aaJ- #mi HbQ i?2 b2M/2` +M UBKTHB+BiHvV Biv Q7 BKTH2K2MiBM; TQHB+B2b 7Q` +`2iBM; i`mbi2/ ;`QmTb
/2+B/2 r?B+? MQ/2 b?QmH/ #2HQM; iQ i?2 b2i Q7 BMi2M/2/ Q7 2M/@TQBMib /B`2+iHv i i?2 M2irQ`F H2p2H- M/ i?2 #BH@
`2+2Bp2`b- #v bT2+B7vBM; i?2 72im`2b UBM i2`Kb Q7 bQ+BH Biv iQ 2z2+iBp2Hv M/ bBKTHv /2H rBi? i?2 T`Q#H2K Q7
`2HiBQMb?BTV Q7 bm+? `2+2Bp2`bX h?2 #Qp2 +T#BHBiB2b BMi2`QT2`#BHBiv KQM; /Bz2`2Mi AQh THi7Q`KbX
`2 /Bb`mTiBp2 r?2M +QKT`2/ iQ +QMp2MiBQMH AS@#b2/ P#pBQmbHv- iQ /Q i?Bb r2 M22/ iQ bi`i 7`QK  T`/B;K
KmHiB+biX aQ+BQ+bi `2HB2p2b i?2 #m`/2M Q7 ;`QmT KM@ i?i +M T`QpB/2 7Q` i?2 2bi#HBb?K2Mi Q7 Tb2m/Q@bQ+BH
;2K2Mi 7`QK M2irQ`F MQ/2b M/ Q7 2tTHB+Bi DQBM T`Q+2@ iB2b #2ir22M /2pB+2b UiQ QT2`i2 i i?2 M2irQ`F Hv2`VX
/m`2b 7`QK /2pB+2bX h?Bb Bb H`2/v pBH#H2 BM bQHmiBQMb Q7 dzbQ+BH M2irQ`Fb
JQ`2Qp2`- a.L Bb +?Qb2M iQ 7+BHBii2 i?2 BKTH2K2Mi@ Q7 AQh /2pB+2bǴ- bm+? b i?2 aQ+BH AMi2`M2i Q7 h?BM;b
iBQM Q7 KmHiB+bi ;`QmTb rBi?  bQ+BH ~pQm` /B`2+iHv i UaAQhV (jy) 7Q` 2tKTH2X >Qr2p2`- i?2v M22/ iQ #2
i?2 M2irQ`F Hv2`X KQp2/ 7`QK i?2 TTHB+iBQM Hv2`- r?2`2BM i?2v ?p2
#22M BMBiBHHv +QM+2Bp2/- /QrM iQ i?2 +QMi`QH THM2 Q7
kXj /pMi;2b Q7  dzbQ+BH@Q`B2Mi2/Ǵ T@ i?2 M2irQ`F Hv2`X AM bQ /QBM;- ;`QmT 2bi#HBb?K2Mi M/
T`Q+? /i 2t+?M;2 KQM; K2K#2`b Q7 bm+? ;`QmTb +M #2
KM;2/ BM  iB;?i2` rv- rBi? BM?2`2Mi ~2tB#BHBiv M/
h?2 mb2 Q7 bQ+BH HBMFb iQ bmTTQ`i M2irQ`F 7mM+iBQMHBiv 2{+B2M+v BM i2`Kb Q7 M2irQ`F `2bQm`+2 mb;2X
Bb MQi M2r BM i?2 i2H2+QKKmMB+iBQMb HM/b+T2X
a2p2`H `QmiBM; T`QiQ+QHb BM rB`2H2bb / ?Q+ (ky)- KQ#BH2 jX aP*AP*ah, P"C1*hAo1a L.
QTTQ`imMBbiB+ M/ /2Hv@iQH2`Mi M2irQ`Fb (kR- kk- kj-
.1aA:L S_AL*ASG1a
k9)- ?p2 #22M /2bB;M2/ iQ #mBH/ mTQM i?2 F2v +QM+2Tib
Q7 bQ+BH M2irQ`F MHvbBb- BX2X- bKHH rQ`H/ T?2MQK2MQM AM i?Bb a2+iBQM r2 /2b+`B#2 ?Qr r2 +M +?B2p2  `2H
(k8) M/ +2Mi`HBivX h?2 7Q`K2` QM2- XFXX +QKKmMBiv- BKTH2K2MiiBQM Q7 i?2 aQ+BQ+bi +QM+2Ti #v `2HvBM; QM
+Tim`2b i?2 7+i i?i +iQ`b rBi?BM  bQ+BH M2irQ`F `2 i?2 +T#BHBiB2b Q7 i?2 aQ7ir`2 .2}M2/ L2irQ`FBM;
b2T`i2/ 7`QK 2+? Qi?2` #v M p2`;2 MmK#2` Q7 7B`Hv T`/B;KX h?2 `2bmHiBM; bQHmiBQM Bb M 2M#H2` 7Q` ;`QmT
HBKBi2/ ?QTbX h?2 Hii2` QM2 b?Qrb i?i bQK2 MQ/2b BM +QKKmMB+iBQMb #b2/ QM bQ+BH MQiBQMb i i?2 M2irQ`F
 +QKKmMBiv `2 i?2 +QKKQM +[mBMiM+2b Q7 Qi?2` Hv2`X
MQ/2bX aQ+BH iB2b KQM; /2pB+2bX .2pB+2b `2 HBF2Hv iQ BMi2`@
AM i?2 7Q`2K2MiBQM2/ rQ`Fb- i?2 FMQrH2/;2 Q7 bQ+BH +i rBi? Qi?2` /2pB+2b rBi? bBKBH` T`Q}H2b M/ ?#Bib-
+?`+i2`BbiB+b U2X;X- MQ/2 +2Mi`HBiv- BM@#2ir22MM2bbV Bb 2X;X- i?Qb2 HQ+i2/ BM i?2 bK2 TH+2- QrM2/ #v i?2 bK2
mb2/ iQ KF2 #2ii2` 7Q`r`/BM; /2+BbBQMb M/ bbBbi i?2 mb2`- T`Q/m+2/ BM i?2 bK2 +QKTMv #`M+?X
`2Hv b2H2+iBQM r?2M /2HBp2`BM; /i iQ i?2 BMi2M/2/ /2b@ am+? iB2b `2 r2HH +Tim`2/ #v i?2 aAQh T`/B;K BM (jy)X
iBMiBQMUbVX h?2`2-  72r #bB+ ivT2b Q7 bQ+BH `2HiBQMb?BTb- /2}M2/
JMv Q7 i?2 bim/B2/ TT`Q+?2b BMpQHp2/ mMB+bi Q` KmH@ ++Q`/BM; iQ mb2`@/2}M2/ TQHB+B2b- `2 BMi`Q/m+2/, +Q@
iB+bi +QKKmMB+iBQMb (ke- kd- k3)X h?2 Bbbm2 Q7 /i QrM2`b?BT Q#D2+i `2HiBQMb?BT UPP_V- +`2i2/ #2ir22M
#`Q/+biBM; BM  JQ#BH2 aQ+BH L2irQ`F- r?2`2 KQ#BH2 /2pB+2b i?i #2HQM; iQ i?2 bK2 QrM2`c +Q@HQ+iBQM Q#D2+i
bQ+BH mb2`b T?vbB+HHv BMi2`+i rBi? 2+? Qi?2`- Bb M@ `2HiBQMb?BT U*GP_V- +`2i2/ #2ir22M biiBQM`v /2pB+2b
Hvx2/ BM (kN)X HQ+i2/ BM i?2 bK2 TH+2c T`2MiH Q#D2+i `2HiBQMb?BT
h?2 Q#D2+iBp2 Q7 i?Bb rQ`F Bb iQ 2tTHQBi bBKBH` +QM+2Tib USP_V- +`2i2/ #2ir22M /2pB+2b Q7 i?2 bK2 KQ/2H- T`Q@
#mi mM/2`  /Bz2`2Mi T2`bT2+iBp2X q2 BK MQi iQ BK@ /m+2` M/ T`Q/m+iBQM #i+?c +Q@rQ`F Q#D2+i `2HiBQMb?BT
T`Qp2 7Q`r`/BM; /2+BbBQMb #v H2p2`;BM; bQ+BH M2irQ`F U*qP_V- +`2i2/ #2ir22M KQpBM; /2pB+2b i?i K22i
T`QT2`iB2b- #mi iQ #2ii2` /Bbb2KBMi2 /i i i?2 M2i@ 2+? Qi?2` i i?2 QrM2`bǶ rQ`FTH+2c bQ+BH Q#D2+i `2@
rQ`F Hv2` rBi?BM /vMKB+HHv +`2i2/ ;`QmTb Q7 bQ+BHHv HiBQMb?BT UaP_V- +`2i2/ b  +QMb2[m2M+2 Q7 7`2[m2Mi
+QMM2+i2/ /2pB+2bX BMi2`+iBQMb #2ir22M KQpBM; /2pB+2bX h?2 7`K2rQ`F
h?2 T`QTQbH ?b i?2 TQi2MiBH Q7  `2H ;K2 +?M;2` Bb [mBi2 ~2tB#H2 M/ Qi?2` ivT2b Q7 `2HiBQMb?BTb +M #2
BM pB2r Q7 i?2 +`2iBQM Q7 i?2 7mim`2 AMi2`M2i Q7 h?BM;b- 2bBHv //2/ QM  T2` mb2@+b2 #bBbX
#v T`QpB/BM; bmT2`BQ` /pMi;2b +QKT`2/ iQ r?i ?b TTHB+iBQMb `2[mB`BM; /i /Bbb2KBMiBQM iQ  bQ+BH
#22M /QM2 bQ 7` BM i?2 HBi2`im`2X ;`QmT Q7 /2pB+2b `2- 7Q` BMbiM+2- bQ7ir`2 mT/i2b, 
AM 7+i- bQ+BH #QM/b MQi QMHv 2Mbm`2 KBMBKmK b2T`@ ;Bp2M bQ7ir`2 Ti+? M22/b iQ #2 b72Hv /2HBp2`2/ iQ HH
iBQM /BbiM+2b #2ir22M +iQ`b- +`m+BH 7Q` 2{+B2Mi M/ i?2 /2pB+2b Q` b2MbQ`b Q7 i?2 bK2 #`M/- KQ/2H- #i+?X
7bi /i T`QT;iBQM- #mi Kv 2M#H2 /i 2t+?M;2 6Q` i?Bb- SP_ `2HiBQMb?BTb b?QmH/ #2 2tTHQBi2/X aBKB@
rBi?BM i`mbi2/ ;`QmTb M/ +`2iBQM Q7 ;`QmTb i?i BM@ H`Hv- bQK2 /i M22/b iQ `2+? HH Qi?2` /2pB+2b #2HQM;@
+Hm/2 +iQ`b #2HQM;BM; iQ /Bz2`2Mi +QKKmMBiB2bX AM aQ@ BM; iQ i?2 bK2 QrM2` BM i?2 +b2 Q7 T2`bQMH #m##H2b, i?2

40 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

PP_ `2HiBQMb?BT Bb TT`QT`Bi2 BM i?Bb b+2M`BQX "mbB@ "v Qp2`bi2TTBM; i?2 ;MQbiB+BbK #Qmi aQ+BQ+bi i`7@
M2bb b2`pB+2b Kv #2 /p2`iBb2/ iQ HH /2pB+2b i?i 2Bi?2` }+ i i?2 M2irQ`F Hv2`- i?2 7QHHQrBM; /pMi;2b `2
`2 +m``2MiHv BM i?2 bK2 `2 U*GP_V Q` Q7i2M pBbBi2/ 2tT2+i2/,
i?2 bK2 TH+2 UaP_VX
h`;2i2/ /i /2HBp2`v b+?2K2bX aQ+BQ+bi BKb iQ Ç /i 7Q`r`/BM; +M Q++m` BM M 2{+B2Mi KMM2`-
2M#H2, 2X;X- #v `2/m+BM; i?2 MmK#2` Q7 /mTHB+i2/ T+F2ib-
M/ bpBM; #M/rB/i? ++Q`/BM;Hvc
Ç  ;Bp2M b2M/2` iQ /Bbb2KBMi2 /i BM  Tmb?@HBF2
KMM2` iQ bT2+B}+ MQ/2b- r?B+? `2 7`B2M/b Qp2`  Ç }Hi2`BM; T`Q+2/m`2b +M #2 2M7Q`+2/ BM@M2irQ`F- b
bQ+BH M2irQ`F Q7 /2pB+2b- ++Q`/BM; iQ T`QT2`Hv /2@ `2[m2bi2/ #v TQi2MiBH /i `2+BTB2Mib- iQ HBKBi i?2
}M2/ }Hi2`b M/ TQHB+B2b UBX2X- i?2 bQ+BH `2HiBQMb?BT KbbBp2 KQmMi Q7 ;2M2`i2/ i`{+c
ivT2Vc Ç M2irQ`F QT2`iQ`b +M #2M2}i 7`QK i`{+ `2/m+@
Ç  MQ/2 iQ bm#b+`B#2 iQ bT2+B}+ bQ+BH@#b2/ iQTB+b iBQM- r?B+? Bb T`iB+mH`Hv +`m+BH 7Q` i?2B` BM7`b@
UBX2X- iQ `2+2Bp2 /i 7`QK 7`B2M/b Q7  ;Bp2M ivT2Vc i`m+im`2b 2tT2+i2/ iQ #2 H`;2Hv Qp2`r?2HK2/ BM i?2
M2` 7mim`2X
Ç  MQ/2 iQ T`BQ`BiBx2 UM/ MQi iQ `2+2Bp2V /i 7`QK
S`Q;`KK#H2 T+F2i i`2iK2MiX _2+2Mi /pM+2@
T`iB+mH` b2M/2`b- 2X;X, iQ 2M7Q`+2 ZQac iQ B/2MiB7v
K2Mib BM M2irQ`FBM; i2+?MQHQ;B2b KF2 i?2 /2THQvK2Mi
i?2 KQ`2 bmBi2/ M/ i`mbi2/ +QKKmMB+iBQM 2M/@
Q7 aQ+BQ+bi i i?2 M2irQ`F Hv2` 2p2M KQ`2 pB#H2X q2
TQBMib 7Q` b2+m`Biv `2bQMbc iQ bp2 `2bQm`+2bX
B/2MiB7v a.L b i?2 F2v 2M#H2` 7Q` aQ+BQ+biX h?MFb
.2THQvK2Mi QTiBQMbX hQ i`;2i i?2 7Q`2K2MiBQM2/ iQ Bib T`Q;`KK#BHBiv- r?B+? `2/m+2b i?2 +QKTH2tBiv Q7
Q#D2+iBp2b- i?2 2MpBbBQM2/ 7`K2rQ`F ?b UBV iQ 2M#H2 M2irQ`F 2H2K2Mib- a.L +M BMD2+i 7Q`r`/BM;f/`QTTBM;
MQ/2b iQ BM/B+i2 BM M ;BH2 KMM2` i?2 72im`2b Q7 i?2 `mH2b M/ T`QT2`Hv KMBTmHi2 ?2/2`b Q7 T+F2ib iQ
2M/TQBMib Q7 /i ~Qrb UBX2X- i?2 b2i Q7 BMi2M/2/ `2+BT@ KF2 KQ`2 2{+B2Mi i?2B` 7Q`r`/BM;X
B2Mib M/fQ` i?2 mi?Q`Bx2/ b2M/2`bV #b2/ QM i?2 /Bb@ am+? TQHB+B2b +M #2 /2}M2/ BM  M2irQ`F TTHB+iBQM-
iM+2 BM  bQ+BH M2irQ`F Q7 /2pB+2b- UBBV iQ T`QT2`Hv M/ rBi? MQ M22/ iQ KQ/B7v i?2 /i THM2 Q7 i?2 mM/2`HvBM;
/vMKB+HHv B/2MiB7v i?2K- UBBBV iQ 7Q`r`/ /i T+F2ib M2irQ`F BM7`bi`m+im`2X
++Q`/BM;HvX
 bi`B;?i7Q`r`/ TT`Q+? iQ ++QKTHBb? i?2 }`bi irQ 9X h>1 _*>Ah1*hl_G 6_J1@
72im`2b +QmH/ #2 iQ `2Hv QM M TTHB+iBQM@Hv2` bQHm@ qP_E
iBQMX 6Q` BMbiM+2- i?2 BMi2M/2/ b2i Q7 `2+2Bp2`b +M #2
bT2+B}2/ #v  ;Bp2M b2M/2` i  ?B;?@H2p2H- 2X;X- i?`Qm;? h?2 KBM 2MiBiB2b Q7 i?2 2MpBbBQM2/ 7`K2rQ`F `2, i?2
K2i/iX h?2M- i?2 `2bmHiBM; `2[m2bi +M #2 b2Mi iQ  aQ+BQ+bi MQ/2b- i?2 a.L M2irQ`F U2M+QKTbbBM; #Qi?
Tm`TQb2@#mBHi T`Qtv r?B+? Bb BM +?`;2 Q7 KTTBM; bm+? brBi+?2b M/ +QMi`QHH2`V- m;K2Mi2/ rBi? i?2 MQiBQM Q7
/i QMiQ AS //`2bb2b Q7 i?2 `2+2Bp2`b- bBKBH` iQ (Rd)X aQ+BQ+bi- M/ i?2 aQ+BQ+bi _2HiBQMb?BT a2`pB+2- b
.2bTBi2 i?2 pB`im2 Q7 bBKTHB+Biv- bm+? M TT`Q+? ?b b?QrM BM 6B;m`2 R M/ /2iBH2/ BM i?2 7QHHQrBM; b2+iBQMbX
i?2 /`r#+F Q7 TQQ` T2`7Q`KM+2 BM i2`Kb Q7 2{+B2M+v
BM i?2 mb;2 Q7 M2irQ`F `2bQm`+2b- bBM+2 /i 7Q`r`/BM; 9XR h?2 aQ+BQ+bi MQ/2b
iQ 2+? BMi2M/2/ /2biBMiBQM Bb T2`7Q`K2/ i i?2 mM/2`@ h?2 aQ+BQ+bi MQ/2b `2 i?2 2M/TQBMib Q7  aQ+BQ+bi
HvBM; M2irQ`F Hv2` BM  KvQTB+ KMM2`X +QKKmMB+iBQMX h?2v `2 H2;+v AQh /2pB+2b U2X;X-
h?mb- Qm` BMi2`2bi Bb QM  M2irQ`F@Hv2` TT`Q+?- +@ bK`iT?QM2b- b2MbQ`bV m;K2Mi2/ rBi? i?2 aQ+BQ+bi
+Q`/BM; iQ r?B+? i?2 72im`2b Q7 i?2 BMi2M/2/ b2i Q7 `2@ amTTQ`i Gv2` Ua+aGV `mMMBM; QM iQT Q7 i?2 i`MbTQ`i
+2Bp2`b Q7  ;Bp2M /i T+F2i UQ` Q7  b2M/2` Q7 mMrMi2/ Hv2`- i?`Qm;? r?B+? i?2v `2 2M#H2/ iQ +`2i2- b2M/
/i T+F2ibV +M #2 i`MbHi2/ BMiQ  M2irQ`F@Hv2` AS M/fQ` `2+2Bp2 aQ+BQ+bi T+F2ibX h?2 a+aG 2tTQb2b i?2
//`2bb- ?2M+2 i`2i2/ U7Q`r`/2/f/`QTT2/V #v M2irQ`F aQ+BQ+bi TTHB+iBQM S`Q;`KKBM; AMi2`7+2b USAbV iQ
MQ/2b- ++Q`/BM;HvX h?2 ivT2 Q7 T`QTQb2/ TT`Q+? Bb BM@ i?2 TTHB+iBQMb i?i rMi iQ mb2 i?2 aQ+BQ+bi +QKKm@
bTB`2/ #v i?2 i`/BiBQMH AS@#b2/ KmHiB+bi- rBi? r?B+? MB+iBQM +QM};m`iBQM 7Q` /i /2HBp2`vX Ai Bb i?`Qm;?
Bi b?`2b  72r bT2+ib- bm+? b i?2 `QmiBM; Q7 T+F2ib i?Bb Hv2` i?i aQ+BQ+bi T+F2ib `2 +`2i2/ M/ `2+2Bp2/
rBi?  KmHiB+bi //`2bb U aQ+BQ+bi //`2bb- BM Qm` #v i?2 2M/ /2pB+2bX
+b2VX >Qr2p2`- KmHiB+bi H+Fb i?2 ~2tB#BHBiv M2+2bb`v
iQ BKTH2K2Mi i?2 7Q`2K2MiBQM2/ +`BiB+H 7mM+iBQMHBiB2b
9Xk h?2 a.L M2irQ`F
7Q` i?2 7mim`2 AMi2`M2i Q7 #BHHBQMb Q7 /2pB+2b- r?BH2 K22i@
BM; i?2 `2[mB`2K2Mib Q7 i?2 2M/ mb2`b M/ i?Qb2 Q7 i?2 h?2 a.L M2irQ`F Bb +QKTQb2/ Q7 i?`22 /Bz2`2Mi THM2b-
M2irQ`F QT2`iQ`bX ++Q`/BM; iQ i?2 H2;+v /2THQvK2MiX

© International Telecommunication Union, 2020 41

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

6B;X R Ĝ aQ+BQ+bi `+?Bi2+im`H 7`K2rQ`FX

h?2 /i THM2 2M+QKTbb2b i?2 a.L brBi+?2bX h?2v MQ/2b i?i #2HQM; iQ i?2 b2i Q7 i?2 TQi2MiBH `2+BT@
`2 a.L@2M#H2/ M2irQ`F MQ/2b r?B+? `2 +QMM2+i2/ iQ B2Mibfmi?Q`Bx2/ b2M/2`b Q7  aQ+BQ+bi T+F2iX
2+? Qi?2` M/ BMi2`+i rBi? i?2 a.L +QMi`QHH2`X "2@
ir22M i?2K- i?2 a.L ;i2rvb `2 i?2 BM;`2bbf2;`2bb >2`2BM-  KDQ` 2H2K2Mi Bb i?2 aQ+BH oB`imH LQ/2
MQ/2b Q7 i?2 a.L M2irQ`FX a.L MQ/2b BMi2`+i rBi? UaoLV- r?B+? `2T`2b2Mib i?2 /B;BiH +QmMi2`T`i Q7 
i?2 a.L +QMi`QHH2` i?`Qm;? i?2 PT2M6HQr UP6V bQmi?@ T?vbB+H /2pB+2X Ai biQ`2b bQK2 K2i/i T`QpB/BM; BM@
#QmM/ BMi2`7+2X 7Q`KiBQM #Qmi i?2 Mim`2 Q7 i?2 /2pB+2 M/  HBbi Q7
h?2 +QMi`QH THM2 BM+Hm/2b i?2 a.L +QMi`QHH2`- r?B+? 7`B2M/b- r?B+? Bb Q`;MBx2/ BM  i#H2 MK2/ 6`B2M/b h@
Qp2`b22b i?2 a.L MQ/2b- ++Q`/BM; iQ bT2+B}+ Q`+?2bi`@ #H2X 6Q` 2+? 7`B2M/ BM i?2 i#H2- i?2 aoL `2+Q`/b i?2
iBQM TQHB+B2b /2}M2/ i i?2 TTHB+iBQM THM2X Ai i`+Fb ivT2UbV Q7 7`B2M/b?BTUbV- /2}M2/ ++Q`/BM; iQ i?2 aAQh
i?2 ;`T? Q7 i?2 M2irQ`F iQTQHQ;v BM i?2 L2irQ`F AM@ T`/B;K M/ i?2 i`mbi H2p2H bbQ+Bi2/ rBi? 2+? 7`B2M/X
7Q`KiBQM "b2 ULA"VX ++Q`/BM; iQ BM7Q`KiBQM BM i?2 h?2 aQ+BH oB`imH LQ/2 _2TQbBiQ`v UaoL_V biQ`2b HH
LA" M/ TQHB+B2b /2}M2/ #v M2irQ`F TTHB+iBQMb- Bi BM@ aoLb bbQ+Bi2/ iQ i?2 T?vbB+H /2pB+2b BM  ;Bp2M `2X
D2+ib `mH2b BM i?2 bQ@+HH2/ ~Qr i#H2b Q7 a.L MQ/2b iQ AM/22/- QM2 aoL_ Bb `2bTQMbB#H2 7Q` T`QpB/B/BM; i?2 /2@
2M#H2 i?2 7Q`r`/BM; Q7 bQ+BQ+bi T+F2ib i?`Qm;? P6 b+`B#2/ b2`pB+2b 7Q` i?2 Q#D2+ib BM  ;Bp2M `2c KQ`2
K2bb;2b (jR)X aoL_b `2 i?2M BMi2`+QMM2+i2/ BM  /Bbi`B#mi2/ bvbi2KX
h?2 7QHHQrBM; KQ/mH2b `2 bbQ+Bi2/ iQ i?2 aoL_X
9Xj h?2 aQ+BQ+bi _2HiBQMb?BT a2`pB+2
Ç h?2 _2HiBQMb?BT JM;2` U_JV Bb `2bTQMbB#H2 7Q`
h?2 aQ+BQ+bi _2HiBQMb?BT a2`pB+2 Ua_aV Bb BKTH2@ i?2 `2HiBQMb?BTbǶ HB72+vH2 KM;2K2Mi- BX2X- /2i2+i@
K2Mi2/ i i?2 TTHB+iBQM THM2- M2ti iQ +QMp2MiBQMH BM;- +`2iBM;- mT/iBM; M/ /2H2iBM; `2HiBQMb?BTbR X
a.L TTHB+iBQMb- M/ Bi T`QpB/2b i?2 7QHHQrBM; KBM
7mM+iBQMHBiB2b, Ç h?2 _2HiBQMb?BT "`Qrb2` U_"V MpB;i2b i?2
6`B2M/b h#H2 iQ }M/ TQi2MiBH `2+BTB2Mib Q7  aQ+BQ@
RX 2bi#HBb?BM; bQ+BH `2HiBQMb?BTb #2ir22M MQ/2bX +bi T+F2i- ++Q`/BM; iQ i?2B` TQbBiBQM BM i?2 bQ+BH
qBi?Qmi HQbb Q7 ;2M2`HBiv- r2 BM?2`Bi +QM+2Tib M/ M2irQ`FX SQHB+B2b 7Q` i?2 bQ+BH M2irQ`F MpB;iBQM
K2i?Q/QHQ;B2b `2;`/BM; i?2 TQHB+B2b 7Q` i?2 2bi#@ `2 /Bb+mbb2/ BM (jk)X
HBb?K2Mi Q7 i?2 bQ+BH HBMFb #2ir22M MQ/2b 7`QK i?2
r2HH@++2Ti2/ aAQh T`/B;K (jy)c Ç h?2 aQ+BQ+bi >M/H2` Ua>V- r?2M2p2` [m2`B2/ #v
i?2 a.L +QMi`QHH2`- T`QpB/2b i?2 K2K#2`b Q7 
kX F22TBM; i`+F Q7 i?2 2bi#HBb?2/ bQ+BH `2HiBQMb?BTbc aQ+BQ+bi ;`QmT- 7i2` [m2`vBM; i?2 _" KQ/mH2-
jX T`QpB/BM; BMi2`7+2b iQr`/b i?2 a.L M2irQ`F M/ R 6Q` /2iBH2/ /2b+`BTiBQM Q7 `2HiBQMb?BTb KM;2K2Mi- i?2
iQ MpB;i2 i?2 bQ+BH M2irQ`F bQ iQ B/2MiB7v i?2 `2/2` Bb `272``2/ iQ (jk)X

42 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

i?`Qm;?  _2T`2b2MiiBQMH aii2 h`Mb72` U_1ahV Ç aQm`+2 AS //`2bb, i?2 AS //`2bb Q7 i?2 bQm`+2
SAX /2pB+2X

aoL_b- HQM; rBi? `2H2pMi 7mM+iBQMHBiB2b UBX2X- _J- _" Ç .2biBMiBQM AS //`2bb,  }t2/ AS //`2bb-
M/ a>V- +M #2 /2THQv2/ b  T22`@iQ@T22` bvbi2K- 7Q` B/2MiB}2/ BM i?Bb TT2` b Ӿԅմդ - bbB;M2/ iQ aQ+BQ@
BMbiM+2 #mBH/BM; mTQM i?2 QM2 /2b+`B#2/ BM (jk)X +bi i?i HHQrb a.L ;i2rvb iQ B/2MiB7v aQ+BQ+bi
Pm` /2bB;M +?QB+2 Bb BK2/ i T`QpB/BM; M BKTH2K2M@ T+F2ibX
iiBQM Q7 a.L@#b2/ ;`QmT +QKKmMB+iBQMb #b2/ QM Ç aQ+BQ+bi h;,  k@#vi2b }2H/ i?i Bb +``B2/ BM@
 /2@7+iQ ;HQ#H AQh `2bQm`+2 /B`2+iQ`v- r?B+? Bb /Bb@ bB/2 i?2 i`MbTQ`i@Hv2` /2biBMiBQM TQ`i M/ Bb mb2/
i`B#mi2/ M/ rBi?Qmi  bBM;H2 THv2` BM +QMi`QH Q7 i?2 bvb@ iQ mMB[m2Hv B/2MiB7v i?2 ivT2 Q7 bQ+BH `2HiBQMb?BT
i2KX .B;BiH `2T`2b2MiiBQMb Q7 T?vbB+H AQh /2pB+2b rBHH M/ Qi?2` TT`QT`Bi2 }Hi2`b U2X;X- MmK#2` Q7 ?QTb-
`mM BM /Bbi`B#mi2/ b2`p2`b M/ +M +`2i2 miQMQKQmbHv TQbbB#H2 TTHB+iBQM Q7 aQ+BQ+bi- 2i+XVX h?2 2M+Q/@
bQ+BH@HBF2 `2HiBQMb?BTb rBi? 2+? Qi?2`X "b2/ QM bm+?  BM; Bb b 7QHHQrb,
/Bbi`B#mi2/ `2bQm`+2 /B`2+iQ`v- BMi2`+iBQMb U#Qi? TQBMi@
iQ@TQBMi M/ TQBMi@iQ@KmHiBTQBMiV #2ir22M AQh `2bQm`+2b Ĝ J2i/i, /2pB+2 K2i/i pBH#H2 7Q` 7m@
#2HQM;BM; iQ /Bz2`2Mi THi7Q`Kb +M #2 bi`B;?i7Q`r`/Hv im`2 TTHB+iBQMbX
2M#H2/X 1+? aoL_ UQ` ;`QmT Q7 aoL_bV +QmH/- BM 7+i- Ĝ _2HiBQM6BHi2`, ivT2 Q7 `2HiBQMb?BT U2X;X-
+QMiBM i?2 BK;2b Q7 i?2 /2pB+2b #2HQM;BM; iQ  ;Bp2M PP_- aP_- *GP_- 2i+XVX
THi7Q`K- Bi +M #2 QrM2/ M/ KBMiBM2/ #v i?2 QrM2` Ĝ 62im`2:`QmT, ivT2 Q7 aQ+BQ+bi b2`pB+2b
Q7 i?2 THi7Q`K UQ` 2p2M i?2 QrM2` Q7 i?2 ;`QmT Q7 AQh M22/2/ #v i?2 TTHB+iBQM U2X;X- :`QmT*`2@
/2pB+2bV- M/ Bi BMi2`+ib BM  T22`@iQ@T22` 7b?BQM rBi? iBQM- aQm`+26BHi2`BM;- Sm#fam#VX
Qi?2` aoL_b +QMbiBimiBM; i?2 a_aX
Ĝ _/Bmb, KtBKmK /BbiM+2- BM MmK#2` Q7
?QTb- 7`QK i?2 bQm`+2X
8X aP*AP*ah AL *hAPL
6B;X k b?Qrb bQK2 2tKTH2b Q7 aQ+BQ+bi h; +QM@
AM i?2 7QHHQrBM;- r2 /2iBH i?2 KBM bi2Tb 7Q` i?2 +`2iBQM
};m`iBQMX
Q7  aQ+BQ+bi T+F2iX h?2M- r2 /2b+`B#2 aQ+BQ+bi /i
/2HBp2`v ++Q`/BM; iQ  Tmb?@#b2/ /Bbb2KBMiBQM- Tm#@ "2BM; aQ+BQ+bi T+F2ib B/2MiB}2/ i?`Qm;? +QMp2MiBQMH
HBb?fbm#b+`B#2 T`Q+2/m`2b iQ bQ+BQ+bi ;`QmTb- b r2HH b Hv2`b j M/ 9 ?2/2` }2H/b- H2;+v Ki+?BM; `mH2b +M #2
}Hi2`BM; ++Q`/BM; iQ bQ+BQ+bi `mH2bX TTHB2/- rBi? MQ M22/ iQ `2bQ`i iQ P6 2tT2`BK2Mi2` }2H/b
(jj)X am+? /2bB;M +?QB+2b rQmH/ 7+BHBii2 i?2 /2THQvK2Mi
8XR *`2iBM;  aQ+BQ+bi T+F2i Q7 aQ+BQ+bi- r?B+? +M/B/i2b Bib2H7 b  b?Q`i@i2`K bQ@
HmiBQM iQ #2 2tTHQBi2/ #v M2irQ`F QT2`iQ`bX
 aQ+BQ+bi T+F2i Bb +`2i2/ r?2M2p2`  /2pB+2 M22/b 6Q` i?2 bF2 Q7 bBKTHB+Biv- i?2 2M+Q/BM; /2b+`B#2/ #Qp2
i?2 b2`pB+2b Qz2`2/ #v i?2 aQ+BQ+bi 7`K2rQ`F- r?B+? `272`b iQ i?2 +b2 r?2`2 i?2 ASp9 Bb mb2/X aBKBH` +QMbB/@
`2 BMi2M/2/ iQ, UBV /Bbb2KBMi2 /i BM  Tmb?@HBF2 KM@ 2`iBQMb ?QH/ 7Q` ASpe T+F2ib- 7Q` r?B+? Ki+?BM; }2H/b
M2`c UBBV BM/B+i2 i?2 bm#b+`BTiBQM iQ  aQ+BQ+bi ;`QmTc +M #2 ?M/H2/ #v P6 bBM+2 p2`bBQM RXk (jj)X
UBBBV Q` iQ }Hi2`fT`BQ`BiBx2 /i 7`QK T`iB+mH` b2M/2`bX 6Q` i?Qb2 +QMbi`BM2/ AQh /2pB+2b #2HQM;BM; iQ GQr
q?2M2p2`  T+F2i Bb +`2i2/- Bi ?b iQ BM/B+i2 r?B+? TQr2` M/ GQbbv L2irQ`Fb UGGLbV- eGQqSL UASpe
QM2 Q7 i?2b2 i?`22 ivT2b Q7 b2`pB+2 Bb `2[m2bi2/X h?2 Qp2` GQr@SQr2` qB`2H2bb S2`bQMH `2 L2irQ`FbV
#Qp2 `2 i?2 ivT2b Q7 aQ+BQ+bi b2`pB+2b bmTTQ`i2/ BM ?2/2` +QKT`2bbBQM K2i?Q/b +M #2 mb2/ (j9) Qp2` i?2
i?2 +m``2Mi BKTH2K2MiiBQM- #mi i?2 b2i Q7 aQ+BQ+bi b2`@ HBMF BMi2`+QMM2+iBM; i?2 /2pB+2b iQ i?2 a.L ;i2rvX
pB+2b +M #2 2bBHv 2ti2M/2/ BM i?2 7mim`2X 6Q` i?2 ASpe ?2/2`b- +QKT`2bbBQM K2i?Q/b Kv HbQ
G2i mb +QMbB/2`  /2pB+2- bv Ӷ- r?B+? +`2i2b  T+F2i z2+i bQm`+2 M/ /2biBMiBQM //`2bb2b- M/ i?2v p`v
rBi? /i iQ #2 b2Mi iQ  aQ+BQ+bi ;`QmTX h?2 TTHB+@ ++Q`/BM; iQ i?2 7+i i?i i?2 bQm`+2 Bb +QKKmMB+iBM;
iBQM BM Ӷ KF2b  `2[m2bi iQ i?2 a+aG pB i?2 pBH#H2 rBi? MQ/2b 2Bi?2` rBi?BM Q` QmibB/2 i?2 qSLX AM i?2
SAb- T`QpB/BM; i?2 7QHHQrBM; BM7Q`KiBQM, UBV i?2 ivT2 Hii2` +b2-  8y T2`+2Mi +QKT`2bbBQM `iBQ +M #2 biBHH
Q7 `2[m2bi2/ aQ+BQ+bi b2`pB+2c UBBV i?2 bQ+BH `2HiBQMb?BT +?B2p2/ #v H2iiBM; i?2 7mHH /2biBMiBQM //`2bb- +``vBM;
U2X;X- PP_- *GP_V ++Q`/BM; iQ r?B+? i?2 aQ+BQ+bi i?2 aQ+BQ+bi //`2bb- #2 i`MbKBii2/X
;`QmT ?b iQ #2 7Q`K2/c UBBBV i?2 bQ+BH /BbiM+2 UMmK@ h*S ?2/2` +QKT`2bbBQM 7Q` AQh b+2M`BQb (j8) Bb biBHH
#2` Q7 ?QTb Qp2` i?2 bQ+BH M2irQ`FV- r?B+? `2T`2b2Mib M QT2M Bbbm2 i i?2 biM/`/BxiBQM H2p2H (je)- MQi T`i
i?2 b+QT2 Q7 i?2 aQ+BQ+bi ;`QmTX Q7 _6* ek3k (j9)X h?2 +QKT`2bbBQM 7Q`2b22b iQ pQB/
h?2 a+aG `2+ib iQ i?2 BM+QKBM; `2[m2bi #v +`2iBM; M b2M/BM; i?2 TQ`i MmK#2`b BM 2+? T+F2i- r?B+? ?Qr@
AS T+F2i rBi? i?2 7QHHQrBM; ?2/2` }2H/b, 2p2` /Q2b MQi z2+i i?2 aQ+BQ+bi +QKKmMB+iBQMb b

© International Telecommunication Union, 2020 43

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

i?2 TQ`i MmK#2` rBi? i?2 aQ+BQ+bi h: Bb `2+QM@ `2+2Bp2 i?2 aQ+BQ+bi T+F2iX Pi?2` a.L MQ/2b- BM@
bi`m+i2/ i i?2 ;i2rvX AM/22/- /2+QKT`2bbBQM Q++m`b bi2/- `2 BMbi`m+i2/ iQ 7Q`r`/ i?2 aQ+BQ+bi T+F2i
i i?2 a.L ;i2rv H2iiBM; aQ+BQ+bi T+F2ib i`p2H rBi? iQ i?2 T?vbB+H +Q``2+i TQ`ib #v Ki+?BM; i?2 aQ+BQ@
+QMp2MiBQMH AS ?2/2` }2H/b BM i?2 a.L M2irQ`FX aBK@ +bi }2H/b pHm2bX
BH` QT2`iBQMb `2 T2`7Q`K2/ i i?2 a.L ;i2rvb i?2
/2biBMiBQMb `2 ii+?2/ iQ- B7 i?2 Hii2` QM2b #2HQM; iQ PM+2 i?2 aQ+BQ+bi ;`QmT Bb +`2i2/- bm#b2[m2Mi aQ+BQ@
 qSLX +bi T+F2ib i`MbKBii2/ #v i?2 bQm`+2 /2pB+2 Kv #2
?M/H2/ #v i?2 a.L ;i2rv rBi? MQ M22/ iQ +QMi+i
8Xk Smb?@#b2/ /i /Bbb2KBMiBQM i?2 a.L +QMi`QHH2`- #mi `i?2` 7Q`r`/2/ ++Q`/BM; iQ
`mH2b H`2/v pBH#H2 BM i?2 ~Qr i#H2X ++Q`/BM; iQ
PM+2 i?2 aQ+BQ+bi T+F2i Bb +`2i2/ rBi? /i iQ #2 i?2 H2;+v a.L BKTH2K2MiiBQM-  iBK2Qmi Bb TTHB2/
/Bbb2KBMi2/- Bi Bb b2Mi #v i?2 bQm`+2 /2pB+2 M/ i`2i2/ iQ `mH2b BMD2+i2/ #v i?2 +QMi`QHH2` BMiQ a.L MQ/2b- iQ
BM i?2 M2irQ`F i?`Qm;? i?2 7QHHQrBM; bi2TbX T`2p2Mi  `mH2 iQ biv BM i?2 i#H2 7Q`  HQM; iBK2 M/
mMM2+2bb`BHv Q++mTv bT+2 BM i?2 ~Qr i#H2 (jj)X qBi?BM
RX h?2 aQ+BQ+bi T+F2i `2+?2b i?2 a.L ;i2rv- Qm` 7`K2rQ`F- bm+?  iBK2Qmi +M #2 b2i iQ `2~2+i i?2
r?B+? i?2 bQm`+2 /2pB+2 Bb +QMM2+i2/ iQX aBM+2- BMB@ HB72iBK2 M/ 7`2[m2M+v Q7 BMi2`+iBQMb rBi?BM i?2 aQ+BQ@
iBHHv-  7Q`r`/BM; `mH2 Bb MQi b2i BM i?2 ~Qr i#H2 +bi ;`QmT- i?2 KQ#BHBiv Tii2`Mb Q7 MQ/2bX
Q7 i?2 a.L ;i2rv- i?2 :QhQ*QMi`QHH2` `mH2
TTHB2b 7Q` BiX >2M+2-  P6 S+F2inAM K2bb;2 Bb 8Xj Sm#HBb?fbm#b+`B#2
Bbbm2/ iQ #2 i`MbKBii2/ iQ i?2 a.L +QMi`QHH2`X
aQ+BQ+bi +M #2 2tTHQBi2/ iQ bmTTQ`i  Tm#@
kX lTQM `2/BM; i?2 ?2/2` Q7 i?2 aQ+BQ+bi T+F2i - k HBb?fbm#b+`B#2 BMi2`+iBQM KQ/2H b r2HHX AM 7+i-  /2@
i?2 a.L +QMi`QHH2` `2HBx2b i?i  aQ+BQ+bi ;`QmT pB+2 +M bm#b+`B#2 iQ `2+2Bp2 T+F2ib Tm#HBb?2/ #v /2pB+2b
Kmbi #2 +`2i2/ U62im`2 }2H/ b2i iQ :`QmT*`2@ B/2MiB}2/ #v i?2B` TQbBiBQM BM i?2 bQ+BH M2irQ`FX 6Q`
iBQMVX h?mb- Bi Bbbm2b  `2[m2bi iQ i?2 a_a- iQ 2tKTH2- bbmK2 i?i /2pB+2 ӷ rMib iQ bm#b+`B#2 iQ
`2i`B2p2 i?2 b2i Q7 /2pB+2b- BMi2M/2/ iQ +i b `2+BT@ `2+2Bp2 T+F2ib ;2M2`i2/ #v Bib 7`B2M/b Q7 ivT2 PP_X
B2Mib Q7 i?2 aQ+BQ+bi T+F2iX A7 i?Bb Bb i?2 +b2- Bi rBHH ;2M2`i2  aQ+BQ+bi T+F2i
rBi? i?2 62im`2:`QmT }2H/ b2i iQ Sm#fam# M/ i?2
jX h?2 a> i`B;;2`b i?2 #`QrbBM; Q7 i?2 bQ+BH M2irQ`F- _2HiBQM6BHi2` }2H/ B/2MiB7vBM; M PP_X
b bT2+B}2/ #27Q`2- M/ `2im`Mb iQ i?2 a.L +QM@ am+? M BM7Q`KiBQM rBHH `2+? i?2 a.L +QMi`QHH2` r?B+?
i`QHH2` i?2 //`2bb2b Q7 i?2 b2i Q7 /2pB+2b Q7 i?2 aQ@ rBHH T2`7Q`K i?2 7QHHQrBM; QT2`iBQMb,
+BQ+bi ;`QmTX
RX Ai b2M/b  [m2`v iQ i?2 a> M/ `2+2Bp2b i?2 B/2MiBiB2b
9X h?2 a.L +QMi`QHH2` `2i`B2p2b 7`QK i?2 LA" i?2 a.L Q7 i?2 /2pB+2b rBi? TQbBiBQM BM i?2 bQ+BH M2irQ`F
MQ/2b BM i?2 b?Q`i2bi Ti?b iQr`/b i?2 BMi2M/2/ `2@ +QMbBbi2Mi rBi? i?2 `2[m2bi #v /2pB+2 ӷX
+2Bp2`b Q7 i?2 aQ+BQ+bi ;`QmTX h?2M- Bi #mBH/b i?2 kX Ai //b i?Bb BM7Q`KiBQM BM  T2M/BM; BMi2`2bi i#H2
`QmiBM; Ti?b #v 2Mbm`BM; i?i a.L MQ/2b #2HQM;@ r?B+? i`+Fb HH bm#b+`BTiBQMb `2+2Bp2/ #v /2pB+2bX
BM; iQ i?2 Ti? iQr`/b KmHiBTH2 `2+2Bp2`b `2+2Bp2 q?2M2p2`  /2pB+2 #2;BMb iQ /Bbb2KBMi2 /i- i?2
 bBM;H2 `mH2 M/ 7Q`r`/ i?2 aQ+BQ+bi T+F2i QMHv a.L +QMi`QHH2` rBHH +?2+F r?2i?2` i?2`2 `2 /2pB+2b
QM+2X >2M+2- Bi BMD2+ib 7Q`r`/BM; `mH2b BM i?2 ~Qr i?i ?p2 bm#b+`B#2/ iQ Bib mT/i2b U2X;X- "VX
i#H2 Q7 BMpQHp2/ a.L MQ/2b ++Q`/BM;Hv- #v b2M/BM;
P6 6HQrnJQ/ K2bb;2bX AM T`iB+mH`- i?2 a.L jX A7 i?Bb Bb i?2 +b2- i?2 a.L +QMi`QHH2` rBHH BMbi`m+i
;i2rvb r?B+? i?2 aQ+BQ+bi /2biBMiBQMb `2 i@ i?2 a.L MQ/2b BM i?2 Ti? iQ ӷ iQ 7Q`r`/ i?2 /i
i+?2/ iQ- rBHH #2 BMbi`m+i2/ #v i?2 a.L +QMi`QHH2` T+F2ib iQ BiX
rBi?  `mH2 i?i, UBV Ki+?2b i?2 aQ+BQ+bi@`2Hi2/
?2/2` }2H/b i?i B/2MiB7v i?2 aQ+BQ+bi +QKKmMB@ 8X9 aQm`+2 }Hi2`BM;
+iBQM M/ UBBV 7Q`2b22b iQ 7Q`r`/ i?2 T+F2i iQ i?2
+Q``2+i T?vbB+H TQ`i 7i2` +?M;BM; i?2 /2biBMiBQM aQ+BQ+bi HHQrb  /2pB+2 iQ b2H2+i i?Qb2 i?i `2 2MiBiH2/
aQ+BQ+bi AS //`2bb rBi? i?2 AS /2biBMiBQM UmMB@ iQ b2M/ T+F2ib iQ Bi- #b2/ QM i?2B` TQbBiBQM BM i?2 bQ+BH
+biV //`2bb b +iBQMX h?Bb Bb iQ 2Mbm`2 i?i HH M2irQ`FX am+?  72im`2 +M #2 mb2/ #Qi? BM  T`Q+iBp2
/2pB+2b #2HQM;BM; iQ i?2 aQ+BQ+bi ;`QmT +Q``2+iHv M/  `2+iBp2 rvX JQ`2 bT2+B}+HHv-

k h?22MiB`2 aQ+BQ+bi T+F2i Bb b2Mi #v i?2 a.L ;i2rv- ?2M+2 

Ç S`Q+iBp2,  /2pB+2 KB;?i /2+B/2 iQ `2+2Bp2 T+F2ib
S+F2inPmi Bb i`MbKBii2/ #v i?2 +QMi`QHH2`- #+F iQ i?2 a.L #v Bib 7`B2M/b QMHv- 7Q` b2+m`Biv `2bQMb Q` iQ bp2 2M@
;i2rv (jj)X 2`;v- +QKTmiiBQMH M/ +QKKmMB+iBQM `2bQm`+2bX

44 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

6B;X k Ĝ 1tKTH2b Q7 aQ+BQ+bi h; +QM};m`iBQMX

Ç _2+iBp2, i?2 +QKTmiiBQMH Q` +QKKmMB+iBQM HQ/ hQ i?Bb Tm`TQb2- r2 #mBHi M 2KmHiBQM THv;`QmM/X AM
7Q`  /2pB+2 Kv #2+QK2 iQQ ?B;?- 2X;X- #2+mb2 Q7  T`iB+mH`- i?2 JBMBM2i M2irQ`F 2KmHiQ` (8) ?b #22M
.Qa ii+FX A7 i?Bb Bb i?2 +b2- i?2 /2pB+2 KB;?i /2@ mb2/- Bi HHQrb i?2 +`2iBQM Q7  M2irQ`F rBi? i?QmbM/b
+B/2 iQ ++2Ti T+F2ib #v  bm#b2i Q7 /2pB+2b- #b2/ Q7 MQ/2b QM i?2 HBKBi2/ `2bQm`+2b Q7  bBM;H2 UpB`imHV
QM i?2B` TQbBiBQM BM i?2 bQ+BH M2irQ`FX AM i?Bb rv K+?BM2X AM T`iB+mH`- Bi 2M#H2b 7bi T`QiQivTBM; M/
aQ+BQ+bi +M #2 2tTHQBi2/ iQ `2HBx2  }`2rHH i?2 2tT2`BK2MiH 2pHmiBQM Q7 P6@2M#H2/ M2irQ`F2/ bvb@
TQHB+B2b Q7 r?B+? +?M;2 /2T2M/BM; QM i?2 +m``2Mi i2KbX h?2 2tT2`BK2MiH b2iiBM; +QMbBbib Q7 i?2 M2irQ`F
HQ/X iQTQHQ;v /2TB+i2/ BM 6B;X jX  7mHH@K2b? BMi2`+QMM2+ib
i?2 +Q`2 a.L MQ/2b- r?B+? `2 i?2 `QQib Q7  i?`22@Hv2`b
 /2pB+2- bv Ӹ- rBb?BM; MQi iQ `2+2Bp2 T+F2ib 7`QK 7i@i`22 iQTQHQ;vX lT iQ kR /2pB+2b `2 ii+?2/ iQ 2+?
MQ/2b rBi? +2`iBM bQ+BH T`QT2`iB2b b2M/b  aQ+BQ@ a.L ;i2rv UMQi HH i?2 /2pB+2b `2 b?QrM BM i?2 };@
+bi T+F2i #v bT2+B7vBM; BM i?2 62im`2:`QmT }2H/ m`2VX PLPa ?b #22M +QMbB/2`2/ b  `272`2M+2 a.L
aQm`+26BHi2`BM;X PM+2 i?2 T+F2i `2+?2b i?2 a.L +QMi`QHH2` BM i?2 +QMi2ti Q7 i?Bb rQ`F- /m2 iQ Bib b+H@
+QMi`QHH2`- i?2 Hii2` QM2 rBHH [m2`v i?2 a>- r?B+? rBHH #BHBiv T`QT2`iB2b M/ Bib ?B;?Hv KQ/mH` `+?Bi2+im`2 (e)X
`2THv rBi? i?2 HBbi Q7 mi?Q`Bx2/ AS //`2bb2bX ++Q`/@ h?2 PLPa +QMi`QHH2` BMi2`+ib rBi? M 2ti2`MH a_a-
BM;Hv- i?2 a.L +QMi`QHH2` rBHH BMb2`i 2Mi`B2b BM i?2 ~Qr r?B+? 2bi#HBb?2b bQ+BH `2HiBQMb?BTb KQM; 2KmHi2/
i#H2 Q7 i?2 a.L ;i2rv r?B+? Ӹ Bb ii+?2/ iQ- iQ /2pB+2b- M/ KM;2b i?2KX
bT2+B7v i?2 7Q`r`/BM; `mH2 7Q` T+F2ib /2biBM2/ iQ Bi b2Mi
h?2 PLPa +QMi`QHH2` M/ i?2 JBMBM2i M2irQ`F 2KmHiQ`
7`QK mi?Q`Bx2/ b2M/2`b M/ i?2 /`QTTBM; `mH2 7Q` i?Qb2
`2 #Qi? `mMMBM; QM i?2 bK2 pB`imH K+?BM2- r?BH2 i?2
r?B+? `2 MQi HHQr2/X
a_a `mMb BM  /Bz2`2Mi QM2X "Qi? i?2b2 pB`imH K+?BM2b
`2 HQ+i2/ BM  T?vbB+H b2`p2` rBi? M AMi2H s2QMU_V
eX S1_6P_JL*1 1oGlhAPL *Sl 18@kejy* pj RX3y :>x tjk T`Q+2bbQ` M/ jdd-3
AM i?Bb b2+iBQM r2 /2b+`B#2 i?2 2MpB`QMK2Mi 7Q` i?2 T2`@ :B" Q7 K2KQ`vX
7Q`KM+2 2pHmiBQMX JQ`2 bT2+B}+HHv- BM a2+iBQM eXR
r2 /2b+`B#2 i?2 iQQHb miBHBx2/ 7Q` i?2 T2`7Q`KM+2 2pH@ eXk aQ+BH `2HiBQMb?BTb b2iiBM;b M/ i`{+
miBQM- M/ BM a2+iBQM eXk r2 /Bb+mbb i?2 b+2M`BQbX h?2 Tii2`Mb
#2M+?K`F miBHBx2/ 7Q` +QKT`BbQM Tm`TQb2b Bb T`2b2Mi2/
BM a2+iBQM eXj- r?2`2b i?2 +QMbB/2`2/ T2`7Q`KM+2 K2i@ h?2 T2`7Q`KM+2 Q7 i?2 T`QTQb2/ bQHmiBQM ?b #22M 2pH@
`B+b `2 B/2MiB}2/ BM a2+iBQM eX9X mi2/ rBi?  b2i Q7 `2T`2b2MiiBp2 AQh i2bi +QM};m`iBQMb
T`QT2`Hv /2bB;M2/ iQ iF2 BMiQ ++QmMi /Bz2`2Mi MmK#2`b
eXR hQQHb M/ `272`2M+2 iQTQHQ;v M/ /Bbi`B#miBQMb Q7 MQ/2b BM i?2 2KmHi2/ iQTQHQ;v- /B7@
72`2Mi T?vbB+H /BbiM+2b #2ir22M bQm`+2b M/ /2biBM@
h?2 7Q+mb Q7 i?2 T2`7Q`KM+2 2pHmiBQM Bb iQ bb2bb aQ@ iBQMb- M/ /Bz2`2Mi ivT2b Q7 b2`pB+2X h?Bb Bb BK2/ i
+BQ+bi BM i?2 +b2 Q7 Tmb?@#b2/ /i /Bbb2KBMiBQM iQ@ KFBM; i?2 Q#iBM2/ `2bmHib b ;2M2`HBx#H2 b TQbbB#H2
r`/b  ;`QmT Q7 /2pB+2bX M/ ?pBM;  +H2` B/2 Q7 i?2 TQi2MiBH M/ HBKBib Q7

© International Telecommunication Union, 2020 45

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

6B;X j Ĝ _272`2M+2 iQTQHQ;vX

aQ+BQ+bi BM KmHiBTH2 b+2M`BQbX 1+? Q7 i?2 i2bi +QM};@ eXkXk a+2M`BQ ", aK`i ?QK2 KQMBiQ`BM;
m`iBQMb ?b #22M KTT2/ QMiQ  mb2 +b2 +?`+i2`Bx2/
:`QmT +QKKmMB+iBQM M22/b,  `M/QKHv b2H2+i2/ /2pB+2
#v i?2 2tTHQBiiBQM Q7  T`iB+mH` ivT2 Q7 bQ+BH `2HiBQM@
BM i?2 2KmHi2/ iQTQHQ;v- `2b2K#HBM;  bK`iT?QM2 Q7 
b?BT #2ir22M i?2 /2pB+2b BMpQHp2/X AM i?Bb rv- ?2HT7mH
mb2` +m``2MiHv i Q{+2- +ib b  b2M/2` M/ Bbbm2b  aQ@
;mB/2HBM2b +M #2 T`QpB/2/ #Qmi i?2 bmBi#BHBiv Q7 i?2
+BQ+bi T+F2i iQ +`2i2  ;`QmT Q7 `2+BTB2Mib K/2 Q7 HH
T`QTQb2/ bQHmiBQM BM i?2 +QMi2ti Q7 /Bz2`2Mi TTHB+iBQM
i?2 bK`i /2pB+2b +QMM2+i2/ iQ i?2 UbK2V ?QK2 ;i2rv-
b+2M`BQb M/- i i?2 bK2 iBK2- Q7 i?2 2z2+iBp2M2bb Q7
r?B+? Bb /Bz2`2Mi 7`QK i?2 QM2 i?2 mb2`Ƕb bK`iT?QM2 Bb
+QKKmMB+iBQMb #b2/ QM 2+? Q7 i?2 TQbbB#H2 bQ+BH@HBF2
ii+?2/ iQX h?2 ;`QmT +M #2 +`2i2/- 7Q` BMbiM+2- iQ
`2HiBQMb?BTb 2bi#HBb?2/ KQM; AQh /2pB+2bX .2iBHb `2
MQiB7v /2pB+2b iQ +QM};m`2  r`K r2H+QK2 7Q` i?2 mb2`X
;Bp2M BM i?2 7QHHQrBM; bm#b2+iBQMbX h#H2 R HbQ bmK@
K`Bx2b i?2 KDQ` 72im`2b +?`+i2`BxBM; 2+? b+2M`BQ- AMpQHp2/ `2HiBQMb?BT ivT2, PP_ UQrM2`b?BTVX
r?B+? `2, i?2 ivT2b Q7 bQ+BH `2HiBQMb?BT Ub?Q`i2M2/ b 1M/TQBMi /Bbi`B#miBQM T`Q}H2, b2M/2` BM  HQ+iBQM M/
_2HXV- i?2 MmK#2` Q7 /2biBMiBQMb Ub?Q`i2M2/ b .ahbV HH /2biBMiBQMb +Hmbi2`2/ BM  /Bz2`2Mi UTQi2MiBHHvV
7Q` 2+? +QKKmMB+iBQM- i?2B` /BbiM+2 7`QK i?2 bQm`+2 `2KQi2 HQ+iBQMX
Ub?Q`i2M2/ b a_*V- M/ i?2B` TQbBiBQM rBi? `272`2M+2 iQ
i?2 +QMbB/2`2/ M2irQ`F iQTQHQ;vX
eXkXj a+2M`BQ *, qB`2H2bb a2MbQ` L2irQ`F
UqaLV KM;2K2Mi
eXkXR a+2M`BQ , aK`i BM/mbi`BH THMi :`QmT +QKKmMB+iBQM M22/b,  `M/QKHv b2H2+i2/ /2pB+2
BM i?2 2KmHi2/ iQTQHQ;v +ib b  b2M/2` M/ Bbbm2b 
:`QmT +QKKmMB+iBQM M22/b, M BM/mbi`BH THMi Bb aQ+BQ+bi ;`QmT +`2iBQM /2biBM2/ iQ HH i?2 /2pB+2b Q7 i?2
2[mBTT2/ rBi? b2p2`H +QMM2+i2/ /2pB+2b Ub2MbQ`b M/ bK2 #`M/- mMB7Q`KHv /Bbi`B#mi2/ BM i?2 iQTQHQ;v iQ /Bb@
+imiQ`bV M/ QM2 Q7 i?2b2 U`M/QKHv b2H2+i2/V #2HQM;@ b2KBMi2  M2r +QM};m`iBQM 7Q` i?2 /2pB+2-  bQ7ir`2
BM; iQ i?2 2KmHi2/ iQTQHQ;v Bbbm2b  aQ+BQ+bi T+F2i mT/i2- Q`  M2r /`Bp2` p2`bBQMX
/2biBM2/ iQ HH i?2 /2pB+2b +QMM2+i2/ iQ i?2 bK2 ;i2@ AMpQHp2/ `2HiBQMb?BT ivT2, SP_ UT`2MiHVX
rvX h?2 ;`QmT +M #2 +`2i2/- 7Q` BMbiM+2- 7Q` i?2 1M/TQBMi /Bbi`B#miBQM T`Q}H2, mMB7Q`K /Bbi`B#miBQM Q7
/Bbb2KBMiBQM Q7 H`Kb- 7Q` ;`QmT +QM};m`iBQM M/ `2@ 2M/TQBMibX
+QM};m`iBQM- 7Q` 7mM+iBQMH i2biBM;X

AMpQHp2/ `2HiBQMb?BT ivT2, *GP_X eXkX9 a+2M`BQ ., aK`i KQ#BHBiv

1M/TQBMi /Bbi`B#miBQM T`Q}H2, HH 2M/TQBMib +Hmbi2`2/ BM :`QmT +QKKmMB+iBQM M22/b, r2 bbmK2 KQ#BH2 /2@
i?2 bK2 `2X pB+2b U2X;X- bK`iT?QM2b- HTiQTbV +``B2/ #v T2QTH2 KQp@
BM; BM  bK`i +BivfbK`i +KTmb M/ BMi2`+iBM; rBi?

46 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Qi?2` /2pB+2b K2i 2Bi?2` BM i?2 M2B;?#Q`?QQ/ Q` +HQb2 Q7@ M/ /vMKB+ aQ+BQ+bi ;`QmT +`2iBQM- ?2M+2 2Mbm`BM;
}+2bf+Hbb`QQKbX h?2 ivT2 Q7 i?2 /i 2t+?M;2/ rBi?BM  7B` +QKT`BbQMX aT2+B}+HHv- i?2 bQm`+2 MQ/2 +QM@
i?2 ;`QmT BM+Hm/2b, BM7Q`KiBQM `2Hi2/ iQ KQ#BHBiv T@ i+ib  T`Qtv BM +?`;2 Q7 BMi2`+iBM; rBi?  aAQh@HBF2
THB+iBQMb- iQm`Bbi BM7Q`KiBQM- /i 7Q` i?2 BKTH2K2M@ THi7Q`K iQ ;2i i?2 b2i Q7 BMi2M/2/ /2biBMiBQMb #2HQM;@
iiBQM Q7 Mv AMi2HHB;2Mi h`MbTQ`iiBQM avbi2Kb TTHB@ BM; iQ i?2 aQ+BQ+bi ;`QmTX h?2 Hii2` QM2 Bb /2b+`B#2/
+iBQMX i?`Qm;? ii`B#mi2bfK2i/i /2}M2/ i i?2 TTHB+iBQM
AMpQHp2/ `2HiBQMb?BT ivT2, aP_X Hv2`- bBKBH`Hv iQ i?2 BM7Q`KiBQM 2M+Q/2/ BM i?2 i;b
1M/TQBMi /Bbi`B#miBQM T`Q}H2, p`B#H2 HQ+iBQM Q7 2M/@ BM aQ+BQ+bi T+F2ibX 7i2` `2i`B2pBM; i?2 HBbi- i?2 T`Qtv
TQBMib BM i?2 ;`QmTX 7Q`r`/b Bi iQ i?2 bQm`+2 MQ/2 r?B+? b2M/b i?2 T+F2i
iQ i?2 /2biBMiBQMb i?`Qm;? KmHiBTH2 mMB+bi 2t+?M;2bX
AM Qi?2` rQ`/b- i?2 +QMi`QHH2` b2ib mT /BbiBM+i `QmiBM;
eXkX8 _2HiBQMb?BTb +`2iBQM
Ti?b 7Q` 2+? /2biBMiBQM M/ bQK2 HBMFb +M #2 b?`2/
b iQ i?2 +`2iBQM Q7 i?2 `2HiBQMb?BTb- i?2b2 ?p2 #22M #v KmHiBTH2 Ti?b iQr`/b /2biBMiBQMb #2HQM;BM; iQ i?2
b2i BM /2i2`KBMBbiB+ rv 2t+2Ti 7Q` i?2 aP_X AM T`iB+m@ bK2 ;`QmTX qBi?Qmi HQbBM; ;2M2`HBiv- r2 bbmK2 i?i
H`- /Bz2`2Mi ;`QmTb Q7 /2pB+2b HBMF2/ rBi? SP_- PP_- i?2 T`Qtv Bb ii+?2/ iQ i?2 `QQi MQ/2 Q7 i?2 iQTQHQ;v
M/ SP_ `2HiBQMb?BTb `2 +`2i2/ bQ b iQ ?p2 7`QK UBX2X- Ԣ BM 6B;X jVX
8 iQ ky `2+BTB2Mib 7Q` 2+? bBKmHi2/ +QKKmMB+iBQMX
>Qr2p2`- i?2 *GP_ `2HiBQMb?BT ?b #22M +`2i2/ #2@ eX9 J2i`B+b
ir22M /2pB+2b i?i `2 +QMM2+i2/ iQ i?2 bK2 ;i2rv
b i?2 +Q@HQ+iBQM ?b iQ #2 bbm`2/X b iQ i?2 aP_ h?2 7QHHQrBM; K2i`B+b ?p2 #22M +QMbB/2`2/ iQ 2pHmi2
`2HiBQMb?BTb mb2/ BM a+2M`BQ .- i?2b2 `2 2bi#HBb?2/ i?2 T2`7Q`KM+2 Q7 i?2 +QKT`2/ b+?2K2b BM i?2 +`2@
#2ir22M /2pB+2b BM i?2 2KmHi2/ iQTQHQ;v ++Q`/BM; iQ iBQM Q7  aQ+BQ+bi ;`QmT M/ /i 2t+?M;2 KQM; Bib
i?2B` T?vbB+H /BbiM+2 M/ 7QHHQr  bBKTH2 T`Q##BHBbiB+ K2K#2`b,
KQ/2HX h?2 T`BM+BTH2 /QTi2/ Bb bm+? i?i i?2 +HQb2` i?2
/2pB+2b- i?2 ?B;?2` i?2 T`Q##BHBiv i?i i?2 irQ /2pB+2b Ç i?2 MmK#2` Q7 P6 bB;MHBM; T+F2ib 2t+?M;2/ #2@
?p2 2bi#HBb?2/  aP_ `2HiBQMb?BTX ++Q`/BM;Hv- /2@ ir22M a.L MQ/2b M/ +QMi`QHH2` iQ #mBH/ `QmiBM;
pB+2b ii+?2/ iQ i?2 bK2 a.L ;i2rv UBX2X- M ++2bb Ti?b iQr`/b i?2 BMi2M/2/ aQ+BQ+bi /2biBMiBQMbX
SQBMiV ?p2 i?2 ?B;?2bi T`Q##BHBiv iQ 2bi#HBb? BiX h?2b2 h?2 K2i`B+ QMHv `272`b iQ i?2 +QMi`QH T+F2ib 2t@
/2pB+2b `2 +?`+i2`Bx2/ #v b?`BM; i?2 bK2 Ti? iQ +?M;2/ iQ T`Q+2bb BM+QKBM; `2[m2bib 7`QK bQ+BQ+bi
`2+? i?2 `QQi MQ/2 UԢ BM 6B;m`2 jV- r?B+? Bb K/2 Q7 9 MQ/2b i i?2 a.L ;i2rv- MK2Hv S+F2inAM-
a.L MQ/2bX q2 #b2 QM i?Bb MmK#2` iQ /2}M2 i?2 MQi@ S+F2inPmi M/ 6HQrnJQ/X h?2 #+F;`QmM/
iBQM iQ /2MQi2 i?2 `2H2pMi T`Q##BHBiv iQ +`2i2  aP_ UT2`BQ/B+V bB;MHBM; 2t+?M;2/ #2ir22M i?2 +QM@
#2ir22M i?2K, ԟ֎֊վӴΚ X 6QHHQrBM; i?2 bK2 T`BM+BTH2- /2@ i`QHH2` M/ i?2 a.L MQ/2b Bb MQi +QMbB/2`2/c
pB+2b b?`BM; i?`22- irQ- Q` QM2 a.L MQ/2b BM i?2 Ti? iQ
`2+? Ԣ- 2bi#HBb?  `2HiBQMb?BT rBi? T`Q##BHBiv ԟ֎֊վӴϯ - Ç i?2 MmK#2` Q7 /i T+F2ib 2t+?M;2/ BMiQ i?2 M2i@
ԟ֎֊վӴϵ - M/ ԟ֎֊վӴφ - `2bT2+iBp2HvX h?2 ?B;?2` ԙ i?2 ?B;?2` rQ`F iQ `2+? HH i?2 BMi2M/2/ /2biBMiBQMb Q7 i?2
i?2 T`Q##BHBiv ԟ֎֊վӴօ - rBi? ԙ ୮ \   ^X h?2 b2iiBM; +QKKmMB+iBM; ;`QmT- QM+2 Bi ?b #22M +`2i2/c i?2
Q7 ԟ֎֊վӴօ mb2/ BM i?2 T2`7Q`K2/ bBKmHiBQMb Bb `2TQ`i2/ BM K2i`B+ +QMbB/2`b i?2 MmK#2` Q7 i`MbKBii2/ T+F2ib
h#H2 kc /Bz2`2Mi +QM};m`iBQMb ?p2 #22M +QMbB/2`2/ iQ T2` HBMF M/ `2 `2T`2b2Mi2/ #v 2Bi?2` aQ+BQ+bi Q`
2pHmi2 i?2 BKT+i Q7 /Bz2`2Mi MmK#2`b Q7 7`B2M/b M/ J@lMB+bi T+F2ibX
i?2B` /Bbi`B#miBQM BM i?2 +QMbB/2`2/ iQTQHQ;vX
6Q` i?2 #2M+?K`F b+?2K2- i?2 `2[m2bi T+F2ib Bbbm2/
#v i?2 bQm`+2 iQr`/b i?2 T`Qtv b r2HH b i?2 bB;MHBM;
eXj "2M+?K`F b+?2K2 K2bb;2b `2[mB`2/ iQ BMbi`m+i i?2 `2H2pMi a.L MQ/2b
h?2 T2`7Q`KM+2 Q7 i?2 T`QTQbH ?b #22M +QKT`2/ iQr`/b Bi `2 HbQ +QMbB/2`2/X
;BMbi M TTHB+iBQM@Hv2` bQHmiBQM r2 `272` iQ b KmH@ h?2 #Qp2 K2i`B+b ?p2 #22M K2bm`2/ i?`Qm;? i?2 r2HH@
iBTH2 mMB+bi UH#2H2/ BM i?2 THQib b J@lMB+biVX LQi2 FMQrM qB`2b?`F M2irQ`F T`QiQ+QH MHvx2`j X
i?i HbQ 7Q` i?Bb #2M+?K`F b+?2K2- r2 `2 7Q+mbBM; *QKT`BbQM 2tT2`BK2Mib ?p2 #22M +QM/m+i2/ r?2M
QM i?2 Tmb?@#b2/ /i /Bbb2KBMiBQM b+2M`BQX h?2 p`vBM; i?2 MmK#2` Q7 /2biBMiBQMb UQ` `2H2pMi T`Q#@
+?QB+2 Q7 i?Bb #2M+?K`F Bb K2Mi iQ [mMiBiiBp2Hv 2b@ #BHBiv b2iiBM;bV M/ `2 p2`;2/ Qp2` ky `mMbX
iBKi2 i?2 #2M2}ib Q7 i?2 aQ+BQ+bi T`QTQbH ;BMbi M j SH2b2MQiB+2 i?i i?2 MHvbBb Q7 i?2 bB;MHBM; BM+m``2/ 7Q` i?2
TTHB+iBQM@Hv2` bQHmiBQMX AM i?2 Hii2` QM2- i?2 M2irQ`F
+`2iBQM Q7 bQ+BH `2HiBQMb?BTb #2ir22M /2pB+2b Bb QmibB/2 i?2
Hv2` Bb ;MQbiB+ #Qmi i?2 +QKKmMB+iBM; ;`QmT- #mi b+QT2 Q7 i?Bb TT2` M/ Bb T2+mHB` Q7 i?2 +QM+2Bp2/ aAQh BK@
Bi Qz2`b i?2 bK2 72im`2b BM i2`Kb Q7 b2M/2`@BMBiBi2/ TH2K2MiiBQMX M BMi2`2bi2/ `2/2` Bb `272``2/ iQ (jd)X

© International Telecommunication Union, 2020 47

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

h#H2 R Ĝ amKK`v Q7 i?2 KBM bQ+BH `2HiBQMb?BTb b2iiBM;bX

a+2M`BQ lb2 +b2 _2HX O.ahb a_*@.ahb SQbBiBQM Q7

.BbiM+2 .ahb
 aK`i BM@ *GP_ 8@ky R ?QT 7Q` HH ii+?2/ iQ
/mbi`v /2biBMiBQMb i?2 bK2
a.L ;i2@
rv U4
b2M/2`V
" aK`i PP_ 8@ky 6Bt2/ 7Q`  ii+?2/ iQ
?QK2 ;Bp2M b2i Q7 i?2 bK2
/2biBMiBQMb a.L ;i2@
rv Uଈ
b2M/2`V
* qaL KM@ SP_ 8@ky R@d ?QTb lMB7Q`KHv
;2K2Mi /Bbi`B#mi2/
BM i?2 iQTQH@
Q;v
. aK`i KQ@ aP_
#BHBiv

h#H2 k Ĝ S`Q##BHBiB2b Q7 aP_ 2bi#HBb?K2MiX

aBK@A. O.2biBMiBQMb ԟ֎֊վӴφ ԟ֎֊վӴϵ ԟ֎֊վӴϯ ԟ֎֊վӴΚ

R 9X8 yXR yXk yXj yX9
k 8X8 yXk yXj yX9 yX8
j 3X9 yXj yX9 yX8 yXe
9 RRXk yX9 yX8 yXe yXd
8 RkXd yX8 yXe yXd yX3
e R8Xj yXe yXd yX3 yXN
d R3X9 yXd yX3 yXN R

dX 1sS1_AJ1LhG _1alGha AM Qi?2` rQ`/b- M a.L MQ/2 `2+2Bp2b  MmK#2` Q7 J@

lMB+bi T+F2ib iQ 7Q`r`/ 2[mH iQ i?2 MmK#2` Q7 /2b@
AM i?Bb b2+iBQM r2 b?Qr i?2 T2`7Q`KM+2 `2bmHibX JQ`2 iBMiBQMb Bi HHQrb iQ `2+?X 6Q` 2+? Q7 i?2K- Bi +QM@
bT2+B}+HHv- BM a2+iBQM dXR r2 bb2bb aQ+BQ+bi BM i2`Kb i+ib i?2 +QMi`QHH2` #v ;2M2`iBM;  S+F2inAM K2b@
Q7 ;2M2`i2/ bB;MHBM; T+F2ibc r?2`2b BM a2+iBQM dXk r2 b;2 M/ rBib 7Q` i?2 +Q``2bTQM/BM; S+F2inPmi M/
rBHH 7Q+mb QM /i T+F2ibX 6HQrnJQ/ rBi? BMbi`m+iBQMb #Qmi i?2 7Q`r`/BM; #2@
?pBQm`X
dXR aB;MHBM; T+F2ib 6Q`  ;Bp2M MmK#2` Q7 /2biBMiBQMb- i?2 ?B;?2bi MmK#2`
h?2 }`bi b2i Q7 `2bmHib BKb iQ MHvx2 i?2 +QMi`QH THM2 Q7 P6 T+F2ib `2 2t+?M;2/ BM +b2 Q7 a+2M`BQ *X AM
bB;MHBM; 7QQiT`BMi BM+m``2/ #v i?2 T`QTQbH rX`XiX i?2 i?2 Hii2` QM2- BM/22/- i?2 /2biBMiBQMb `2 bT`2/ Qp2`
#2M+?K`F b+?2K2X 6B;X 9 `2TQ`ib i?2 MmK#2` Q7 2t@ i?2 iQTQHQ;v M/ i?2 `QmiBM; Ti? iQr`/b i?2K Kv
+?M;2/ P6 T+F2ib r?2M p`vBM; i?2 MmK#2` Q7 /2b@ BMpQHp2 b2p2`H a.L MQ/2b UM/ ;i2rvbVX a+2M`BQ "
iBMiBQMb Q7 i?2 aQ+BQ+bi ;`QmT 7Q` i?2 b+2M`BQb @*- 7QHHQrb rBi?  HQr2` MmK#2` Q7 2t+?M;2/ P6 T+F2ibX
r?2`2b i?2 `2bmHib 7Q` b+2M`BQ . `2 b?QrM BM 6B;X 8UVX AM a+2M`BQ - BMbi2/- QMHv  bBM;H2 a.L ;i2rv Bb
Ai +M #2 +H2`Hv Q#b2`p2/ i?i 7Q` i?2 J@lMB+bi T@ BM +?`;2 Q7 aQ+BQ+bi T+F2i 7Q`r`/BM;X Ai Bb i?2 QMHv
T`Q+? i?2 K2i`B+ bB;MB}+MiHv BM+`2b2b rBi? i?2 MmK@ a.L MQ/2 i`MbKBiiBM; M/ `2+2BpBM; P6 T+F2ibX
#2` Q7 /2biBMiBQMb- BM HH i?2 +QMbB/2`2/ b+2M`BQbX am+? AM i?2 T`QTQb2/ aQ+BQ+bi bQHmiBQM- i?2 +QMi`QHH2` Bb BM
 i`2M/ Bb /m2 iQ i?2 7+i i?i i?2 2M/@iQ@2M/ +QKKm@ +?`;2 Q7 #mBH/BM; `QmiBM; Ti?b iQr`/b i?2K bQ iQ
MB+iBQM Ti? iQr`/b 2+? bBM;H2 /2biBMiBQM M22/b iQ pQB/ i?2 7Q`r`/BM; Q7 i?2 bK2 aQ+BQ+bi T+F2i Qp2`
#2 /Bb+Qp2`2/ rBi? i?2 bmTTQ`i Q7 i?2 a.L +QMi`QHH2`X i?2 bK2 HBMFX

48 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

>2M+2- mMHBF2 i?2 #2M+?K`F b+?2K2- BM Qm` T`QTQbH-

i?Qb2 a.L MQ/2b r?B+? #2HQM; iQ i?2 Ti?b iQr`/b /B7@
72`2Mi /2biBMiBQMb `2+2Bp2 QMHv  bBM;H2 aQ+BQ+bi T+F2i
iQ 7Q`r`/ M/  bBM;H2 6HQrnJQ/ 7`QK i?2 +QMi`QHH2`X
h?2 ;BM Q7 aQ+BQ+bi rX`XiX J@lMB+bi BM i2`Kb Q7 2t@
+?M;2/ P6 T+F2ib ;2ib KQ`2 `2K`F#H2 b i?2 MmK@
#2` Q7 /2biBMiBQMb BM+`2b2bX 6Q` BMbiM+2- BM a+2M`BQ
*- Bi Tbb2b 7`QK  7+iQ` Q7 `QmM/ e 7Q` }p2 /2biBMiBQMb
iQ  7+iQ` Q7 KQ`2 i?M R9 7Q` ir2Miv /2biBMiBQMbX
Ai Bb rQ`i? Q#b2`pBM; i?i- BM aQ+BQ+bi-  bBM;H2
6HQrnJQ/ K2bb;2 Kv +QMp2v KmHiBTH2 `mH2b iQ #2
BMD2+i2/ BMiQ M a.L MQ/2X AM T`iB+mH`- h#H2 j `2@
TQ`ib- 7Q` a+2M`BQ - i?2 bBx2 Q7 i?2 6HQrnJQ/ K2b@
b;2- b K2bm`2/ i i?2 a.L ;i2rv- r?B+? i?2 bQm`+2
M/ i?2 /2biBMiBQMb `2 #Qi? ii+?2/ iQX 6Q` i?2 aQ@
+BQ+bi T`QTQbH- i?2 bBx2 `2bQM#Hv BM+`2b2b rBi? i?2
MmK#2` Q7 /2biBMiBQMb iQ ++QKKQ/i2 i?2 +iBQM `mH2
7Q` 2+? Q7 i?2KX h?2 `mH2 bT2+B}2b i?2 T?vbB+H Qmi@ UV a+2M`BQ 
Tmi TQ`i b r2HH b i?2 +?M;2 Q7 i?2 AS //`2bb 7`QK
aQ+BQ+bi iQ mMB+biX 6Q` J@lMB+bi- 2+? 6HQrnJQ/
+``B2b  bBM;H2 `mH2- bBM+2 Bib BMD2+iBQM Bb Bbbm2/ T2` 2+?
J@lMB+bi T+F2i i`p2`bBM; M a.L MQ/2X h?2 bBx2
BM+`2b2b Q7 H2bb i?M  7+iQ` Q7 j 7Q` i?2 aQ+BQ+bi T@
T`Q+? +QKT`2/ iQ J@lMB+bi- BM i?2 +b2 Q7 ir2Miv
/2biBMiBQMbX
.2bTBi2 i?2 H`;2` bBx2 Q7 6HQrnJQ/ T+F2ib- Bi +M #2
2bBHv BM72``2/ i?i- Qp2`HH- i?2 P6 bB;MHBM; 7QQiT`BMi
Q7 i?2 T`QTQbH- BM i2`Kb Q7 MmK#2` Q7 2t+?M;2/ #vi2b-
Bb bB;MB}+MiHv HQr2` i?M J@lMB+biX HbQ- i?2 T`QTQbH
#2ii2` b+H2b rBi? i?2 bBx2 Q7 i?2 aQ+BQ+bi ;`QmTX
aBKBH` iQ i?2 #2M+?K`F b+?2K2- i?2 T`QTQbH 2tT2`B@
2M+2b i?2 H`;2bi bB;MHBM; BM a+2M`BQ *- r?2`2BM KmHiB@
TH2 a.L MQ/2b- BMpQHp2/ BM 7Q`r`/BM; aQ+BQ+bi T+F2ib
iQ /2biBMiBQMb- bT`2/ Qp2` i?2 iQTQHQ;v- M22/ iQ #2 BM@
bi`m+i2/X
aBKBH` +QMbB/2`iBQMb ?QH/ 7Q` a+2M`BQ .- 6B;X 8UVX U#V a+2M`BQ "
HbQ BM bm+?  +b2- i?2 T`QTQb2/ aQ+BQ+bi bQHmiBQM Bb
H2bb b2MbBiBp2 iQ i?2 bBKmHiBQM b2iiBM;b UBX2X- bBx2 Q7 aQ@
+BQ+bi ;`QmT M/ Bib +QM};m`iBQM BM i2`Kb Q7 T`QtBKBiv
Q7 /2biBMiBQMb rX`XiX i?2 bQm`+2V i?M i?2 #2M+?K`FX

dXk .i T+F2ib

_2bmHib BM 6B;X e b?2/ 7m`i?2` HB;?i BMiQ i?2 2{+B2M+v
Q7 i?2 +QKT`2/ b+?2K2b BM /2HBp2`BM; i?2 /i T+F@
2ibX aBKBH` iQ i?2 P6 bB;MHBM;- HbQ i?2 MmK#2` Q7
2t+?M;2/ aQ+BQ+bi T+F2ib BM+`2b2b rBi? i?2 MmK#2`
Q7 /2biBMiBQMbc i?2 ?B;?2bi pHm2b `2 2tT2`B2M+2/ 7Q`
a+2M`BQ * M/ i?2 HQr2bi QM2b BM a+2M`BQ X
b  ;2M2`H `2K`F- Bi +M #2 Q#b2`p2/ i?i i?2 T`QTQbH
Bb H2bb b2MbBiBp2 iQ BM+`2b2b BM i?2 MmK#2` Q7 /2biBM@
iBQMb r?2M +QKT`2/ iQ i?2 #2M+?K`FX h?Bb ?TT2Mb
#2+mb2 i?2 +QMi`QHH2` #mBH/b i?2 `QmiBM; Ti?b iQ pQB/
i?i T+F2ib `2 `2/mM/MiHv i`MbKBii2/ Qp2`  ;Bp2M U+V a+2M`BQ *

6B;X 9 Ĝ aQ+BQ+bi obX JmHiBTH2 lMB+bi, 1t+?M;2/ P6 T+F2ib

7Q` aQ+BQ+bi ;`QmT +`2iBQM r?2M p`vBM; i?2 MmK#2` Q7 /2biBM@
iBQMb- BM /Bz2`2Mi b+2M`BQbX

© International Telecommunication Union, 2020 49

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

h#H2 j Ĝ aBx2 UBM #vi2bV Q7 i?2 6HQrnJQ/ T+F2i 7Q` a+2M`BQ

X

O.2biBMiBQMb J@lMB+bi aQ+BQ+bi

R Rdk Rdk
8 Rdk k8k
Ry Rdk jjk
R8 Rdk 9Rk
ky Rdk 9Nk

HBMF b?`2/ #v KQ`2 /2biBMiBQMbX

h?Bb Bb MQi i?2 +b2 7Q` i?2 J@lMB+bi bQHmiBQM r?2`2
7Q`r`/BM; /2+BbBQMb `2 b2T`i2Hv iF2M 7Q` 2+? /i
T+F2i- ++Q`/BM; iQ i?2 //`2bb Q7 i?2 BMi2M/2/ /2biB@
MiBQMX
UV P6 T+F2ib 2t+?M;2/ iQ +`2i2 i?2 aQ+BQ+bi ;`QmT q?2M `272``BM; iQ a+2M`BQ - i?2 J@lMB+bi TT`Q+?
Hrvb b2M/b irB+2 b KMv /i T+F2ib b i?2 T`QTQbHX
h?Bb Bb M Q#pBQmb +QMb2[m2M+2 Q7 i?2 7+i i?i- 7i2` `2@
+2BpBM; i?2 /2biBMiBQMb HBbi- 7Q` i?2 J@lMB+bi TT`Q+?
i?2`2 `2 irQ T+F2ib- 7Q` 2+? /2biBMiBQM- i`p2HBM; BMiQ
i?2 iQTQHQ;vX PM2 T+F2i i`p2Hb 7`QK i?2 bQm`+2 iQ i?2
a.L ;i2rv- M/ i?2 Qi?2` QM2 7`QK i?2 a.L ;i2rv
iQ i?2 +Q``2bTQM/BM; /2biBMiBQMX h?Bb /Q2b MQi TTHv
7Q` i?2 aQ+BQ+bi TT`Q+?- r?2`2 i?2`2 Bb QMHv i?2 /i
T+F2i 7`QK i?2 a.L ;i2rv iQ 2+? /2biBMiBQMX
AKT`Qp2K2Mib ;2i H`;2` 7Q` Qi?2` b+2M`BQbX
U#V 1t+?M;2/ /i THM2 T+F2ib BM i?2 iQTQHQ;v AM a+2M`BQ "- KQ`2 a.L MQ/2b `2 BMpQHp2/ BM i?2 `Qmi@
BM; Ti?- /2bTBi2 i?2 7+i i?i HH i?2 /2biBMiBQMb `2
6B;X 8 Ĝ aQ+BQ+bi obX JmHiBTH2 lMB+bi, T2`7Q`KM+2 K2i`B+b +QMM2+i2/ iQ i?2 bK2 a.L ;i2rvX >2M+2- KQ`2 /i
mM/2` /Bz2`2Mi bBKmHiBQM b2iiBM;b 7Q` a+2M`BQ .X h?2 /Bz2`2Mi
bBKmHiBQM `mMb U7`QK φ iQ ϨV +Q``2bTQM/ iQ i?2 /Bz2`2Mi [email protected] T+F2ib i`p2H BMiQ i?2 M2irQ`F- 2bT2+BHHv 7Q` i?2 J@
Q7 h#H2 kX lMB+bi bQHmiBQMX am+?  i`2M/ Bb KQ`2 `2K`F#H2 BM
a+2M`BQ *- /m2 iQ i?2 H`;2` bT`2/ Q7 /2biBMiBQMb Qp2`
i?2 iQTQHQ;vX  bBKBH` i`2M/ Bb Q#b2`p2/ 7Q` a+2M`BQ .
BM 6B;X 8U#VX
LQi bm`T`BbBM;Hv- BKT`Qp2K2Mib Q7 aQ+BQ+bi rX`XiX J@
lMB+bi `2 ;`2i2` BM a+2M`BQ " +QKT`2/ iQ a+2M`BQ
*X AM/22/- BM a+2M`BQ " i?2 Ti? iQr`/b HH BMi2M/2/
/2biBMiBQMb Bb i?2 bK2 7`QK i?2 bQm`+2 iQ i?2 a.L
250
Sociocast, Scenario A
Sociocast, Scenario B
Sociocast, Scenario C ;i2rvX >2M+2- BM aQ+BQ+bi- i?2 a.L +QMi`QHH2` Dm/B@
Number of sociocast packets

+BQmbHv Bbbm2b `mH2b i?i T`2p2Mi 7`QK 7Q`r`/BM; /mTHB@

200 M-Unicast, Scenario A
M-Unicast, Scenario B
M-Unicast, Scenario C
+i2/ T+F2ib Qp2` i?2 bK2 HBMFbX
150

100
3X .Aa*laaAPL L. *PL*GlaAPL
AM i?Bb TT2` r2 ?p2 T`QTQb2/ M/ MHvx2/ i?2 #2@
50 ?pBQm` Q7 M `+?Bi2+im`H 7`K2rQ`F 2M+QKTbbBM; HH
i?2 2MiBiB2b- 7mM+iBQMHBiB2b- M/ T`Q+2/m`2b i?i bmTTQ`i
0
5 10 15 20
 7`2b? M2r M2irQ`F@Hv2` ;`QmT /Bbb2KBMiBQM K2i?Q/-
Number of destinations BX2X- aQ+BQ+bi- #v H2p2`;BM;  bQ7ir`2@/2}M2/ M2irQ`F
TT`Q+?X
6B;X e Ĝ aQ+BQ+bi obX JmHiBTH2 lMB+bi, 1t+?M;2/ /i THM2 _2bmHib +?B2p2/ i?`Qm;? M 2KmHiBQM i2bi#2/ b?Qr i?2
T+F2ib r?2M p`vBM; i?2 MmK#2` Q7 `2+2Bp2`b- BM /Bz2`2Mi b+2M`@
BQbX
#2ii2` b+H#BHBiv Q7 i?2 T`QTQbH BM i2`Kb Q7 P6 bB;@
MHBM; M/ /i T+F2i `2/mM/M+v r?2M +QKT`2/ iQ
M TTHB+iBQM@Hv2` #2M+?K`F b+?2K2- mM/2` /Bz2`2Mi
`2T`2b2MiiBp2 AQh b+2M`BQbX

50 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

AKT`Qp2K2Mib `2 +?B2p2/ #v H2p2`;BM;  Tm`TQb2@ deNR9RVX

#mBHi M2irQ`F TTHB+iBQM BM i?2 +QMi`QHH2` UBV BM +?`;2
Q7 B/2MiB7vBM; i?2 b2i Q7 aQ+BQ+bi /2biBMiBQMb #v BM@ _161_1L*1a
i2`+iBM; rBi? M 2ti2`MH aAQh THi7Q`K U72im`2 BK@
TH2K2Mi2/ i i?2 TTHB+iBQM Hv2` #v i?2 #2M+?K`F (R) X H@6m[?- JX :mBxMB- JX JQ?KK/B-
b+?2K2V M/ UBV `2bTQMbB#H2 7Q` bK`iHv #mBH/BM; `QmiBM; JX H2/?`B- M/ JX vvb?- dzAMi2`M2i Q7 i?BM;b, 
Ti?b iQr`/b KmHiBTH2 `2+2Bp2`b bQ b iQ pQB/ T+F2i bm`p2v QM 2M#HBM; i2+?MQHQ;B2b- T`QiQ+QHb- M/ T@
/mTHB+iBQM Qp2` HBMFbX a.L HHQrb iQ KM;2 i?2 BKTH2@ THB+iBQMb-Ǵ A111 *QKKmMB+iBQMb am`p2vb  hm@
K2MiiBQM Q7 bm+? 7mM+iBQMHBiB2b i i?2 +QMi`QH THM2 BM iQ`BHb- pQHX Rd- MQX 9- TTX kj9dĜkjde- kyR8X
 ~2tB#H2 M/ T`Q;`KK#H2 KMM2`- rBi? MQ +?M;2b (k) *X .BQi- "X LX G2pBM2- "X GvH2b- >X Ebb2K- M/
BM i?2 7Q`r`/BM; 2H2K2Mib- ?2M+2 KFBM; i?2 /2pBb2/ .X "H2MbB272M- dz.2THQvK2Mi Bbbm2b 7Q` i?2 AS KmH@
7`K2rQ`F T`+iB+HHv pB#H2 i  HQr BKTH2K2MiiBQM iB+bi b2`pB+2 M/ `+?Bi2+im`2-Ǵ A111 M2irQ`F-
+QbiX pQHX R9- MQX R- TTX d3Ĝ33- kyyyX
"2M2}ib Q7 i?2 T`QTQbH `2 /2}MBi2Hv H`;2 r?2M #B;
;`QmTb Q7 /2biBMiBQM /2pB+2b `2 +Hmbi2`2/ iQ;2i?2`- b (j) GX ixQ`B- X A2`- M/ :X JQ`#BiQ- dzaQ+BQ+bi, 
rBiM2bb2/ #v `2bmHib `272``BM; iQ a+2M`BQ ", i?2 P6 bB;@ M2r M2irQ`F T`BKBiBp2 7Q` AQh-Ǵ A111 *QKKmMB@
MHBM; Bb `2/m+2/ #v  7+iQ` ?B;?2` i?M Ry M/ i?2 +iBQMb J;xBM2- pQHX 8d- MQX e- TTX ekĜed- kyRNX
MmK#2` Q7 2t+?M;2/ /i T+F2ib b?`BMFb #v KQ`2 i?M
(9) .X E`2mix- 6X JX _KQb- SX 1X o2`BbbBKQ-
 7+iQ` Q7 8 U7Q` ir2Miv /2biBMiBQMbVX h?2 HQr2` ;BMb
*X 1X _Qi?2M#2`;- aX xQ/QHKQHFv- M/ aX l?@
7Q` aQ+BQ+bi T+F2ib rX`XiX P6 bB;MHBM; `2 /m2 iQ i?2
HB;- dzaQ7ir`2@/2}M2/ M2irQ`FBM;,  +QKT`2?2M@
7+i i?i aQ+BQ+bi HbQ `2bQ`ib iQ KmHiBTH2 mMB+bib 7Q`@
bBp2 bm`p2v-Ǵ S`Q+22/BM;b Q7 i?2 A111- pQHX Ryj-
r`/BM; BM i?2 Hbi ?QT 7`QK i?2 a.L ;i2rv iQr`/b
MQX R- TTX R9Ĝde- kyR8X
i?2 BMi2M/2/ /2biBMiBQMb- iQ 2Mbm`2 bm++2bb7mH `2+2TiBQM
i i?2 TTHB+iBQM Hv2`X Ai +M #2 7m`i?2` 2bBHv BM72``2/ (8) "X GMix- "X >2HH2`- M/ LX J+E2QrM- dz M2irQ`F
UHi?Qm;? MQi b?QrM BM `2bmHibV i?i BKT`Qp2K2Mib ;2i BM  HTiQT, `TB/ T`QiQivTBM; 7Q` bQ7ir`2@/2}M2/
2p2M H`;2` b i?2 /BbiM+2 #2ir22M i?2 bQm`+2 M/ i?2 M2irQ`Fb-Ǵ BM S`Q+X Q7 i?2 Ni? *J aA:*PJJ
b2i Q7 /2biBMiBQMb BM+`2b2bX qQ`Fb?QT QM >Qi hQTB+b BM L2irQ`Fb- TX RN- *J-
Pp2`HH- i?2 T`QTQbH Bb 2bT2+BHHv bmBi2/ 7Q` Tmb?@#b2/ kyRyX
/i /Bbb2KBMiBQM iQ H`;2 aQ+BQ+bi ;`QmTb ?B;?Hv +Hmb@
i2`2/ M/ 7` 7`QK i?2 bQm`+2- r?B+? r2HH `2b2K#H2b i?2 (e) dzPLXG"- dzAMi`Q/m+BM; PLPa @  a.L M2irQ`F
+b2 Q7 KmHiBTH2 /2pB+2b Q7  bK`i ?QK2 U2X;X- TTHB@ QT2`iBM; bvbi2K 7Q` b2`pB+2 T`QpB/2`b-Ǵ kyR9X
M+2bV iQ #2 `2KQi2Hv +QM};m`2/ #v i?2 mb2`Ƕb bK`i@ (d) aX@>X a?2M- dz1{+B2Mi ao* KmHiB+bi bi`2KBM; 7Q`
T?QM2X pB/2Q +QM72`2M+BM; rBi? a.L +QMi`QH-Ǵ A111 h`Mb@
AM i?2 Qi?2` +b2b- i?2 ;BMb `2 HbQ bB;MB}+Mi M/ H@ +iBQMb QM L2irQ`F M/ a2`pB+2 JM;2K2Mi- kyRNX
rvb ?B;?2` i?M  7+iQ` Q7 kX
h?2 +?B2p2/ 2M+Qm`;BM; `2bmHib KQiBpi2 mb iQ 7m`i?2` (3) "X EMmibbQM- >X Gm- qX sm- M/ "X >QTFBMb- dzS22`@
2tTHQ`2 i?Bb 72`iBH2 `2b2`+? `2 r?B+? ?b H`;2 `QQK iQ@T22` bmTTQ`i 7Q` KbbBp2Hv KmHiBTHv2` ;K2b-Ǵ
7Q` BKT`Qp2K2MibX h?2 2z2+iBp2M2bb Q7 i?2 T`QTQbH BM BM A111 AL6P*PJ kyy9- pQHX R- A111- kyy9X
?M/HBM; Qi?2` aQ+BQ+bi 72im`2b- HBF2 bQm`+2 }Hi2`BM;
(N) sX aX amM- uX sB- aX .xBMK`B`- sX aX >mM;-
M/ Tm#HBb?fbm#b+`B#2- M22/b iQ #2 T`+iB+HHv 2tTHQ`2/X
.X qm- M/ hX 1X L;- dz_2Tm#HB+, .i KmHiB+bi
b  7m`i?2` +?HH2M;2- AQh /2pB+2b #2HQM;BM; iQ aQ+BQ@
K22ib ?v#`B/ `+F@H2p2H BMi2`+QMM2+iBQMb BM /i
+bi ;`QmTb Kv KQp2 HQM; /BbiM+2b #2ir22M /Bz2`@
+2Mi2`-Ǵ BM kyR3 A111 kei? AMi2`MiBQMH *QM@
2Mi ++2bb TQBMibX >2M+2- i`+FBM; i?2B` TQbBiBQMb i
72`2M+2 QM L2irQ`F S`QiQ+QHb UA*LSV- TTX ddĜ3d-
i?2 pB`imH +QmMi2`T`ib UaoL M/ aoL_V- b r2HH b
A111- kyR3X
KM;BM; i?2 7Q`r`/BM; `mH2b bbQ+Bi2/ iQ i?2K BM
i?2 a.L MQ/2b- #2+QK2 p2`v /B{+mHi M/ 2MiBH T`QT2` (Ry) sX GB- uX@*X hBM- :X G2/rB+?- uX JBb?`- sX >M-
rQ`F`QmM/b r?B+? rBHH #2  bm#D2+i Kii2` Q7 7mim`2 M/ *X w?Qm- dz*QMbi`BM2/ QTiBKBxiBQM Q7 KmHiB@
BMp2biB;iBQMbX +bi `QmiBM; 7Q` rB/2 `2 +QMi`QH Q7 bK`i ;`B/-Ǵ
A111 h`Mb+iBQMb QM aK`i :`B/- kyR3X
*ELPqG1.:1J1Lh
(RR) JX >Qbb2BMB- .X hX ?K2/- aX a?B`KQ?KK/B- M/
h?Bb rQ`F rb T`iBHHv bmTTQ`i2/ #v i?2 1m`QT2M LX .X :2Q`;Mb- dz bm`p2v Q7 TTHB+iBQM@Hv2`
lMBQMǶb >Q`BxQM kyky `2b2`+? M/ BMMQpiBQM T`Q@ KmHiB+bi T`QiQ+QHb-Ǵ A111 *QKKmMB+iBQMb am`@
;`K mM/2` i?2 *P:@GP T`QD2+i U;`Mi ;`22K2Mi MQX p2vb M/ hmiQ`BHb- pQHX N- MQX R@9- TTX 83Ĝd9- kyydX

© International Telecommunication Union, 2020 51

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

(Rk) aX AbHK- LX JmbHBK- M/ CX qX irQQ/- dz bm`@ (kj) EX q2B- sX GBM;- M/ EX sm- dz bm`p2v Q7
p2v QM KmHiB+biBM; BM bQ7ir`2@/2}M2/ M2irQ`F@ bQ+BH@r`2 `QmiBM; T`QiQ+QHb BM /2Hv iQH2`Mi M2i@
BM;-Ǵ A111 *QKKmMB+iBQMb am`p2vb  hmiQ`BHb- rQ`Fb, TTHB+iBQMb- itQMQKv M/ /2bB;M@`2Hi2/
pQHX ky- MQX R- TTX j88Ĝj3d- kyRdX Bbbm2b-Ǵ A111 *QKKmMB+iBQMb am`p2vb  hmiQ`B@
Hb- pQHX Re- MQX R- TTX 88eĜ8d3- kyRjX
(Rj) wX Ha22/- AX ?K/- M/ AX >mbbBM- dzJmHiB+bi@
BM; BM bQ7ir`2 /2}M2/ M2irQ`Fb,  +QKT`2?2MbBp2 (k9) JX sBQ- CX qm- M/ GX >mM;- dz*QKKmMBiv@r`2
bm`p2v-Ǵ CQm`MH Q7 L2irQ`F M/ *QKTmi2` TTHB@ QTTQ`imMBbiB+ `QmiBM; BM KQ#BH2 bQ+BH M2irQ`Fb-Ǵ
+iBQMb- pQHX Ry9- TTX eRĜdd- kyR3X A111 h`Mb+iBQMb QM *QKTmi2`b- pQHX ej- MQX d-
TTX Re3kĜReN8- kyRjX
(R9) X /Kb- CX LB+?QHb- M/ qX aB/F- dz_6* jNdj-
T`QiQ+QH BM/2T2M/2Mi KmHiB+bi@/2Mb2 KQ/2 USAJ@ (k8) .X CX qiib- dzL2irQ`Fb- /vMKB+b- M/ i?2 bKHH@
.JV, S`QiQ+QH bT2+B}+iBQM U`2pBb2/V-Ǵ i2+?X `2TX- rQ`H/ T?2MQK2MQM-Ǵ K2`B+M CQm`MH Q7 bQ+BQH@
kyy8X Q;v- pQHX Ry8- MQX k- TTX 9NjĜ8kd- RNNNX

(R8) "X *BM- aX .22`BM;- AX EQmp2Hb- "X 62MM2`- M/ (ke) qX :Q- ZX GB- "X w?Q- M/ :X *Q- dzJmHiB@
X h?v;`DM- dz_6* jjde- AMi2`M2i :`QmT JM@ +biBM; BM /2Hv iQH2`Mi M2irQ`Fb,  bQ+BH M2i@
;2K2Mi S`QiQ+QH- p2`bBQM j-Ǵ i2+?X `2TX- m;mbi rQ`F T2`bT2+iBp2-Ǵ BM S`Q+22/BM;b Q7 *J JQ#B@
kyyeX >Q+- TTX kNNĜjy3- *J- kyyNX

(Re) >X >QH#`QQF M/ "X *BM- dz_6* 9eyd- aQm`+2@ (kd) qX :Q- ZX GB- "X w?Q- M/ :X *Q- dzaQ+BH@
aT2+B}+ JmHiB+bi 7Q` AS-Ǵ i2+?X `2TX- m;mbi kyyeX r`2 KmHiB+bi BM /Bb`mTiBQM@iQH2`Mi M2irQ`Fb-Ǵ
A111f*J h`Mb+iBQMb QM L2irQ`FBM; UhPLV-
(Rd) X o2MFi`KMB 2i HX- dzJQ#BHBiv6B`bi,  pQHX ky- MQX 8- TTX R88jĜR8ee- kyRkX
KQ#BHBiv@+2Mi`B+ M/ i`mbirQ`i?v BMi2`M2i `+?Bi2+@
im`2-Ǵ *J aA:*PJJ *QKTmi2` *QKKmMB+iBQM (k3) sX >m- hX >X *?m- oX *X G2mM;- 1X *X@>X L;B-
_2pB2r- pQHX 99- MQX j- TTX d9Ĝ3y- kyR9X SX E`m+?i2M- M/ >X *X *?M- dz bm`p2v QM KQ#BH2
bQ+BH M2irQ`Fb, TTHB+iBQMb- THi7Q`Kb- bvbi2K
(R3) *X X J`+QM/2b- hX SX aMiQb- X SX :Q/Qv- `+?Bi2+im`2b- M/ 7mim`2 `2b2`+? /B`2+iBQMb-Ǵ A111
*X *X oB2H- M/ *X X h2Bt2B`- dz*bi~Qr, *H2M@ *QKKmMB+iBQMb am`p2vb  hmiQ`BHb- pQHX Rd- MQX j-
bHi2 KmHiB+bi TT`Q+? mbBM; BM@/pM+2 Ti? T`Q@ TTX R88dĜR83R- kyR8X
+2bbBM; BM T`Q;`KK#H2 M2irQ`Fb-Ǵ BM kyRk A111
avKTQbBmK QM *QKTmi2`b M/ *QKKmMB+iBQMb (kN) CX 6M- CX *?2M- uX .m- qX :Q- CX qm- M/ uX amM-
UAa**V- TTX yyyyN9ĜyyyRyR- A111- kyRkX dz:2Q+QKKmMBiv@#b2/ #`Q/+biBM; 7Q` /i /Bb@
b2KBMiBQM BM KQ#BH2 bQ+BH M2irQ`Fb-Ǵ A111 h`Mb@
(RN) JX w?Q- "X CB- JX qm- >X um- M/ uX sm- +iBQMb QM S`HH2H M/ .Bbi`B#mi2/ avbi2Kb- pQHX k9-
dzaQ7ir`2 /2}M2/ M2irQ`F@2M#H2/ KmHiB+bi 7Q` MQX 9- TTX dj9Ĝd9j- kyRjX
KmHiB@T`iv pB/2Q +QM72`2M+BM; bvbi2Kb-Ǵ BM kyR9
A111 AMi2`MiBQMH *QM72`2M+2 QM *QKKmMB+@ (jy) GX ixQ`B- X A2`- :X JQ`#BiQ- M/ JX LBiiB- dzh?2
iBQMb UA**V- TTX RdkNĜRdj8- A111- kyR9X bQ+BH BMi2`M2i Q7 i?BM;b UaAQhVĜr?2M bQ+BH M2i@
rQ`Fb K22i i?2 BMi2`M2i Q7 i?BM;b, *QM+2Ti- `+?B@
(ky) .X Eib`Qb- LX .BKQFb- M/ GX hbbBmHb- dzaQ+BH i2+im`2 M/ M2irQ`F +?`+i2`BxiBQM-Ǵ *QKTmi2`
M2irQ`F MHvbBb +QM+2Tib BM i?2 /2bB;M Q7 rB`2H2bb M2irQ`Fb- pQHX 8e- MQX Re- TTX j8N9Ĝjey3- kyRkX
/ ?Q+ M2irQ`F T`QiQ+QHb-Ǵ A111 M2irQ`F- pQHX k9-
MQX e- TTX kjĜkN- kyRyX (jR) LX J+E2QrM- hX M/2`bQM- >X "HF`Bb?MM-
:X S`mHF`- GX S2i2`bQM- CX _2t7Q`/- aX a?2MF2`-
(kR) uX w?m- "X sm- sX a?B- M/ uX qM;- dz bm`p2v M/ CX hm`M2`- dzPT2M6HQr, 2M#HBM; BMMQpiBQM
Q7 bQ+BH@#b2/ `QmiBM; BM /2Hv iQH2`Mi M2irQ`Fb, BM +KTmb M2irQ`Fb-Ǵ *J aA:*PJJ *QKTmi2`
SQbBiBp2 M/ M2;iBp2 bQ+BH 2z2+ib-Ǵ A111 *QK@ *QKKmMB+iBQM _2pB2r- pQHX j3- MQX k- TTX eNĜd9-
KmMB+iBQMb am`p2vb  hmiQ`BHb- pQHX R8- MQX R- kyy3X
TTX j3dĜ9yR- kyRkX
(jk) GX ixQ`B- *X *KTQHQ- "X .- _X :B`m- X A2`-
(kk) EX qX 2i HX- dz1tTHQBiBM; bKHH rQ`H/ T`QT2`iB2b :X JQ`#BiQ- M/ aX Zmii`QTMB- dz1M?M+BM; B/2M@
7Q` K2bb;2 7Q`r`/BM; BM /2Hv iQH2`Mi M2irQ`Fb-Ǵ iB}2`fHQ+iQ` bTHBiiBM; i?`Qm;? bQ+BH BMi2`M2i Q7
A111 h`Mb+iBQMb QM *QKTmi2`b- pQHX e9- MQX Ry- i?BM;b-Ǵ A111 AMi2`M2i Q7 h?BM;b CQm`MH- pQHX e-
TTX k3yNĜk3R3- kyR8X MQX k- TTX kNd9ĜkN38- kyR3X

52 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

(jj) dzPT2M~Qr brBi+? bT2+B}+iBQM @ p2`bBQM RXjXR PT2M *Hm/B *KTQHQ Ua2MBQ`
L2irQ`FBM; 6QmM/iBQM UPL6V-Ǵ a2Ti2K#2` kyRkX J2K#2`- A111V `2+2Bp2/
i?2 Kbi2`Ƕb M/ S?X.X
(j9) CX >mB M/ SX h?m#2`i- dz*QKT`2bbBQM 7Q`Ki 7Q` /2;`22b BM i2H2+QKKmMB+@
ASpe /i;`Kb Qp2` B222 3ykXR8X9@#b2/ M2irQ`Fb- iBQMb 2M;BM22`BM; 7`QK i?2
_6* ek3k-Ǵ a2Ti2K#2` kyRRX J2/Bi2`@ `M2 lMBp2`bBiv Q7
(j8) X v/B- .X _Qb- M/ GX hQmiBM- dzh*S ?2/2` _2;;BQ *H#`B- BM kyyd
+QKT`2bbBQM 7Q` eGQqSL-Ǵ AMi2`M2i .`7i U/`7i@ M/ kyRR- `2bT2+iBp2HvX AM
v/B@QHQrTM@i+T?+@yyV- rQ`F BM T`Q;`2bb- kyRyX kyy3- b?2 rb  oBbBiBM;
S?X.X aim/2Mi rBi? i?2 SQ@
(je) *X :QK2x- X `+B@JQ`2i- M/ CX *`Qr+`Q7i- dzh*S HBi2+MB+Q /B hQ`BMQ M/ 
BM i?2 AMi2`M2i Q7 h?BM;b, 7`QK Qbi`+BbK iQ T`QKB@ .. 62HHQr rBi? i?2 lMB@
M2M+2-Ǵ A111 AMi2`M2i *QKTmiBM;- pQHX kk- MQX R- p2`bBiv Q7 S/2`@ #Q`M- :2`KMv- BM kyR8X a?2 Bb
TTX kNĜ9R- kyR3X +m``2MiHv M bbQ+Bi2 S`Q72bbQ` Q7 i2H2+QKKmMB+iBQMb
rBi? i?2 J2/Bi2``M2 lMBp2`bBiv Q7 _2;;BQ *H#`BX
(jd) GX ixQ`B- *X *KTQHQ- "X .- _X :B`m- X A2`- >2` KBM `2b2`+? BMi2`2bib BM+Hm/2 p2?B+mH` M2irQ`F@
:X JQ`#BiQ- M/ aX Zmii`QTMB- dzaK`i /2pB+2b BM BM;- 8:- M/ 2/;2 +QKTmiBM;X
i?2 bQ+BH HQQTb, *`Bi2`B M/ H;Q`Bi?Kb 7Q` i?2 +`2@
iBQM Q7 i?2 bQ+BH HBMFb-Ǵ 6mim`2 :2M2`iBQM *QK@ MiQMBQ A2` Bb T`Q72b@
Tmi2` avbi2Kb- pQHX Nd- TTX jkdĜjjN- kyRNX bQ` Q7 h2H2+QKKmMB+iBQMb
i i?2 lMBp2`bBiv Q7 *H@
lh>P_a #`B- AiHvX >2 ;`/m@
GmB;B ixQ`B US?.- kyyyV i2/ BM *QKTmi2` 1M;BM22`@
Bb T`Q72bbQ` Q7 h2H2+QKKm@ BM; i i?2 lMBp2`bBiv Q7 *H@
MB+iBQMb i i?2 lMBp2`@ #`B- AiHv- M/ `2+2Bp2/ 
bBiv Q7 *;HB`B- r?2`2 ?2 Jbi2` .BTHQK BM AM7Q`K@
H2/b i?2 +iBpBiB2b Q7 i?2 iBQM h2+?MQHQ;v 7`QK *1@
J*G# H#Q`iQ`v UJmH@ 6_A1GfSQHBi2+MB+Q /B JB@
iBK2/B *QKKmMB+iBQMbV HMQ- AiHv- M/  S?X.X /2@
rBi? `QmM/ ky {HBi2/ `2@ ;`22 7`QK i?2 lMBp2`bBiv Q7
b2`+?2`bX aBM+2 kyR3- ?2 *H#`BX 6`QK RNN9 iQ
?b #22M i?2 +QQ`/BMiQ` Q7 RNN8 ?2 ?b #22M rBi? aB2K2Mb : BM JmMB+?- :2`KMv-
i?2 Kbi2` /2;`22 +Qm`b2 BM M/ 7`QK RNNd iQ kyRN rBi? i?2 lMBp2`bBiv Q7 _2;;BQ *H@
AMi2`M2i h2+?MQHQ;v 1M;B@ #`BX >2 ?b Tm#HBb?2/ KQ`2 i?M jyy TT2`b BM ?B;?@
M22`BM; i i?2 lMBp2`bBiv Q7 *;HB`BX >Bb `2b2`+? BM@ [mHBiv DQm`MHb M/ +QM72`2M+2b M/ ?b ;Bp2M b2p2`H
i2`2bib 7HH BM i?2 `2 Q7 AMi2`M2i Q7 h?BM;b UAQhV- rBi? hmiQ`BHb M/ BMpBi2/ bT22+?2b i BMi2`MiBQMH 2p2Mib
T`iB+mH` `272`2M+2 iQ i?2 /2bB;M Q7 2z2+iBp2 H;Q`Bi?Kb QM i?2 iQTB+b Q7 AQh- aQ+BH@AQh- M/ 8: M2irQ`FbX >2 Bb
7Q` i?2 `2HBxiBQM Q7 bQ+BH M2irQ`Fb KQM; +QMM2+i2/ HbQ b2`pBM; b 1/BiQ` BM *?B27 7Q` *QKTmi2` L2irQ`Fb-
/2pB+2b iQ +`2i2 i?2 aQ+BH AQh T`/B;KX >Bb BMi2`2bib 1Hb2pB2`X>Bb `2b2`+? BMi2`2bib BM+Hm/2 rB`2H2bb M/ KQ@
HbQ 7HHb BM i?2 `2 Q7 ZmHBiv Q7 1tT2`B2M+2 UZQ1V- #BH2 8: M2irQ`Fb M/ AMi2`M2i Q7 h?BM;bX
rBi? T`iB+mH` TTHB+iBQM iQ i?2 KM;2K2Mi Q7 b2`@
:Bmb2TT2 JbbBKBH@
pB+2b M/ `2bQm`+2b BM M2r ;2M2`iBQM M2irQ`Fb 7Q` KmHiB@
BMQ JBHQii rb #Q`M
K2/B +QKKmMB+iBQMbX Gi2Hv- ?2 HbQ TTHB2b i?2 bim/v
BM *iMB- aB+BHv UAiHvV
Q7 ZQ1 iQ AQh b2`pB+2bX >2 b2`p2b `2;mH`Hv BM i?2 +QM72`@
QM Jv yd- RN3eX >2
2M+2 Q`;MBxBM; +QKKBii22 Q7 i?2 b2+iQ` M/ b bbQ+Bi2
`2+2Bp2/ ?Bb Kbi2` /2;`22
M/ ;m2bi 2/BiQ` Q7 b2p2`H BMi2`MiBQMH DQm`MHb UA111
i ǴlMBp2`bBi¨ /2;HB bim/B
AQh DQm`MH- / >Q+ L2irQ`Fb- A111 A111 PT2M CQm`@
/B *iMBǴ BM kyRd 7i2`
MH Q7 i?2 *QKKmMB+iBQMb aQ+B2iv- A111 *QKKmMB+@
 bBt@KQMi? +QHH#Q`iBQM
iBQMb J;xBM2- 2i+XVX
rBi? ǴhûHû+QK S`Bb h2+?Ǵ-
S`BbU6`M+2V BM kyRe-
r?B+? H2/ iQ i?2 `2HBxiBQM Q7 ?Bb Kbi2` /2;`22 i?2bBbX
>2 Bb +m``2MiHv 2M`QHH2/ i i?2 Hbi v2` Q7 S?. BM
AM7Q`KiBQM 1M;BM22`BM; i ǴlMBp2`bBi¨ J2/Bi2``M2 /B

© International Telecommunication Union, 2020 53

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

_2;;BQ *H#`BǴ- _2;;BQ *H#`B UAiHvV r?2`2 Bb biBHH

+QQT2`iBM; rBi? i?2 *LAh `2b2`+? mMBi Q7 *iMBX
>Bb KBM `2b2`+? +iBpBiB2b 7Q+mb QM a.L M/ i?2 AQhǶb
rQ`H/- rBi?  T`iB+mH` BMi2`2bi BM i?2 lob M/ i?2
b2+m`Biv bvbi2KbX

:B+QKQ JQ`#BiQ `2@

+2Bp2/ i?2 Hm`2 /2;`22
+mK Hm/2 BM 1H2+i`QMB+
1M;BM22`BM; M/ i?2 S?X.
7`QK i?2 lMBp2`bBiv Q7 *i@
MB UAiHvV BM RNNe M/ kyyy-
`2bT2+iBp2HvX aBM+2 RNNN
iQ kyyR ?2 rb  `2b2`+?
2M;BM22` i :2Q`;B h2+?
UiHMi- laVX aBM+2 kyyR
?2 Bb rBi? i?2 .2T`iK2Mi
Q7 1H2+i`B+- 1H2+i`QMB+- M/
*QKTmi2` 1M;BM22`BM; Q7 i?2 lMBp2`bBiv Q7 *iMB
r?2`2 ?2 Bb +m``2MiHv 7mHH T`Q72bbQ` Q7 i2H2+QKKmMB+@
iBQMbX >Bb `2b2`+? BMi2`2bib 7Q+mb QM MHvbBb M/ /2bB;M
Q7 rB`2H2bb M2irQ`Fb M/ AMi2`M2i Q7 h?BM;bX
aHpiQ`2 Zmii`QTMB
`2+2Bp2/ i?2 "a+X M/
Ja+X /2;`22b BM *QKTmi2`
1M;BM22`BM; BM kyRe M/
kyRd- `2bT2+iBp2Hv- 7`QK
lMBp2`bBiv Q7 *iMB-
AiHvX aBM+2 m;mbi kyRd
?2 Bb rBi? i?2 *QMbQ`xBQ
LxBQMH2 AMi2`mMBp2`bBi`BQ
T2` H2 h2H2+QKmMB+xBQMB b
 _2b2`+? 1M;BM22`X >Bb `2b2`+? BMi2`2bib 7Q+mb QM
AQh +QMM2+iBpBiv M/ +QKTmiBM; TT`Q+?2bX

54 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

THE INTERNET OF METAMATERIAL THINGS AND THEIR SOFTWARE

ENABLERS

Christos Liaskos1,2 , Georgios G. Pyrialakos2,3 , Alexandros Pitilakis2,3 , Ageliki Tsioliaridou2 , Michail

Christodoulou2,3 , Nikolaos Kantartzis2,3 , Sotiris Ioannidis4,2 , Andreas Pitsillides5 , Ian F. Akyildiz5
1
Computer Science Engineering Dept., University of Ioannina, Ioannina, Greece, 2 Foundation for Research and
Technology - Hellas, Heraklion, Greece, 3 Electrical and Computer Engineering Dept., Aristotle University, Greece,
4
Technical University of Chania, Crete, Greece, 5 Computer Science Dept., University of Cyprus, Nicosia, Cyprus,
NOTE: Corresponding author: Christos Liaskos, [email protected]

Abstract – A new paradigm called the Internet of MetaMaterial Things (IoMMT) is introduced in this paper where
artificial materials with real-time tunable physical properties can be interconnected to form a network to realize com-
munication through software-controlled electromagnetic, acoustic, and mechanical energy waves. The IoMMT will
significantly enrich the Internet of Things ecosystem by connecting anything at any place by optimizing the physical
energy propagation between the metamaterial devices during their lifetime, via “eco-firmware” updates. First, the means
for abstracting the complex physics behind these materials are explored, showing their integration into the IoT world.
Subsequently, two novel software categories for the material things are proposed, namely the metamaterial Application
Programming Interface and Metamaterial Middleware, which will be in charge of the application and physical domains,
respectively. Regarding the API, the paper provides the data model and workflows for obtaining and setting the physical
properties of a material via callbacks. The Metamaterial Middleware is tasked with matching these callbacks to the
corresponding material-altering actuations through embedded elements. Furthermore, a full stack implementation of
the software for the electromagnetic metamaterial case is presented and evaluated, incorporating all the aforementioned
aspects. Finally, interesting extensions and envisioned use cases of the IoMMT concept are discussed.
Keywords – Internet of Things, metamaterials, programming interface, software-defined networking.

1. INTRODUCTION that has enabled the creation of artificial materials with

real-time tunable physical properties [2, 3]. Metama-
Recent years have witnessed the advent of the Internet- terials are based on the fundamental idea stating that
of-Things (IoT), denoting the interconnection of every the physical properties of matter stem from its atomic
electronic device and the smart, orchestrated automa- structure. Therefore, one can create artificially struc-
tion it entails [1]. Vehicles, smart phones, sensors, home tured materials (comprising sufficiently small elemen-
and industrial appliances of any kind expose a function- tary “units” of composition and geometry) to yield any
ality interface expressed in software, allowing for devel- required energy manipulating behavior, including types
opers to create end-to-end workflows. As an upshot, not found in natural materials. Metamaterials manipu-
smart buildings and even smart cities that automatically lating electromagnetic (EM) energy were the first kind
adapt, e.g., power generation, traffic and heat manage- of metamaterials to be studied in depth, mainly due
ment to the needs of residents, have been devised in to the relative ease of manufacturing as low-complexity
recent years. This current IoT potential stems from ex- electronic boards [3].
posing and controlling a high-level functionality of an
Going beyond EM waves, the collectively termed elas-
electronic device, such as turning on/off lights and air-
todynamic metamaterials can manipulate acoustic, me-
conditioning units based on the time of day and temper-
chanical and structural waves, whereas thermodynamic
ature. This paper proposes the expansion of the IoT to
and quantum-mechanic metamaterials have also been
the level of physical material properties, such as electri-
postulated [4]. Elastodynamic metamaterials, empow-
cal and thermal conductivity, mechanical elasticity, and
ered by recent advances in nano- and micro-fabrication
acoustic absorption. This novel direction is denoted as
(e.g. additive manufacturing/3D printing), can ex-
the Internet of MetaMaterial Things (IoMMT), and can
hibit effective/macroscopic nonphysical properties such
have groundbreaking potential across many industrial
as tunable stiffness and absorption/reflection, extreme
sectors, as outlined in this paper. There are two key
mass-volume ratios, negative sonic refraction, etc [5].
enablers for the proposed IoMMT:
Their cell-size spans several length scales, depending on
Key Enabler 1: the application: acoustic cloaking/anisotropy/isolation,
The first key enabler of the proposed IoMMT are the ultra-lightweight and resilient materials, devices for
metamaterials, the outcome of recent research in physics medical/surgical applications and food/drug adminis-

© International Telecommunication Union, 2020 55

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

tration, MEMS, anti-seismic structures, etc. Tunability Sensor Networked

IoT
MetaMaterials
Passive part Gateway
of elastodynamic metamaterials can be achieved with Actuate
Communicate
Actuator &
electric, magnetic, optical, thermal or chemical stimuli. Sense
EM Compute
In a nutshell, their operation is as follows: Impinging Behavior
EM waves create inductive currents over the material, Source
Label
SPLIT
NETWORK
which can be modified by tuning the actuator elements Source
Label
FOCUS API
within it (e.g., simple switches) accordingly. The Huy- Source
Label
ABSORB MIDDLEWARE
gens principle states that any EM wavefront departing Source
Label Alter
POLARIZATION
from a surface can be traced back to an equivalent cur- Source
Label
STEER
rent distribution over a surface [3]. Thus, in principle, Source
Label
PHASE Alter

metamaterials can produce any custom departing EM

wave as a response to any impinging wave, just by tun-
ing the state of embedded switches/actuators. Such EM
interactions are shown in Fig. 1 (on the right side). The Fig. 1 – Networked metamaterial structure and possible energy
wave interactions [8].
same principle of operation applies to mechanical, acous-
tic and thermal metamaterials [6].
Key Enabler 2: collimate()
absorb()
The second key enabler of the IoMMT is the concept
of networked metamaterials. These will come with an
application programming interface (API), an accompa-
nying software middleware and a network integration refract()
steer()

architecture that enable the hosting of any kind of en-

ergy manipulation over a metamaterial in real time
(e.g., steering, absorbing, splitting of EM, mechanical,
avoid()
thermal or acoustic waves), via simple software call- follow()

backs executed from a standard PC (desktop or lap- steer()

focus()
top), while abstracting the underlying physics. The goal steer()
focus()
is to constitute the IoMMT directly accessible to the steer()
focus()
IoT and software development industries, without car-
ing for the intrinsic and potentially complicated phys-
ical principles. Regarding the IoMMT potential, large
scale deployments of EM metamaterials in indoor setups
have introduced the groundbreaking concept of pro-
grammable/intelligent wireless environment (Fig. 2) [7]. Fig. 2 – The programmable wireless environment introduced
in [7], is created by coating walls with networked metamateri-
By coating all major surfaces in a space (e.g., indoors) als. This allows for customized wireless propagation-as-an-app per
with EM metamaterials, the wireless propagation can communicating device pair, introducing novel potential in data
be controlled and customized via software. As detailed rates, communication quality, security and wireless power trans-
in [7] this can enable the mitigation of path loss, fad- fer.
ing and Doppler phenomena, while also allowing waves ˆ The acoustic metamaterials can surround noisy
to follow improbable air-routes to avoid eavesdroppers devices or be applied on windows to provide a
(a type of physical-layer security). In cases where the more silent environment, but to also harvest energy
device beamforming and the EM metamaterials in the which can be added to a system such as a smart-
space are orchestrated together, intelligent wireless envi- household.
ronments can attain previously unattainable communi-
cation quality and wireless power transfer [7]. Extend- Assuming a central controller to optimize a given
ing the EM case, we envision the generalized IoMMT IoMMT deployment allows for further potential. For
deployed as structural parts of products, as shown in instance, one can allow for quickly “patching” of over-
Fig. 3: looked physical aspects (e.g., poor ecological perfor-
ˆ EM interference and unwanted emissions can be mance) of IoMM-enabled products during operation,
harvested by IoMM-coated walls and be trans- without overburdening the product design phase with
formed back to usable EM or mechanical energy. such concerns. The “patching” may also be deferred in
the form of “eco-firmware”, distributed via the Internet
ˆ Thermoelectric and mechanical metamaterials can to ecologically tune a single product or horizontal sets
micro-manage emanated heat and vibrations from of products.
devices, such as any kind of motor, to recycle it In this context, the principal contributions of the paper
as energy while effectively cooling it. The same are as follows:
principle can be applied to a smart household or a
noisy factory. ˆ We propose the concept of the IoMMT and discuss

56 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

agation within a space, in every physical domain, i.e.,

for any physical material property and corresponding
information-carrying wave. For instance, control over
the equivalent RLC parameters of an electric load con-
trols the power that can be delivered to it by an EM
wave. Moreover, the presented software is a mature
prototype platform for the development of IoMMT ap-
plications. This constitutes a major leap towards a new
research direction. On the other hand, other research
directions have proposed and explored the Internet of
NanoThings [9]. Although similarly named, these direc-
tions are not related to the IoMMT, as they are about
embedding nano-sized computers into materials in order
to augment the penetration level of applications (e.g.,
sense structural, temperature, humidity changes within
a material, rather than just over it, etc.), and not to
control the energy propagation within them.
The remainder of this paper is organized as follows,
devoting a section to each of the principal contribu-
tions of our work. In Section 2 we provide the related
work overview and the necessary prerequisite knowledge
(a) Conceptual IoMMT deployment within a smart home.
for networked metamaterial. In Section 3 we present
the architecture for integrating the IoMMT in exist-
ing Software-Defined Networks (SDNs) and systems. In
Section 4 we present the novel metamaterial API, and
Section 5 follows with the description of the Metamate-
rial Middleware and its assorted workflows. In Section 6
we present the implemented version of the software for
the EM metamaterial case, along with a description of
the employed evaluation test bed. Finally, novel real-
istic applications enabled with our new paradigm are
discussed in Section 7, and we conclude the paper in
Section 8.

(b) Conceptual IoMMT deployment within single products. 2. PREREQUISITES AND RELATED
WORK
Fig. 3 – Envisioned applications of the IoMMT in smart houses
and products. Metamaterials are simple structures that are created by
its architecture and interoperability with existing periodically repeating a basic structure, called a cell or a
network infrastructures. meta-atom [3]. Some examples across physical domains
are shown in Fig. 4. The planar (2D) assemblies of meta-
ˆ We define two novel categories of software: the atoms, known as metasurfaces, are of particular interest
Metamaterial API and the Metamaterial Middle- currently [10, 11]. For instance, EM are currently heav-
ware, which enable any software developer to in- ily investigated by the electromagnetic/high-frequency
teract with a set of networked metamaterials, in community, for novel communications, sensing and en-
a physics-agnostic manner. We establish the data ergy applications. [12–14].
models, workflows, and test bed processes re- A notable trait of metamaterials is that they are sim-
quired for profiling and, subsequently, componetiz- ple structures and, therefore, there exists a variety of
ing metamaterials. techniques for generally low-cost and scalable produc-
tion [3]. The techniques such as printed circuit broads,
ˆ We present an implemented and experimentally flexible materials such as Kapton, 3D printing, Large
verified version of the metamaterial API and the Area Electronics, bio-skins and microfluidics have been
Metamaterial Middleware for the EM case. successfully employed for manufacturing [3].
ˆ We highlight promising, new applications empow- In each physical domain, a properly configured metama-
ered by the featured IoMMT concept. terial has the capacity to steer and focus an incoming
energy wave towards an arbitrary direction or even com-
In this aspect, the potential of our IoMMT paradigm pletely absorb the impinging power. In the EM case this
is the first to offer true control over the energy prop- capability can be exploited for advanced wireless com-

© International Telecommunication Union, 2020 57

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

tion, which has not been proposed before. Additionally,

our paper focuses more on the networking approaches
for metamaterials, which has only been treated in our
previous work [8], and only for the EM case. Finally,
the work of Chen et al. [27] also advocates for the use
of metamaterial in any physical domain for distributed
energy harvesting, e.g., in a smart house or a city. How-
ever, software enablers and networking considerations
are not discussed or solved in [27]. Moreover, the energy
manipulation type is restricted to harvesting which can
be viewed as a subset of our proposed IoMMT potential.

Metamaterials: Principles of Operation, Classi-

fication and Supported Functionalities
A conceptual metamaterial is illustrated in Fig. 5 [3].
Basically, a metamaterial consists of periodically re-
peated meta-atoms arranged in a 3D grid layout, with
Fig. 4 – Energy manipulation domains of artificial materials: the metasurfaces being a sub-case. In particular, unit
(a) Electromagnetic [20] (b) Mechanical [21] (c) Acoustic [22] (d)
Thermoelectric [23].
cells comprise passive and tunable parts, required in
reconfigurable metamaterials as well as optional inte-
munications [7, 15–19], offering substantially increased grated sensory circuits, which can extract information
bandwidth and security between two communicating of the incident energy wave. Furthermore, tunable parts
parties. are crucial for metamaterials, as they enable reconfig-
The potential stemming from interconnected metama- urability and switching between different functions. For
terials has begun to be studied only recently [8]. The illustration, in EM metamaterials at microwave frequen-
perspective networking architecture and protocols [7,8], cies, the tunable parts embedded inside the unit cells
metamaterial control latency models [24], and smart en- can be voltage-controlled resistors (varistors) and/or ca-
vironment orchestration issues have been recently stud- pacitors (varactors), micro-electromechanical switches
ied for the EM case [25, 26]. (MEMS), to name a few [3, 19].
Notably, a similarly named concept, i.e., the Internet On the other hand, in mechanical and acoustic meta-
of NanoThings [9], was recently proposed to refer to materials, the tunable parts can be micro-springs with
materials with embedded, nano-sized computing and a tunable elasticity rate [28, 29]. The meta-atoms may
communicating elements. In general, these materials also form larger groups, called super-atoms or super-
are derived from miniaturizing electronic elements and cells, repeated in specific patterns that can serve more
placing them over or embedding them into fabrics and complex functionalities, as discussed later in this paper.
gadgets, to increase their application-layer capabilities. Lastly, the software-defined metamaterials include a
For instance, this could make a glass window become gateway [7], i.e., an on-board computer, whose main
a giant, self-powered touchpad for another IoT device. tasks are to: i) power the whole device and ii) control
Originally. the concept of software-defined metasurfaces (get/set) the state of the embedded tunable elements,
wa based on the nano-IoT as the actuation/control en- iii) interoperate with the embedded sensors, and iv)
abler [17]. Nano-devices can indeed act as the con- interconnect with the outside world, using well-known
trollers governing the state of the active cells, offer- legacy networks and protocol stacks (e.g. Ethernet).
ing manufacturing versatility and extreme energy effi- The relative size of a meta-atom compared to the wave-
ciency. Nonetheless, until nano-IoT becomes a main- length of the excitation (impinging wave) defines the en-
stream technology, other approaches can be adopted ergy manipulation precision and efficiency of a metama-
for manufacturing software-defined metasurfaces, as re- terial. For example, EM metasurfaces share many com-
ported in the related physics-oriented literature [6]. It is mon attributes with classic antenna-arrays and reflect-
also noted that nano-IoT as a general concept is about arrays. Antenna arrays can be viewed as independently
embedding nano-sized computers into materials in order operating antennas, being very effective for coarse beam
to augment the penetration level of applications (e.g., steering as a whole. Reflect-arrays typically consist of
sense structural, temperature, humidity changes within smaller elements (still subwavelength), permitting more
a material, rather than just over it, etc.), and not specif- fine-grained beam steering and a very coarse polariza-
ically to control the energy propagation within them. tion control. Metamaterials comprise orders of magni-
In contrast, our work refers specifically to the case of tude smaller meta-atoms, and may also include tunable
metamaterials and the capabilities they offer for the ma- elements and sensors. Their meta-atoms are generally
nipulation of energy across physical domains. Moreover, considered tiny with regard to the exciting wavelength,
our paper introduces the software enablers for this direc- hence allowing full control over the form of the departing

58 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Departing waves:
Directions, {D'(i)}.
Global switching
Frequency, {f'(i)}. frequency fsw Impinging wave:
Bandwidth, {B'(i)}. Direction, D.
Block size << 1/fmax
Amplitudes, {A'(i)}. Frequency, fmax.
Phases,{φ'(i)}. Bandwidth, B.
Polarizations, {J'(i)}. Amplitude, A.
if (fsw>fmax) Phase, φ.
Modulations, {M(i)'}. Polarization, J.
Modulation, M.

IoT Chipset-based
gateway, NICs, and
power supply
connection

Passive Commands
block Tunable
>STEER();
and/or >FOCUS();
sensory >FILTER();
IC >ABSORB();
Tunable block and Sensory >PHASE_ALTER();
IC-to-gateway
interconnection fabric offering Middleware >POLAR_ALTER();
Switch or Bus-type control >MODULATE();
>SENSE();

META-API

Fig. 5 – Overview of the metasurface/metamaterial structure and operating principles.

energy wave. ˆ Wavefront: Steering (reflecting or refracting), split-

Regardless of their geometry and composition, the oper- ting, focusing, collimating, beamforming, scatter-
ating principle of metamaterials remains the same. As ing.
depicted in Fig. 5, an impinging wave of any physical
nature (e.g., EM, mechanical, acoustic, thermal) excites ˆ Bandwidth: Filtering.
the surface elements of a metamaterial, initiating a spa-
tial distribution of energy over and within it. We will ˆ Modulation: Requires embedded actuators that can
call this distribution “exciting-source”. On the other switch states fast enough to yield the targeted mod-
hand, well-known and cross-domain principles state that ulation type [30].
any energy wavefront, which we demand to be emitted
by the metamaterial as a response to the excitation, can ˆ Frequency: Filtering, channel conversion.
be traced back to a corresponding surface energy distri-
bution denoted as “producing-source” [3, 6]. Therefore, ˆ Doppler effect mitigation and non-linear effects [8].
a metamaterial configures its tunable elements to cre-
ate a circuit that morphs the exciting-source into the
Additionally, sensing impinging waves may be consid-
producing-source. In this way, a metamaterial with high
ered one of the above functionalities and, as an outcome,
meta-atom density can perform any kind of energy wave
the embedded sensors can extract information of any of
manipulation that respects the energy preservation prin-
the above parameters related to the incident wave.
ciple. Arguably, the electromagnetism constitutes a very
In this aspect, the role of the contributed metamate-
complex energy type to describe and, as a consequence,
rial API is to model these manipulation types into a
manipulate in this manner, as it is described by two de-
library of software callbacks with appropriate parame-
pendent vectors (electric and magnetic field) as well as
ters. Then, for each callback and assorted parameters,
their relative orientation in space, i.e. polarization (me-
the Metamaterial Middleware produces the correspond-
chanical, acoustic and thermal waves can be described
ing states of the embedded tunable elements that in-
by a single scalar field in space). As such, incoming EM
deed yield the required energy manipulation type. In
waves can be treated in more ways than other energy
other words, a metamaterial coupled with an API and
types. The common types of EM wave manipulation via
a Metamaterial Middleware can be viewed as a hypervi-
metamaterials, reported in the literature [3], can desig-
sor that can host metamaterial functionalities upon user
nate a set of high-level functionality types as follows:
request [8].
ˆ Amplitude: Filtering (band-stop, -pass), absorp- In the following, we focus on EM metamaterials which,
tion. as described, yield the richest API and most complex
Metamaterial Middleware. The expansion to other en-
ˆ Polarization: Waveplates (polarization conversion, ergy domains is discussed via derivation in Section 7.
modulation).

© International Telecommunication Union, 2020 59

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

API callbacks into metamaterial hardware directives.

SDN app ecosystem Envisioned IoMMT apps
A notable trait of the Metamaterial Middleware is that
Device discovery
Device access control Wireless Power Energy isolation and
and registration
app
& permissions app Transfer security it is divided into two parts, in terms of system deploy-
Device location discovery app Propagation optimization ment [32]:

Northbound Inteface IoMMT API 1. The metamaterial manufacturing stage component,

a complex, offline process requiring special meta-
SDN Controller IoMMT Middleware
DB material measurement and evaluation setups (dis-
Southbound Inteface - Unified IoT and IoMMT interface cussed in Section 5), and

2. The metamaterial operation stage component,

which operates in real time based on a codebook.
SDN
Existing IoT, Networking and
This codebook is a database populated once by
Controller Sensor infrastructure the manufacturing stage component and contains
IoMMT API
callback
Metamaterial unit
a comprehensive set of configurations for all meta-
connected to the
SDN Controller
material API callbacks, supported by a given meta-
Con
material.
tain
ed a
nd o
ptim
ized
IoMMT API
callback pro
pag
atio
IoMMT API
callback The operation stage component simply retrieves con-
Interference-protected n
area Environment figurations from the codebook and optionally combines
OR enabling
Avoided obstacle
OR software- them as needed, using an interleaving process described
Avoided untrusted
device defined in Section 5.
propagation
Transmitter Notably, other studies propose the use of online machine
Receiver
learning as a one-shot process, which can be more prac-
tical when response time is not a major concern [33].
However, in this work we propose the aforementioned
Fig. 6 – SDN schematic display of the system model and the separation in deployment, to ensure the fastest opera-
entire workflow abstraction.
tion possible overall, thus covering even the most de-
3. NETWORKED METAMATERIALS manding cases.
AND SDN WORKFLOWS It is noted that SDN is not a choice due to restric-
tions, but rather a choice due to compatibility. In the
Many metamaterials deployed within an environment software-defined metasurfaces presented in this paper, a
can be networked through their gateways. This means key point is the abstraction of physics via an API that
that they may become centrally monitored and config- allows networking logic to be reused in PWEs, without
ured via a server/access point in order to serve a partic- requiring a deep understanding of physics. SDN has
ular end objective. (among other things) already introduced this separa-
An example is given in Fig. 6, where a set of meta- tion of control logic from the underlying hardware and
materials is designed with the proper commands for its administrative peculiarities. Therefore, we propose
energy wave steering and focusing, in order to route an integration of PWE within SDN to better convey the
the energy waves exchanged between two wireless users, logical alignment of the two concepts.
thus avoiding obstacles or eavesdroppers. Other ap-
plications include wireless power transfer and wireless 4. APPLICATION PROGRAMMING
channel customization for an advanced quality of service INTERFACE FOR METAMATERI-
(QoS) [7, 13]. Such a space, where energy propagation
becomes software defined via metamaterials is called a
ALS
programmable wireless environment (PWE) [8]. In the following, we consider a metamaterial in the form
As shown in [7], the PWE architecture is based of a rectangular tile. The term tile is used to refer to a
on the software-defined networking (SDN) principles. practical metamaterial product unit, which can be used
The PWE server is implemented within an SDN con- to cover large objects such as walls and ceilings in a
troller [31]; the southbound interface abstracts the meta- floorplan.
material hardware, treating metamaterial devices as A software process can be initiated for any metamaterial
networking equipment that can route energy waves (e.g. tile supporting a unique, one-to-one correspondence be-
similar to a router, albeit with a more extended and tween its available switch element configurations and a
unique parameterization). Thus, the metamaterial API large number of metamaterial functionalities. The meta-
constitutes a part of the northbound SDN interface, material tiles in this work incorporate tunable switch
atop of which the security, QoS and power transfer con- elements, which dictate the response of each individual
cepts can be implemented as SDN controller applica- cell, locally. In this way, providing an arrangement of
tions. On the other hand, the Metamaterial Middleware all the tile cells allows the tuning of the “concerted”
is part of the SDN middleware, translating metamaterial metamaterial response of the entire tile.

60 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Associate Tile Enumerate Database

Identifier Supported Tiles Interrupt
user
Initialize Configuration
Environment Set Tile Handle
State Database Interrupt
Enumerate Sup-
ported EM fucntions Get Tile Gateway
State Interrupt
(a) Tile Health Hypersurface Hardware error
Check Gateway Communication error

HyperSurface Functionality

<Variables> <Functionality> <Fun_Type>

ID ID ID
TID TypeID Type “Steering”
Type “R,X,D” FreqID
Ni DoAinID <DoA>
Mj DoAoutID null
ID
Wavefront PolID
theta 0
<Tile> SplitID phi 0
HEADERS ID
N <Beam_Splitting>
POLARIZATION M ID
CS_REFERENCE Size Nb 3
class variables </> Dir “30 0;40 30;0 0”
SOURCE_LOCATION
int ID, N, M;
SOURCE_ATTRIBUTES double physical_size; </> class variables
Variables[Vn][N][M] TileSwitchSet; int ID; String Type;
DIRECTION_OF_ARRIVAL ArrayList<Configurations> Confs; Frequency f;
DIRECTION_OF_DEPARTURE DirectionOfArricval doa_in;
DirectionOfArricval doa_out;
...
DATA Configuration DoAin

DIRECTION_RANGE_1 <Configuration> <States>

EM_FIELD ID ID
DIRECTION_RANGE_2 TID R 0
FID X 65
EM_FIELD VID
D null
SID
(b) (c)

Fig. 7 – (a) Case diagram of the main functions supported by the three basic entities. Tasks highlighted with the database icon indicate
that a set of data is to be retrieved from the Configuration Database. (b) Wavefront description in data object format. (c) A simplified
overview of the structure of the Configuration Database. The ¡Tile¿ table hosts all information regarding a tile’s physical implementation.
The ¡Functionality¿ table combines a set of metamaterial parameters to define new functionalities. Both tables are combined in the
¡Configuration¿ table with a set of entries from the ¡States¿ table to compose a new configuration that supports the functionality FID on
tile TID (VID refers to an entry in the table ¡Variables¿).

© International Telecommunication Union, 2020 61

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

In this section, we present the API that grants access to is parameterized by a discrete variable corresponding to
the tile’s metamaterial applications by defining an ab- the number of outgoing beams, their directivity ampli-
stract representation of the metamaterial, its switch ele- tudes, and an appropriate number of (θ, ϕ) pairs, in-
ment configurations and their respective functionalities. dicating the steering angles. The most complex func-
Specifically, the API resides between a user, operating a tionality can be generally described by a custom scat-
common PC (desktop, smartphone, etc.) and a tile gate- tering pattern and represented herein by a collection of
way, linked to the network of switch element controllers. variables that indicate the reflected power towards all
The case diagram of the proposed concept, presented in directions within the tile’s viewing area.
Fig. 7(a), involves the following main entities: It is, also, worth noting that the data objects being
ˆ A Configuration Database which stores all informa- passed as arguments in the callbacks are primarily de-
tion regarding the tiles, the switch element config- scriptions of wavefronts. A simplified data structure is
urations, and their corresponding functionalities. illustrated in Fig. 7(b). Hence, a wavefront is described
by a type (string identifier), such as “Planar”, “Elliptic”,
ˆ A User which initiates all API callbacks though ei- “Gaussian”, “Custom”, etc. For each type, a series of
ther the source code or button click events in the headers defines the location and attributes of the cre-
Graphical User Interface (GUI). ating source (for impinging wavefronts only) as well as
the coordinate system origin with respect to which all
ˆ The HyperSurface Gateway which represents the
distances are measured. Moreover, the direction of ar-
electronic controller of the hardware.
rival and departure are arrays that can be used to define
ˆ An Interrupt handling service which acts as a per- multiple impinging or departing wavefronts at the same
sistent daemon, receiving and dispatching com- time. Notably, the information within the headers may
mands to the Hypersurface Gateway. be sufficient to produce any value of the wavefront via
simply analytical means. In such cases, the data part
4.1 Data Structures of the Metamaterial API can be left empty. In custom wavefronts, the data is
populated accordingly. A mechanism for defining peri-
In the configuration Database (DB), each tile is associ- odicity is supplied via the notion of ranges (i.e. coor-
ated with an element array S that represents all possible dinate ranges where the energy field is approximately
arrangements of switch element states on the metama- equal), to potentially limit the size of the overall data
terial under study. Each switch element is represented object.
by either a discrete or a continuous variable, creating a The parameters that represent the functionalities and
mathematical space of V1 × N × M + · · · + Vn × N × M configurations of a tile constitute the set of variables
dimensions, where Vi is the number of elements of the that are exposed to the programmer through the meta-
same type (e.g. capacitors) and N, M the number of material API. They are organized in a unified manner
unit cells towards the two perpendicular directions. Fur- within the Database, as shown in Fig. 7(c) which pro-
thermore, every object in this space corresponds to a dif- vides an illustration of the unique association between
ferent state of S and therefore a different configuration. all primary tables. Particularly, the ¡Tile¿ table stores
As an example, a tile with two controllable resistive and all information of a tile’s hardware implementation, such
one diode elements per unit cell is parameterized by a as the number of variables per unit cell and the type of
2 × N × M + 1 × N × M array, where the first and second switch elements. The ¡Functionality¿ table stores the
sets span a continuous [Rmin , . . . , Rmax ] and a discrete representation scheme described in subsection 2 for all
[0, 1] range, respectively. In this case, 0 and 1 corre- available metamaterial functionalities. Each parameter
spond to the OFF and ON states of the diode. This associated with a functionality is organized in a separate
representation will then acquire the following form table, including a table that stores an identification vari-
F ←− [(d1 , d2 , i1 )n=1 , . . . , (d1 , d2 , i1 )n=N ×M ] (1) able representing the type of functionality. This table
ID enumerates all possible operations supported by the
where d1 , d2 are double-type variables, i1 is an inte- tile, including full power absorption, wavefront manipu-
ger variable, and F is the appointed functionality. The lation (steering, splitting, etc.), and wavefront sensing.
primitive data types of all variables should be selected Finally, the ¡Configuration¿ table combines, in an exclu-
so as to minimize the total parameter space of com- sive manner, both primary tables (¡Tile¿ and ¡Function-
bined states without any loss of relevant information. ality¿) to link each stored functionality with a specific
This lays a better optimized communication and com- set of switch element states, acquired from the pool of
putational burden to both the API and the Compiler, available entries in the secondary table ¡States¿.
especially during the compilation process where a siz-
able amount of mathematical computations is required. 4.2 API Callbacks and Event Handling
Accordingly, all functionalities are, also, associated with
their own representation and classified pertinent to their Using the Database as a reference point, the API is re-
own type and defining parameters. For instance, a com- sponsible for interpreting a configuration array to the
plex beam-splitting and polarization control operation proper set of hardware commands, when a suitable call-

62 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

back is executed. In general, an API callback can refer corresponding protocol. This notifies the intra-tile con-
to a number of common requests such as: trol network to assign the switch element states to their
suitable values. Finally, feedback from a successful or
ˆ Detect the number and type of accessible tiles in
failed configuration setup is received from the tile, noti-
the environment.
fying the user. The state of the newly set configuration
ˆ Get the current state of all switch elements or set is evaluated through either the identification of failed or
them to a specific configuration. unresponsive switches, or by activating the sensing app,
in controlled conditions, as a self-diagnosing tool for the
ˆ Check the health status and handle interrupts from tile.
the tiles or the Database. A more advanced API callback may involve the assign-
Prior to any other callback, the API follows an initi- ment of a secondary or supplementary functionality, on
ation process, while the software detects all presently top of an already existing operation. For instance, the
active and connected (discoverable) tiles by broadcast- Metamaterial Middleware may receive independent re-
ing a corresponding network message. The tiles report quests from different users to steer the wavefront of sev-
their location and a unique identifier, e.g. a fixed value, eral point power sources (i.e. the users’ cellphones) to-
that associates all tiles with the same hardware speci- wards the direction of a single nearby network hotspot.
fications. The API validates the support of the active This can be handled by the API in many ways. In the
tiles by checking if the identifier exists in the tile list case where several tiles are present in the environment,
present in the database. It, then, retrieves the switch each tile can be repurposed to host a separate func-
element arrays that correspond to these tiles and re- tionality, distributing all users to their own active tiles.
mains idle until a new “get” or “set” request is received When this is not feasible, the API can divide a single
for a currently active or new configuration, respectively. tile into separate areas and associate the respective el-
This means that the API is now open to receive new ement switches to different configurations (the division
functionality requests from a user, physically operating is usually performed in equal-sized rectangular patterns,
the software, or generate its own requests by reacting to but interlacing can, also, be used). Lastly, two function-
unexpected changes in the environment of devices linked alities can be combined into a single one, when a cor-
to the MS network. When a new functionality request is responding physical interpretation exists. For example,
received, the API retrieves one of the available configu- two separate steering operations, from the same source,
rations from the Database and translates it to a proper may be combined into a single dual-splitting operation,
set of element states on an active metamaterial tile. The expressed by a single unified pattern on the metamate-
corresponding API callback process is illustrated in Fig. rial.
8. In particular, the Caller (user) executes a metama- In other cases, the configuration resolver may need to
terial Function Deployment request, which, in turn, in- combine several functionalities to produce a special new
vokes the Configuration Resolver, identifying a tile that application. This occurs when a functionality is pa-
supports the requested functionality. Next, the resolver rameterized by a continuous variable (e.g. a steering
queries the Database and returns a configuration that angle), while the Metamaterial Middleware can eval-
matches the intended metamaterial application, looking uate and store only a finite number of entries in the
for a proper entry in the ¡Configuration¿ table. The API database. In such a scenario, the resolver seeks the two
creates a string command, using the tile identifier and a closest matching entries through an appropriate mini-
hardware representation of the element state variables, mization function (e.g. for a beam-steering operation,
which is, then, conveyed to the tile Gateway using the the minimum distance between the requested and cur-
rently stored steering angles), whereas performing an
user
interpolation of the switch state values.
Configuration
Resolver
To ensure a seamless operation, the API is reinforced
with a set of specialised algorithms to handle unex-
pected failures in the hardware or communication net-
(?) TID supporting FID NxM work. So, a tile must include all necessary identification
matrix
capabilities (e.g. the ability to identify power loss in its
Hypersurface switch elements), and be able to notify the Metamaterial
Gateway NxM Middleware in case of failure. The API is, then, respon-
states
sible for handling these errors by ensuring no loss of the
Callback
consumes current functionality [32]. For illustration, if a group of
switch elements is stuck to an unresponsive state, the
Fig. 8 – A metamaterial Function Deployment request initiates an Metamaterial Middleware may instantly seek the clos-
API callback on the tile (TID). The Configuration Resolver seeks
an appropriate configuration that supports the selected function-
est matching functionality with a fixed configuration for
ality (FID). The matrix of states is conveyed to the HyperSurface the faulty elements. In more severe cases of demanding
Gateway, where it is translated to a set of corresponding hardware human intervention, like an unresponsive network (af-
states. ter a certain timeout), the API is also responsible for

© International Telecommunication Union, 2020 63

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

informing an available user. API Callback:

(HSF type, Impinging wave, Required departing wavefront )

5. THE METAMATERIAL MIDDLE- Phased array model Neural network

WARE Workflow model equivalent rays model
Phased array + circuit
equivalent model Raw EM field model
For the metamaterial to be reconfigured between dif-
Analytical model-
ferent functionalities, a physical mechanism for locally based initial
tuning each unit cell response must be infused [8]. In the solution Period bounds
context of the present work, we assume that the response detection

of the unit cells is controlled by variable impedance loads Smart solution- Effective variable
connected to the front side metallization layer of the space reduction discretization
Goodness-of-Fit
metamaterial, where structures such as the resonant metric Physics-derived
Model-based variable restrictions
patch pair resides [3]. The loads are complex valued Regression calibration
variables, comprising resistors and capacitors or induc- Approach (XCSF)

tors. The value of the i-th load, Zi = Ri + jXi , com- Optimization engine
selection & initialization
prises two parameters: its resistance (Ri > 0) and re-
actance (Xi = −(ωCi )−1 or Xi = +ωLi ), for capacitive
Fitness Function
and inductive loads, respectively. The loads are, thus, Simulations-driven

electromagnetically connected to the surface impedance Optimization Loop

Machine learning-
of the “unloaded” unit cell and by tuning their values Measurements-driven
based accelerator
we can regulate the unit cell response, e.g. the ampli-
tude and phase of its reflection coefficient. The latter is Store optimal (Populate training
naturally a function of frequency and incoming ray di- configuration to DB dataset)

rection and polarization. When the metamaterial unit

Fig. 9 – The Metamaterial Middleware functionality optimization
cells are properly “orchestrated” by means of tuning the workflow. The workflow seeks to match an analytical metamate-
attached (Ri , Xi ) loads, the desired functionality (global rial model and its parameters to a specific parameterized API
response) of the metamaterial is attained. callback. A selected analytical model is first calibrated. Then, an
iterative process (simulation or measurement-based) optimizes the
In the most rigorous approach, the metamaterial re- input parameters of the model that best yield the API callback.
sponse can be computed by full-wave simulations,
which implement Maxwell’s laws, given the geome- band simulation for the response of a unit cell of volume
try and metamaterial properties of the structure as ( λ5 )3 on a contemporary desktop computer could take
well as a complex vector excitation, i.e. the imping- several minutes, especially if the cell includes fine sub-
ing wave polarization and wavefront shape (phase and wavelength features. The memory and CPU resources
amplitude profile). The full-wave simulation captures scale-up linearly for metasurfaces comprising hundreds
the entire physical problem and, hence, does not re- of thousands of unit cells. Moreover, full-wave simula-
quire a metamaterial-level abstraction for the structure. tions do not explicitly unveil the underlying principles
Frequency-domain solvers, which assume linear media that govern the metamaterial functionality.
and harmonic excitation (i.e. the same frequency com-
ponent in both the excitation and the response), are the 5.1 Functionality Optimization Workflow:
prime candidates for full-wave simulation. They typ- Metamaterial Modeling and State Cali-
ically discretize the structure’s volumes or surfaces at
bration
a minimum of λ/10 resolution, formulate the problem
with an appropriate method (e.g. the finite-element or In this section we establish the optimization workflow
the boundary-element method) and, then, numerically of Fig. 9 that drives the calibration process of the meta-
solve a large sparse- or full-array system to compute the material via an appropriate approximation model. Here
response, in our case, the scattered field. Conversely, calibration denotes the matching of actual active ele-
time-domain solvers assume a pulsed excitation, cov- ment states (e.g., the states of a tunable varactor) to
ering a predefined spectral bandwidth and iteratively the corresponding model parameter values (e.g., phase
propagate it across the structure, solving Maxwell’s difference per cell in the reflectarray model). In this
equations to compute its response; they, typically, re- workflow, the optical scattering response is initially in-
quire a dense discretization of the structure, e.g. a vestigated and, then, the solution is hill-climbed via an
minimum of λ/20 resolution. From this process, it be- optimization loop relying on either field measurements
comes evident that metamaterial with a wide aperture, or full-wave simulations. The approximate models are
i.e. spanning over several wavelengths along the max- as follows.
imum dimension, require high computational resources The simplest model is the phased antenna-array anal-
in the full-wave regime, scaling linearly, when paramet- ysis, where each single unit cell is treated as an inde-
ric simulations need to be performed to optimize the pendent antenna, excited by a single impinging ray and
structure and/or the response. For instance, broad- emitting a single ray in response, with a local phase and

64 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

amplitude alteration. Assuming a metamaterial consist- phase, which is identical as before. The key difference
ing of M × N unit cells, the scattered E-field complex and merit is that the calibration is, now, extremely pre-
amplitude pattern at a given frequency can be calcu- cise with regard to the full-wave simulations, while it
lated by the envelope (coherent superposition) of all rays takes much less time to complete, as detailed in the cor-
scattered from the metamaterial [2] responding study of [2].
An intermediate solution, combining the precision of
M X
N
X the circuit model and the automation of the antenna-
E(θ, ϕ) = Amn ejαmn fmn (θmn , ϕmn )
array model, is an equivalent propagation model, men-
m=1 n=1
tioned here for the sake of completion. The main idea
· Γmn ejγmn fmn (θ, ϕ)ejΦmn (θ,ϕ) . (2) is to introduce a generic mechanism to capture the
In (2), ϕ and θ are the azimuth and elevation angles in cross-interactions among meta-atoms (as opposed to the
the scattering direction, (θmn , ϕmn ) denotes the direc- strict, physics-derived nature of the circuit model) and
tion of the wavefront ‘ray’ incident on the mn-th cell, then proceed with automatic model calibration, avoid-
Amn and αmn are the amplitude and phase of the in- ing the need for expert input. The equivalent ray model
cident wavefront on the mn-th cell, Γmn and γmn form uses a neural network approach as the generic cross-talk
the reflection coefficient (amplitude and phase) of the descriptor [35]. A short summary is as follows. Each
mn-th cell, while fmn defines the scattering pattern of meta-atom is mapped to a neural network node, and
the mn-th cell, which, according to reciprocity, is iden- the locally impinging wave amplitude and phase are its
tical for the incident and scattered direction, and, in inputs. Then, we clone this layer (omitting the inputs)
this work, is assumed that fmn (θ, ϕ) = cos(θ). Finally, and form a number of intermediate, fully connected lay-
Φmn (θ, ϕ) is the phase shift in the mn-th cell stemming ers (usually 3-5), thereby emulating a recurrent network
from its geometrical placement, as with a finite number of steps. We define links per node
(shared among all node clones), which define an alter-
Φmn (θ, ϕ) = k sin θ [dx m cos ϕ + dy n sin ϕ] + φ0 (θ, ϕ), ation of the local phase and amplitude, and its distribu-
(3) tion to other neighboring meta-atoms/nodes. Next, we
where dx,y are the rectangular unit-cell lateral dimen- proceed to calibrate the model via feed-forward/back-
sions, k = 2π/λ is the wavenumber in the medium en- propagation, thereby obtaining a match between R, X,
closing the metamaterial, and φ0 is the reference phase Γmn , and γmn values. Nonetheless, despite its auto-
denoting the spherical coordinate system center, typi- mated nature, a major drawback of this model is the
cally in the middle of the metamaterial aperture. Given need for considerable computational resources, without
a uniform, single-frequency impinging wave, any depart- which the model loses its value, since it becomes re-
ing wavefront is essentially a Fourier composition of the stricted only to very simple metamaterial designs.
individual meta-atom responses. Thus, we can, also, cal- Since computational complexity is a concern regard-
culate the meta-atom amplitudes Γmn and phases γmn less of the chosen model, the Metamaterial Middle-
that yield a desired departing wavefront, by applying ware workflow allows the user to define solution reduc-
an inverse Fourier transform, as elaborately discussed tion across three directions. First, meta-atoms may be
in [2]. The calculated Γmn and γmn values must be grouped into periodically repeated super-cells. Thus,
mapped to the Ri and Xi values that generate them, the optimization workflow needs only to optimize the
since the latter are the actual tunable metamaterial pa- configuration parameters of a super-cell, as opposed
rameters. This process requires a set of simulations yet to optimizing the complete metamaterial. Second, the
it can be automated: existing model calibration tech- range of possible R and X values per meta-atom can be
niques, such as the Regression and Goodness of Fit can discretized into regular or irregular steps, reducing the
be employed [34]. solution space further1 . Finally, some R and X values
The shortcoming of the antenna-array approach is that or ranges can be discarded due to the physical nature
the coupling between adjacent unit cells (e.g., compare of the optimization request. For instance, if we seek
against Fig. 5) is not properly accounted for, which can to optimize a wave steering approach with an empha-
result to model imprecision [2]. To this aim, the Meta- sis on minimal losses over the metamaterial (maximum
material Middleware user is presented with an alterna- reflection amplitude), the Ohmic resistance R needs to
tive model. It utilizes the phased array and equivalent receive its boundary value. On a related track, ma-
circuit model, which assumes not only the transmitting- chine learning-based approaches can quickly estimate
responding antenna per meta-atom, but, also, circuit the performance deriving from one set of R and X val-
elements that interconnect them and account for the ues, thereby discarding non-promising ones and acceler-
cross-meta-atom metamaterial interactions. The disad- ating convergence [36].
vantage of this approach is that an expert needs to define Subsequently, the Metamaterial Middleware workflow
this circuit model, that is generally unique per metama- moves to the optimization stage, where it attempts to
terial design [3]. Once this model has been selected and 1 Notably,contemporary optimization engines already incorporate
provided in the proper format, the optimization work- equivalents to this direction, as they are able to detect strongly
flow of Fig. 9 continues, once again, with the calibration and loosely connected inputs-outputs [34]).

© International Telecommunication Union, 2020 65

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

hill-climb the initial solution detected via any of the API Callback:
(HSF type, Impinging wave, HSF Configuration , Resulting departing wavefront)
described approximate models. At this point, the work-
flow is compatible to any modern optimization engine, Machine-learning based
which receives an input solution and outputs one or estimation
more proposed improvements upon it at each iteration.
Herein, we stress the existence of engines that, also, in- Estimation Y Store estimated
departing wavefront
corporate machine learning mechanisms, to accelerate confidence
and calculated
sufficient
the optimization cycle [34]. The optimization can be metrics to DB

based either on full-wave simulations or a real measure- foreach metric

ment test bed, described in Section 6. The optimiza- Analysis-driven N

tion metric can be any reduction of the produced de- Wavefront metric
Simulations-driven Evaluation type calculation
parting wavefront. Various metrics relevant to antenna (reduction)
and propagation theory may be extracted, namely: the
Measurements-driven
number of main lobes (beam directions), the directiv-
ity of main lobes, the side (parasitic) lobes and their Store departing
wavefront and (Populate training
levels, the beam widths, etc. Such metrics can be used calculated metrics data set)
to quantify the metamaterial performance for the re- to DB
quested functionality, e.g. the main lobe directivity and
beam width measures how “well” a metamaterial steers Fig. 10 – Workflow for profiling a metamaterial functionality.
The workflow seeks to produce a data set that describes the
an incoming wavefront to a desired outgoing direction. metamaterial behavior for any impinging wave type that does not
Lastly, the hill-climbed metamaterial configuration per- match the one specified in the current metamaterial configuration.
taining to the metamaterial API callback is stored into An exhaustive evaluation takes place first for a wide set of possi-
a database for any future use by metamaterial users. ble impinging waves. For intermediate impinging wave cases, the
workflow can rely on estimations produced by machine learning
Finally, we note that multiple simultaneous functionali- algorithms or simple extrapolation means, provided that it yields
ties can be supported by interlacing different scattering an acceptable degree of confidence.
profiles across the metamaterial. In general, this is pre-
formed by spatially mixing the profiles in phasor form always be illuminated by the intended wavefront [24].
For instance, user mobility can alter the impinging wave-
Nc
X front in a manner that has limited relation to the in-
Amn ejαmn = Ac,mn ejαc,mn , (4) tended one and, consequently, to the running metama-
c=1 terial configuration. As such, there is a need for fully
profiling a metamaterial, i.e. calculate and cache its
where c iterates over single, ”low-level” functionalities expected response for each intended metamaterial con-
and n, m are the unit cell indices. Typically, low-level figuration, but, also, for each possible (matching or not)
functionalities correspond to simple beam steering op- impinging wavefront of interest. This profiling process
erations, which are produced exclusively by phase vari- is outlined in Fig. 10.
ations on the metamaterial (Ac,mn = 1). In this case,
The profiling process begins by querying the existing
a ”high-level” functionality will correspond to a multi-
cache (part of the DB) or trained model for the given
splitting operation with variable spatial distribution of
metamaterial and an estimation (or existing calculated
Amn amplitude, raising the hardware requirements for
outcome) of the expected metamaterial response for a
the metamaterial. Therefore, a metamaterial with no
given impinging wave. If it exists, this response is stored
absorption capabilities (and thus no control over Amn )
into a separate profile entry for the metamaterial in the
will have limited access to high-level operations, unless
Metamaterial Middleware DB2 . If the response needs
a mathematical approximation is to be applied, skew-
to be calculated anew, the process proceeds with either
ing the scattering response from its ideal state. As dis-
an analysis-, simulation- or measurement-driven eval-
cussed in Section 6, a method for minimizing amplitude
uation. Therefore, the choice is given as a means to
variations has been successfully investigated by increas-
facilitate the expert into reducing the required compu-
ing the number of secondary parasitic lobes. Such a
tational time, as allowed per case. Then, the profiler
problem can be easily reformulated into an optimization
proceeds to, also, calculate all possible reductions of
task, where an optimal match to the ideal high-level op-
the departing wavefront, e.g. the number of main lobes
eration can be pursued under specific constrains (e.g.
(beam directions), the directivity of the main lobes, the
Amn > const., ∀m, n).
side (parasitic) lobes and their levels, the beam widths,
etc. Finally, once all required impinging wavefronts have
5.2 The Metamaterial Functionality Profiler been successfully processed, the profiling process is con-
The optimization workflow of Fig. 9 opts for the best 2 Incase of an estimated response, the user has control over the
metamaterial configuration for a given, specific pair of process to filter out estimations with low confidence. However,
impinging and departing wavefronts. However, in real the selected estimation engine must be able to provide a confi-
deployments, it is not certain that a metamaterial will dence degree for this automation.

66 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

cluded. set of element states, through the use of numerical

It is clarified, that the middleware operations are one- simulations.
time only, i.e., once the database containing the behav-
ior profile of a metasurface is complete, it can be used ˆ The experimental evaluation of the exported config-
in any application setting in the real world by any tile uration through physical measurement of a meta-
of the same type. material prototype. Notably, the presented soft-
ware has been verified experimentally, and a full
6. SOFTWARE IMPLEMENTATION report can be found online [20].
AND EVALUATION
ˆ The final storing of all configuration parameters
Employing the concepts of the previous sections, we into the configuration DB to complete the function
developed a complete Java implementation of the de- optimization process.
scribed software. The software is subdivided into two
integral modules: i) an implementation of the metama- In this context, Fig. 11(a) depicts all the individual steps
terial API that handles the communication and alloca- in separate panels. If a new configuration is to be de-
tion of existing configurations and ii) the Metamaterial fined for a tile already present in the configuration DB,
Middleware that populates the configuration DB with then, the first step can be skipped. Alternatively, the
new data (new tiles, configurations, and functionalities). user must input all essential parameters of the unit cell
The Metamaterial Middleware incorporates a full GUI structure, i.e. the number and type of all variables that
environment, guiding the user through a step-by-step correspond to the sum of reconfigurable metamaterial
process to produce new configurations. It utilizes all elements. The definition of a new configuration begins
available theoretical and computational tools for the ac- with the parameterization of the desired functionality
curate characterization of a metamaterial tile. Further- (Fig. 11(a.2)). The current implementation supports
more, it offers direct access to the configuration DB, plane wave or point source inputs (for far- and near-field
manually, via a custom-made Structured Query Lan- energy sources) and a set of output options correspond-
guage (SQL) manager or through the automated pro- ing to all basic metamaterial functionalities, discussed
cess following a successful metamaterial characteriza- in Section 2. Here, we select a beam splitting operation
tion. Through this process, all the necessary data re- and proceed to the first main step of the characteriza-
lated to a newly produced configuration become explic- tion process.
itly available to the API. The analytical evaluation of the energy scattering pro-
A microwave metasurface was selected to demonstrate file is performed in the software locally and in real time.
the capabilities of the developed concepts and methods During this step, the Metamaterial Middleware calcu-
for software-tunable metamaterials. We adapted the de- lates the proper scattered field response of the imping-
sign of [37], where a set of RF diodes can be employed to ing wave, for each unit cell at the N × M tile, via either
toggle the reflection-phase of each cell between 16 states. the analytical methods of Section 5 or through an op-
We numerically extracted the response of the metasur- timization process. The scattered fields are evaluated
face (i.e., its scattering pattern), and finally used the for each unit cell as a double complex variable (A1 eiφ1
developed software to demonstrate how the metasurface and A2 eiφ2 , magnitude and phase of the reflected TE
response can be controlled. For the practical demonstra- and TM polarizations, respectively), a process physi-
tion of the developed software in the same measurement cally correct under the condition that the unit cell is
environment (anechoic chamber) with a simpler, 1-bit a subwavelength entity. This simply implies that the
metasurface hardware, we redirect the reader to [20,30], input of the “physical size” field in Fig. 11(a.1) must
since the hardware manufacturing topic is quite exten- comply with this specification or a warning message will
sive and clearly beyond the software aspects that con- appear. In our example, the selected tile consists of bi-
stitute the focus of this paper. nary elements (D stands for diodes), which may only
In the following, we list and comprehensively describe control the phase of the co-polarized scattered field (φ1
the steps undertaken during a optimization process, as term). By clicking the “View suggested” button, we an-
seen through the GUI environment of the Metamaterial alytically calculate and display φ0 − φ1 (φ0 is the phase
Middleware. In summary, this process involves: of the incoming wavefront), which should give an indi-
cation of the diode states at the metamaterial. Alterna-
ˆ The definition of a new unit cell structure and tile tively, a similar result can be extracted by launching the
array (if required). metaheuristic optimizer, either via a blind optimization
process (all-0 initial solution) or an assisted optimiza-
ˆ The parameterization of a new functionality.
tion, using the analytically evaluated profile as an ini-
ˆ The analytical evaluation of the scattering profile tial solution. The latter practice leads to more refined
on the metamaterial for the selected functionality. results, over an analytical evaluation, by considering the
finite size of the tile and the non-infinitesimal size of the
ˆ The association of the metamaterial profile to the unit cell.

© International Telecommunication Union, 2020 67

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

3. Evaluate the scattering profile 4. Perform switch-state calibration

1. Design a virtual tile 40x40 unit cellls
analytical outcome

Phase φ Amplitude A

0 A=1,φ=2π
blind optimization
outcome
- binary space

Phase φ
Amplitude A = 1
2. Define a new functionality
(a)

Save configuration to Database Perform evaluation measurement

(b) (c)

Mast Positioner Controller

angle Rx Antenna
Rx
Antenna

VNA

Tx
Antenna
Turn Table angle Compiler Host

(d)

Fig. 11 – (a) Snapshots of the Metamaterial Middleware GUI during the metamaterial characterization process. The user may select an
existing tile or create a new one by drawing its unit cell topology and variable elements within. A new functionality for this tile is assigned
by selecting the desired parameters for the impinging and outgoing waves. The process begins by evaluating the scattering response profile,
calculated as a phase and amplitude array for the selected metamaterial function. During the final step the Metamaterial Middleware
seeks the set of optimal states that can realize the scattering response into an actual hardware configuration. (b) After a successful
characterization process the configuration is saved to the DB. Herein, as an instance, a snapshot of the DB manager that was created
during development of the software is presented. (c) The Metamaterial Middleware offers extensive experimental capabilities through a
dedicated module. (d) Utilizing the hardware present in an anechoic chamber, the Metamaterial Middleware was able to acquire full
scattering diagrams of a metamaterial tile, mounted on the available positioning equipment. The three-dimensional (3D) schematic, shown
on the left, is updated live as the turntable or the positioner head rotates during a measurement or optimization process. On the right
hand side, the Metamaterial Middleware host (white laptop) connects via Ethernet to the VNA (measuring the Rx/Tx antennas) and to
the positioner controller.

68 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

In the final step, we seek to match the scattering field Algorithm 1 Physical Element Calibration. High-
response (i.e. the (A1 eiφ1 , A2 eiφ2 )i,j pairs calculated lighted commands run in a separate thread.
in the previous step) to the appropriate set of element Set index to i = 0, j = 0
states, such as the resistance R and reactance X (ca- while index does not exceed max do
pacitance or inductance) of the loads, for all unit cells Get Sij ; Get T ol
indexed by i, j. The correspondence between the scat- if !(exists Sdb where |Sdb − Sij | < T ol) then
tering response of a unit cell and its physical structure while searching do
constitutes a highly complex and demanding propaga- Select node where state == f ree
tion problem. Actually, the search for a proper set of Set to busy
states for a fully defined response constitutes an inverse Initiate parallel thread;
problem with closed-form solutions available only for Optimizer: Fixes (R, L, C, D) variables;
very simple unit cell types. As such, it demands the use Optimizer → Requests new simulation;
of highly efficient optimization algorithms and strong node ← Returns Ssim
computation power to perform the necessary numerical Sdb = Ssim ;
simulations. In our implementation, we employ a cluster if |Ssim − Sij | < T ol then
of interconnected computer clients (that serve as simu- searching = f alse
lation nodes) receiving instructions through a TCP/IP Set (R, L, C, D) variables as optimal
network from the host Metamaterial Middleware run- end if
ning on the main machine. In a lightweight scenario, Set node state to f ree
a single-cell simulation assigned to one of the clients end while
can be completed in less than a minute but this may Iterate index
grow substantially when the complexity of the design is end if
increased. An acceptable convergence is expected to be end while
reached within a few thousands simulation trials. In per-
spective, the prototype studied in this work displayed a process and a new series of simulations begins un-
typical evaluation time of 8 to 12 hours for a full char- til a sufficiently close convergence to the targeted
acterization of its configuration space and all possible scattering field response Sij =(A1 eiφ1 , A2 eiφ2 )i,j is
impinging plane waves (θ,φ sets) at the operation fre- achieved. Prior to each simulation, the optimizer
quency of the hardware. The evaluation is performed looks up all entries in the associated table of the
through Algorithm 1 (see next page) whose steps are configuration DB, in case a result (Sdb ) is already
described below. Specifically, the user: stored. If not, the simulation will start and the
result will be stored afterwards. Over time, this
ˆ selects one of the three options: i) a gradient based process can populate the DB with enough results,
optimizer, recommended for a small number of vari- making option (iii) of the initial step a highly effi-
ables (less than four), ii) a metaheuristic optimizer, cient method for evaluating new functionalities for
recommended for a higher count of variables (more this particular tile.
than three), and iii) a database-based optimization
process that configures the proper switch-states, By completing the previous steps, the software has suc-
through simulation results already stored in the cessfully defined a new configuration for the chosen tile,
configuration DB. which, now, remains to be stored in the configuration
DB. Before doing so, we can apply an additional evalu-
ˆ suggests a convergence limit T ol to the optimizer
ation step by conducting an experimental measurement
in the Tolerance text box for all variable targets,
on a physical prototype (if available). Our current im-
ˆ fills the IP list with all available simulation nodes plementation is able to assess multi-splitting or absorp-
and initiates a connection. All nodes will launch the tion functionalities by measuring the full scattering dia-
installed simulation software, open the correspond- gram on the front metamaterial hemisphere, which can
ing geometry model, and remain idle until further then be compared to the scattering profile produced
instructions are received. by the software. As presented in the corresponding
technical report [20], several tests with a metamaterial
ˆ modifies, if necessary, the value range, step and ini- unit have already been successfully conducted in a fully
tial value for each variable in the unit cell. equipped anechoic chamber (Fig. 11). This test bed
incorporates a variety of algorithms, meeting individual
ˆ begins the optimization process by clicking the needs for accuracy and speed for various cases of scat-
“RUN” button. The software will iterate over tering response measurements (e.g. a full 3D-pattern
all unit cells (from top-left (0, 0) to bottom-right versus a 2D-slice might be required for arbitrary lobe
(N, M )), seeking a set of optimal values for the scattering). A simple case is outlined in Algorithm 2,
variables of each cell. During this sequence, each where the user has previously set the appropriate pa-
unit cell (i, j) initiates an independent optimization rameters in the GUI (Fig. 11(c)). The GUI automates

© International Telecommunication Union, 2020 69

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

00 01 10 11 Incident plane wave

Dual splitting Quad splitting
20x20 pattern

E2 linear
normalized
Circular beam Dual-circular beam
1 pattern (40x40) pattern
Dual splitting - shifted Quad splitting - shifted

Ε2

0
(a)
E2 linear
normalized

E2 logarithmic (dB) Phase φ Amplitude A

(b)
0 Ε2 20 0 φ 2π 0 Α 1

(c)

Fig. 12 – (a) Element states and far-field scattering diagram of a 4-bit metamaterial array for 6 different cases. (b) Element states and
scattering diagrams for triple- and quintuple-beam scattering, assuming continuously adjustable states with absorption capabilities. (c)
Comparison between a tile with resistor elements (left) and a tile without (right) for an in-plane triple-beam splitting functionality. A
non-uniform amplitude pattern (Amn ) can eliminate all side-lobes and provide increased security to the signal. The incident field in all cases
is a vertically impinging plane wave, chosen for simplicity; yet any plane wave or point source input can be considered by a corresponding
shift of the phase for each unit cell.

70 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Algorithm 2 Evaluate Scattering Response, 3D, slow

rotation case. Highlighted commands refer to HW in-
structions.
Set N ← number of equidistant points on the hemi-
sphere
Set P [N ] ← Struct of (θ, φ) points
for all elements i in P do
Send rotation command → Positioner
(Mast - P [i].theta, Head - P [i].phi, Speed)
Receive ← Positioner feedback
Refresh 3D Figure
Send measurement command → VNA
E[i] = Receive ← S21 power Fig. 13 – Arbitrary functionality optimization test. A smiley face-
Refresh Plot graph shaped scattering pattern is successfully produced (left). The cor-
responding metamaterial element configuration (different meta-
end for atom states expressed in colormap) is shown to the left.
Save (P [i],E[i]) to DB
unit cell with reflection phases (−90, 0, 90, 180) and full
the process of measuring metamaterial devices in the reflection amplitude (A = 1) [14,38]. The outcome com-
anechoic chamber by supporting several communication plies fully with the results provided by the corresponding
interfaces for the following equipment: authors, indicating that the Metamaterial Middleware
may cooperate with arbitrary hardware configurations
ˆ Vector network analyzer (VNA): produces the en- and thus be compatible with any reasonable design in
ergy signal and receives the response (i.e. S21 - a future diverse metamaterial market. Following these
parameter) from the antenna setup. results, we, also, test four exclusive cases that highlight
the additional capabilities of our software. In particu-
ˆ Positioner: allows the mechanical support of the
lar, Fig. 12(b) displays the results for a triple- and a
metamaterial and antenna devices. Its controller
quintuple-beam splitting case, while Fig. 12(c) shows
can instruct the rotation of both heads (towards
the optimization outcome for an in-plane triple-beam
θ, φ), allowing a complete characterization of the
splitting functionality. For the latter case, the inte-
scattering profile.
grated theoretical algorithms were able to eliminate all
ˆ metamaterial controllers: A metamaterial hosting side lobes by suggesting a non-uniform pattern for the
reconfigurable elements incorporates a communica- amplitude A of the co-polarized scattered field. This
tion network for the explicit control of its element particular case demands a tile with controllable absorp-
states. The Metamaterial Middleware implements tion elements (resistors). Finally, in Fig. 13 we proceed
the proper interface for the evaluation prototypes to showcase the optimization of an arbitrary departing
(serial port connection, WiFi, and Bluetooth have wavefront formation. A smiley-face shaped scattering
been integrated). The same interfaces are, also, pattern is selected as the required energy wave response
used for the metamaterial API developed in Sec- of the metamaterial to a planar impinging wave. The
tion 4. optimization process successfully produces the required
wavefront, and the corresponding metamaterial element
A supplementary note is that the final switch-state con- states are shown to the right of Fig. 13.
figuration can be re-evaluated using the same meta-
heuristic optimizer utilized in step 3 of Fig. 12(a) 7. DISCUSSION: THE TRANSFOR-
via actual experimental results. The optimizer starts MATIONAL POTENTIAL OF THE
with the software-defined configuration as an initial so-
IOMMT AND FUTURE DIREC-
lution and gradually adjusts the switch-state matrices
to more optimally converge to the pursued functional- TIONS
ity under true operational conditions. The implemented While the potential of the IoMMT paradigm alone may
algorithm follows the template of Algorithm 2, where be worth the investigation, here we evaluate its practical
N correspond to the number or optimization variables opportunities affecting the industry, the end users and
(e.g. the number of scattering lobes) and a second f or the environment, namely:
loop nests the existing loop, seeking to maximize the
Sumi (E[i]) metric. ˆ How can the IoMMT prolong the life cycle of prod-
For further evaluation purposes, we validate a number of ucts across deployment scales?
indicative examples from the literature, based on previ-
ously measured and simulated results. Hence, Fig. 12(a) ˆ How can the IoMMT help maintain a high-speed
presents the optimization outcome for a 4-bit metamate- product development pace, without sacrificing eco-
rial array, which can switch over four available states per logical concerns during the product design phase?

© International Telecommunication Union, 2020 71

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

To these ends, we believe that the concept of Circular Micro-manage energy losses Energy manipula on as
and parasi c dissipa ons via an app framework
Economy (CE) and its associated performance indices so ware
is a fitting framework for the initial exploration and
evaluation of the IoMMT paradigm [39, 40]. CE seeks SONIC Electromagne c, Mechanical,
Acous c and Thermal proper es API
to make technological products reusable, repairable and
Northbound Interface
recyclable across their lifetime (i.e., development, pur- EM
chase, usage and disposal) by introducing cross-product Smart Environment Configura on

and cross-manufacturer interactions. Instead of the tra- Consistency and Orchestra on

ditional, linear order of life cycles phases, i.e., i) raw re- MECH Material Mul -physics M-ware
source acquirement, ii) processing, iii) distribution, iv) Monitoring SONIC HEAT
Modules EM MECH
its use and v) disposal, the CE advocates to create links
THERMAL Southbound Interface
from disposal to all preceding phases, promoting i) re-
Inteconnectivity, deployment and configuration
processing or refurbishment, ii) redeployment and redis-
tribution, and iii) multiple uses. IoMMT varia ons &
AI-driven composi on
However, according to the literature [41], the CE intro-
profiling and rec-
duces a paradoxical tension in the industry: While the Equipment-level ommenda on systems
Outdoors,
industry is pressed for faster growth and, hence, a faster (Motors, Cooling/ Indoors- vehicular, IoMMT data and energy
Hea ng/Insula ng level smart-city models
product development rate, CE can introduce a series of surfaces)
design considerations that make for a slower product de-
velopment rate. In this view, the paradigm of IoMMT Fig. 14 – Envisioned future research directions for the Internet
of MetaMaterials.
is by its nature impactful for the energy and ecologi-
cal footprint of multiple products, across disciplines and als, to match the requirements and specifications of any
scales. envisioned application.
The fact that it enables the tuning and optimization of Moving to the networking layer, research can follow the
the physical properties of matter allows for a tremen- proposed northbound/southbound abstraction model
dous impact both in terms of quality of service per inspired from the SDN paradigm. This will provide the
product and scale, but also for energy savings in a hor- necessary platform for: i) Interconnecting the IoMMT
izontal manner. Moreover, the IoMMT can contribute to the vast array of existing networked devices and
a software-driven way for optimizing material proper- assorted standards, models and protocols. ii) Pro-
ties. Using this new technology, the industrial players vide the necessary software abstractions, to open the
can maintain a fast-paced product design, where energy field of energy micromanagement to software develop-
efficiency and sustainability can be upgraded program- ers (i.e., without specialty in Physics), enabling the
matically via “eco-firmware” during its use, thereby of- energy-propagation-as-an-app paradigm. In this aspect,
floading the product design phase of such concerns. we also envision the need for algorithms optimizing
An important future research direction of the IoMMT is IoMMT deployments for minimal-investment-maximal-
to quantify the financial savings stemming from adopt- control, and orchestrating IoMMT deployments for any
ing this technique. While the Circular Economy-derived set of generic energy micro-management objective.
paradoxical tensions have been around for long and are
hard to eradicate, IoMMT can facilitate their resolu- The control time granularity depends on the application
tion by quantifying them, potentially aligning fast-paced scenario and the volatility of the factors affecting the en-
marketing with environmental sustainability. ergy propagation within an environment. In an indoors
Apart from the CE line of work, future work will seek wireless communications setting, such as the one stud-
to provide the theoretical and modeling foundations of ied in [7], where the Intelligent Wireless Environment
the IoMMT. Our vision, overviewed in Fig. 14, is for needs to continuously adapt to the position of user de-
a full-stack study of this new concept, covering: the vices, the control granularity can be considered to be
physical layer modeling, the internetworking and com- 10-25 msec (i.e., randomly walking users running a mo-
munications layer, and the application layer. bile application).
At the physical layer, future research needs to classify Finally, at the application layer research can define key-
and model metamaterials in a functionality-centric way, applications of the IoMMT at multiple scales, start-
and introduce fitting Key-Performance Indicators for ing from the internals of equipment, such as motors,
each offered energy manipulation type. Metasurface- heating, cooling and insulating surfaces. This will pro-
internal control variations (technologies and ways of vide the basic units for IoMMT incorporation to de-
monitoring embedded active elements) can be taken into vices spanning home appliances (ovens, refrigerators,
account, and aspire to deduce models covering the as- washers, heating and cooling units), electronics (from
pects of control data traffic, energy expenditure and interference cancellation, to smart cooling) and build-
feature-based manufacturing cost estimation. This will ing materials (acoustic, thermal and mechanical insula-
enable the creation of the first, cross-physical domain tors). Various scales can be taken into consideration,
profiling and recommendation system for metamateri- from indoor (smart-house) and outdoor (city-level, ve-

72 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

hicular networks) IoMMT deployments. no. 6, pp. 527–537, Dec. 2015. [Online]. Available:
https://doi.org/10.1109/jiot.2015.2413397
8. CONCLUSION [2] F. Capolino, Theory and phenomena of metamate-
In this paper, we pursue the systematic expansion of the rials. CRC press, 2017.
concept of software-defined metamaterials by establish-
[3] A. Li, S. Singh, and D. Sievenpiper, “Metasurfaces
ing the key software elements of a metamaterial network.
and their applications,” Nanophotonics, vol. 7,
With this goal in mind, we introduce two special cate-
no. 6, pp. 989–1011, 2018.
gories of software, the Metamaterial Middleware, which
can methodically produce novel configurations for a sin- [4] M. Kadic, T. Bückmann, R. Schittny, and M. We-
gle metamaterial tile and the metamaterial API, which gener, “Metamaterials beyond electromagnetism,”
is in charge of supervising the energy propagation within Reports on Progress in physics, vol. 76, no. 12, p.
a metamaterial-coated space. The two components were 126501, 2013.
studied and developed autonomously, and then merged
into a unified application via a common layer of abstrac- [5] J. U. Surjadi, L. Gao, H. Du, X. Li, X. Xiong,
tion. N. X. Fang, and Y. Lu, “Mechanical metamate-
For the API, we explored the means to interpret a meta- rials and their engineering applications,” Advanced
material, its configuration space, and the supported en- Engineering Materials, vol. 21, no. 3, p. 1800864,
ergy manipulation functionalities via a group of well- 2019.
defined software objects. The key objective was to suc- [6] M. Pishvar and R. L. Harne, “Foundations for soft,
cessfully conceal the physical layer of the metamaterial, smart matter by active mechanical metamaterials,”
and only expose the essential parameters for configuring Advanced Science, p. 2001384, 2020.
an operation. In light of this definition, the API is capa-
ble of instructing its environment to steer, focus, absorb [7] C. Liaskos, A. Tsioliaridou, A. Pitsillides, S. Ioan-
or split the incoming signals through a simple interface, nidis, and I. Akyildiz, “Using any surface to real-
without any reference to the underlying physics. ize a new paradigm for wireless communications,”
The Metamaterial Middleware was developed by means Commun. ACM, vol. 61, pp. 30–33, 2018.
of various theoretical and computational tools, includ-
ing analytical algorithms that assess the scattering re- [8] C. Liaskos, A. Tsioliaridou, S. Nie, A. Pitsillides,
sponse, full-wave simulations that accurately evaluate S. Ioannidis, and I. Akyildiz, “On the network-
a unit cell response, and experimental modules that layer modeling and configuration of programmable
can perform physical measurements of metamaterial de- wireless environments,” IEEE/ACM Trans. Netw.,
vices. Through a step-by-step process we defined a ro- vol. 27, no. 4, pp. 1696–1713, 2019.
bust characterization methodology, supporting a large [9] I. F. Akyildiz and J. M. Jornet, “The internet
set of metamaterial operations with no restriction on the of nano-things,” IEEE Wireless Communications,
specifications for the metamaterial. We also discuss the vol. 17, no. 6, pp. 58–63, 2010.
prospect of applying different optimization algorithms
for each stage of the metamaterial characterization pro- [10] S. B. Glybovski, S. A. Tretyakov, P. A. Belov, Y. S.
cess. Finally, the evaluation of the middleware outcomes Kivshar, and C. R. Simovski, “Metasurfaces: From
demonstrated that the IoMMT paradigm can be em- microwaves to visible,” Phys. Rep., vol. 634, pp.
ployed successfully within a unified framework. 1–72, 2016.
[11] H.-T. Chen, A. J. Taylor, and N. Yu, “A review
ACKNOWLEDGMENT of metasurfaces: Physics and applications,” Rep.
This work was supported by the European Union’s Hori- Prog. Phys., vol. 79, no. 7, pp. 076 401(1–40), 2016.
zon 2020 research and innovation programme-project
[12] C. Huang, C. Zhang, J. Yang, B. Sun, B. Zhao, and
C4IIoT, GA EU833828. The authors also acknowl-
X. Luo, “Reconfigurable metasurface for multifunc-
edge FETOPEN-RIA project VISORSURF. All pre-
tional control of electromagnetic waves,” Adv. Opt.
sented software modules were conceived, designed and
Mater., vol. 5, no. 22, pp. 1 700 485(1–6), 2017.
implemented in full by the Foundation for Research and
Technology-Hellas (FORTH) and G. Pyrialakos served [13] Ö. Özdogan, E. Björnson, and E. Larsson, “In-
as the lead developer. telligent reflecting surfaces: Physics, propagation,
and pathloss modeling,” IEEE Wireless Commun.
REFERENCES Lett., vol. 9, no. 5, pp. 581–585, 2019.
[1] J. Pan, R. Jain, S. Paul, T. Vu, A. Saifullah, [14] Q. Zhang, X. Wan, S. Liu, J. Yin, L. Zhang,
and M. Sha, “An internet of things framework and T. Cui, “Shaping electromagnetic waves us-
for smart energy in buildings: Designs, prototype, ing software-automatically-designed metasurfaces,”
and experiments,” IEEE Internet Things J., vol. 2, Sci. Rep., vol. 7, no. 1, pp. 3588(1–11), 2017.

© International Telecommunication Union, 2020 73

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[15] A. P. Mosk, A. Lagendijk, G. Lerosey, and M. Fink, [25] C. Liaskos, S. Nie, A. Tsioliaridou, A. Pitsillides,
“Controlling waves in space and time for imaging S. Ioannidis, and I. Akyildiz, “End-to-end wireless
and focusing in complex media,” Nature photonics, path deployment with intelligent surfaces using in-
vol. 6, no. 5, pp. 283–292, 2012. terpretable neural networks,” IEEE Transactions
on Communications, 2020.
[16] X. Tan, Z. Sun, D. Koutsonikolas, and J. M. Jornet,
“Enabling indoor mobile millimeter-wave networks [26] F. Mathioudakis, C. Liaskos, A. Tsioliaridou,
based on smart reflect-arrays,” in IEEE INFOCOM S. Nie, A. Pitsillides, S. Ioannidis, and I. Akyildiz,
2018-IEEE Conference on Computer Communica- “Advanced physical-layer security as an app in pro-
tions. IEEE, 2018, pp. 270–278. grammable wireless environments,” in 2020 IEEE
21st International Workshop on Signal Processing
[17] C. Liaskos, A. Tsioliaridou, A. Pitsillides, I. F. Aky-
Advances in Wireless Communications (SPAWC).
ildiz, N. V. Kantartzis, A. X. Lalas, X. Dimitropou-
IEEE, 2020, pp. 1–5.
los, S. Ioannidis, M. Kafesaki, and C. Soukoulis,
“Design and development of software defined meta-
[27] Z. Chen, B. Guo, Y. Yang, and C. Cheng,
materials for nanonetworks,” IEEE Circuits and
“Metamaterials-based enhanced energy harvesting:
Systems Magazine, vol. 15, no. 4, pp. 12–25, 2015.
A review,” Physica B: Condensed Matter, vol. 438,
[18] F. Liu et al., “Programmable metasurfaces: State pp. 1–8, 2014.
of the art and prospects,” in 2018 IEEE In-
ternational Symposium on Circuits and Systems [28] B. Florijn, C. Coulais, and M. van Hecke, “Pro-
(ISCAS), vol. 2018-May. IEEE, May 2018, p. grammable mechanical metamaterials,” Phys. Rev.
8351817. Lett., vol. 113, no. 17, pp. 175 503(1–4), 2014.

[19] T. Saeed, V. Soteriou, C. Liaskos, A. Pitsillides, [29] S. A. Cummer, J. Christensen, and A. Alù, “Con-
and M. Lestas, “Toward fault-tolerant deadlock- trolling sound with acoustic metamaterials,” Nat.
free routing in hypersurface-embedded controller Rev. Mat., vol. 1, no. 3, pp. 16 001(1–13), 2016.
networks,” IEEE Networking Letters, vol. 2, no. 3,
pp. 140–144, 2020. [30] L. Zhang, X. Q. Chen, S. Liu, Q. Zhang, J. Zhao,
J. Y. Dai, G. D. Bai, X. Wan, Q. Cheng,
[20] The VISORSURF project consortium, “A hy- G. Castaldi, V. Galdi, and T. J. Cui, “Space-
pervisor for metasurface functionalities: progress time-coding digital metasurfaces,” Nat. Commun.,
report,” European Commission Project VI- vol. 9, no. 1, pp. 4334(1–11), 2018.
SORSURF: Public Report, Aug-2019, [On-
line:] https://ec.europa.eu/research/participants [31] Y. E. Oktian, S. Lee, H. Lee, and J. Lam, “Dis-
/documents/downloadPublic? documen- tributed SDN controller system: A survey on de-
tIds=080166e5c560b376&appId=PPGMS. sign choice,” Computer Netw., vol. 121, pp. 100–
111, 2017.
[21] T. Frenzel, M. Kadic, and M. Wegener, “Three-
dimensional mechanical metamaterials with a [32] C. Liaskos, A. Pitilakis, A. Tsioliaridou, N. Kan-
twist,” Science, vol. 358, no. 6366, pp. 1072–1074, tartzis, and A. Pitsillides, “Initial UML def-
2017. inition of the hypersurface compiler middle-
[22] M. R. Haberman and M. D. Guild, “Acous- ware,” European Commission Project VISOR-
tic metamaterials,” Physics Today, vol. 69, SURF: Public Deliverable D2.2, 31-Dec-2017, [On-
no. 6, pp. 42–48, Jun. 2016. [Online]. Available: line:] http://www.visorsurf.eu/m/VISORSURF-
https://doi.org/10.1063/pt.3.3198 D2.2.pdf.

[23] M. S. Dresselhaus, G. Chen, M. Y. Tang, R. Yang, [33] N. Ashraf, M. Lestas, T. Saeed, H. Taghvaee,
H. Lee, D. Wang, Z. Ren, J.-P. Fleurial, and S. Abadal, A. Pitsillides, and C. Liaskos, “Ex-
P. Gogna, “New directions for low-dimensional tremum seeking control for beam steering using hy-
thermoelectric materials,” Advanced materials, persurfaces,” in 2020 IEEE International Confer-
vol. 19, no. 8, pp. 1043–1053, 2007. ence on Communications Workshops (ICC Work-
shops). IEEE, 2020, pp. 1–6.
[24] C. Liaskos, S. Nie, A. Tsioliaridou, A. Pitsil-
lides, S. Ioannidis, and I. F. Akyildiz, “Mobility- [34] M. Laguna and R. Marti, “The OptQuest callable
aware beam steering in metasurface-based pro- library,” in Optimization Software Class Libraries.
grammable wireless environments,” in ICASSP Springer, 2003, pp. 193–218.
2020-2020 IEEE International Conference on
Acoustics, Speech and Signal Processing (ICASSP). [35] C. Liaskos, A. Tsioliaridou, S. Nie, A. Pitsillides,
IEEE, 2020, pp. 9150–9154. S. Ioannidis, and I. Akyildiz, “An interpretable

74 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

neural network for configuring programmable wire- AUTHORS

less environments,” in IEEE Int. Workshop Sig-
nal Processing Adv. Wireless Commun. (SPAWC
2019), 2019, pp. 1–5.
[36] H. Taghvaee, A. Jain, X. Timoneda, C. Liaskos,
S. Abadal, E. Alarcón, and A. Cabellos-Aparicio,
“Radiation pattern prediction for metasurfaces: A
neural network based approach,” arXiv preprint
arXiv:2007.08035, 2020.
[37] Y. Saifullah, A. B. Waqas, G.-M. Yang, F. Zhang, Christos Liaskos received a Diploma in Electrical and
and F. Xu, “4-bit optimized coding metasurface for Computer Engineering from the Aristotle University of
wideband rcs reduction,” IEEE Access, vol. 7, pp. Thessaloniki (AUTH), Greece in 2004, an MSc degree
122 378–122 386, 2019. in Medical Informatics in 2008 from the Medical School,
AUTH and PhD degree in Computer Networking from
[38] S. Liu, T. J. Cui, L. Zhang, Q. Xu, W. Qiu, X. Wan,
the Dept. of Informatics, AUTH in 2014. He has pub-
J. Q. Gu, W. X. Tang, M. Q. Qi, J. G. Han, W. L.
lished work in several journals and conferences, such as
Zhang, X. Y. Zhou, and Q. Cheng, “Convolution
IEEE Transactions on: Networking, Computers, Vehic-
operations on coding metasurface to reach flexi-
ular Technology, Broadcasting, Systems Man and Cy-
ble and continuous controls of terahertz beams,”
bernetics, Networks and Service Management, Commu-
Adv. Science, vol. 3, no. 10, pp. 1 600 156(1–12),
nications, INFOCOM. He is currently an assistant pro-
Jul. 2016.
fessor with the University of Ioannina, Greece, and an
[39] W. R. Stahel, “The circular economy,” Nature, vol. affiliated researcher at the Foundation of Research and
531, no. 7595, pp. 435–438, 2016. Technology, Hellas (FORTH). His research interests in-
clude computer networks, security and nanotechnology,
[40] C. Liaskos, A. Tsioliaridou, and S. Ioannidis, “To- with a focus on developing nanonetwork architectures
wards a circular economy via intelligent metamate- and communication protocols for future applications.
rials,” in 2018 IEEE 23rd International Workshop
on Computer Aided Modeling and Design of Com-
munication Links and Networks (CAMAD). IEEE,
2018, pp. 1–6.
[41] T. Daddi, D. Ceglia, G. Bianchi, and M. D. de Bar-
cellos, “Paradoxical tensions and corporate sus-
tainability: A focus on circular economy business
cases,” Corporate Social Responsibility and Envi-
ronmental Management, vol. 26, no. 4, pp. 770–780,
2019. George Pyrialakos received his Diploma and Ph.D.
degree from the Dept. of Electrical and Computer En-
gineering, Aristotle University of Thessaloniki (AUTH),
Greece, in 2013 and 2019, respectively. He is currently a
post-doc researcher in AUTH, affiliated with the Foun-
dation for Research and Technology Hellas (FORTH).
His research interests include computational electromag-
netics, metamaterials, photonics, and condensed matter
physics.

Alexandros Pitilakis received his Diploma and Ph.D.

degree from the Dept. of Electrical and Computer En-

© International Telecommunication Union, 2020 75

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

gineering, Aristotle University of Thessaloniki (AUTH),

Greece, in 2005 and 2013. He, also, holds an MSc. de-
gree from the ENST Paris, 2007. He is a post-doc re-
searcher in AUTH, affiliated with the Foundation for
Research and Technology Hellas (FORTH). His research
interests include computational electromagnetics, meta-
materials, and nonlinear optics.

Nikolaos Kantartzis is a professor at the School of

Electrical and Computer Engineering, Aristotle Univer-
sity of Thessaloniki (AUTH), Greece, from where he
received the Diploma and Ph.D. degrees in 1994 and
1999, respectively. His primary research interests in-
clude computational electromagnetics, metamaterials,
graphene, EMC/EMI problems, microwaves, and nano
devices. Dr. Kantartzis is a Senior Member of IEEE,
an ICS and ACES member.
Ageliki Tsioliaridou received a Diploma and PhD
degrees in Electrical and Computer Engineering from
the Democritus University of Thrace (DUTH), Greece,
in 2004 and 2010, respectively. Her research work is
mainly in the field of Quality of Service in computer
networks. Additionally, her recent research interests lie
in the area of nanonetworks, with a specific focus on ar-
chitecture, protocols, security and authorization issues.
She has contributed to a number of EU, ESA and Na-
tional research projects. She is currently a researcher
at the Foundation of Research and Technology, Hellas Sotiris Ioannidis received a BSc degree in Mathemat-
(FORTH). ics and an MSc degree in Computer Science from the
University of Crete in 1994 and 1996 respectively. In
1998 he received an MSc degree in Computer Science
from the University of Rochester and in 2005 he received
his PhD from the University of Pennsylvania. Ioanni-
dis held a Research Scholar position at the Stevens In-
stitute of Technology until 2007, and since then he is
Research Director at the Institute of Computer Science
of the Foundation for Research and Technology - Hel-
las. Since November 2017 he is a member of the Eu-
ropean Union Agency for Network and Information Se-
Michail Christodoulou received his Diploma and curity (ENISA) Permanent Stakeholders Group (PSG).
Ph.D. degree in Electrical and Computer Engineering His research interests are in the area of systems, net-
from the Dept. of Electrical and Computer Engineering, works, and security. Ioannidis has authored more than
Aristotle University of Thessaloniki (AUTH) in 1994 100 publications in international conferences and jour-
and 2006, respectively. In 1995, he received an MSc. nals, as well as book chapters, including ACM CCS,
degree from the University of Bradford, UK. In 2014, ACM/IEEE ToN, USENIX ATC, NDSS, and has both
he joined AUTH as a Postdoctoral Research Fellow. chaired and served in numerous program committees
His research interests include RF circuits, antennas, RF in prestigious international conferences. Ioannidis is
MEMS, and electromagnetic measurements. a Marie-Curie Fellow and has participated in numer-
ous international and European projects. He has coor-
dinated a number of European and National projects
(e.g. PASS, EU-INCOOP, GANDALF, SHARCS) and
is currently the project coordinator of the THREAT-
ARREST, I-BiDaaS, BIO-PHOENIX, IDEAL-CITIES,
CYBERSURE, and CERTCOOP European projects.

76 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Academy of Sciences, in Moscow, Russia, since May

2018. He serves on the Advisory Board for the newly es-
tablished research center called Technology Innovation
Institute (TII) in Abu Dhabi, United Arab Emirates
since June 2020. He is the President of the Truva Inc.
since March 1989 and Scientific Advisor for the newly es-
tablished company Airanaculus since April 2020. He is a
Visiting Distinguished Professor with the SSN College of
Andreas Pitsillides is a Professor in the Depart- Engineering in Chennai, India since October 2019. Dr.
ment of Computer Science, University of Cyprus, heads Akyildiz is an Adjunct Professor with the Department
NetRL, the Networks Research Laboratory he founded of Electrical and Computer Engineering at the Univer-
in 2002, and is appointed Visiting Professor at the sity of Iceland since September 2020. He is the Ken
University of the Witwatersrand (Wits), School of Byers Chair Professor Emeritus in Telecommunications
Electrical and Information engineering, Johannesburg, at the Georgia Institute of Technology, and the Direc-
South Africa. Earlier (2014-2017) Andreas was ap- tor of the Broadband Wireless Networking Laboratory
pointed Visiting Professor at the University of Johan- and Former Chair of the Telecom Group from 1985-
nesburg, Department of Electrical and Electronic En- 2020. He is also former Finnish Distinguished Professor
gineering Science, South Africa. His broad research with the University of Tampere, Finland, supported by
interests include communication networks (fixed and the Finnish Academy of Sciences from 2012-2016. He
mobile/wireless), Nanonetworks and Software Defined is the founder of NanoNetworking Center and Former
Metasurfaces/Metamaterials, the Internet- and Web- of Honorary Professor at the University of Politecnica de
Things, Smart Spaces (Home, Grid, City), and Internet Cataluna in Barcelona from 2008-2017. Dr. Akyildiz is
technologies and their application in Mobile e-Services, also former Distinguished Professor and Founder of the
especially e-health, and security. He has a particular Advanced Wireless Networks Lab with the King Ab-
interest in adapting tools from various fields of applied dulaziz University in Jeddah, Saudi Arabia from 2011-
mathematics such as adaptive non-linear control theory, 2018. He is also the Founder of the Advanced Wireless
computational intelligence, game theory, and recently Sensor Networks lab and former ExtraOrdinary Profes-
complex systems and nature inspired techniques, to sor with the University of Pretoria, South Africa from
solve problems in communication networks. He has pub- 2007-2012.
lished over 270 referred papers in flagship journals (e.g. He is the Founder and Editor in Chief of the newly es-
IEEE, Elsevier, IFAC, Springer), international confer- tablished ITU J-FET (International Telecommunication
ences and book chapters, 2 books (one edited), par- Union Journal for Future and Evolving Technologies
ticipated in over 30 European Commission and locally since July 2020. Dr. Akyildiz is Editor-in-Chief Emer-
funded research projects as principal or co-principal in- itus of Computer Networks Journal (Elsevier) (1999-
vestigator, received several awards, including best pa- 2019), the founding Editor-in-Chief Emeritus of the
per, presented keynotes, invited lectures at major re- Ad Hoc Networks Journal (Elsevier) (2003-2019), the
search organisations, short courses at international con- founding Editor-in-Chief Emeritus of the Physical Com-
ferences and short courses to industry, and serves/served munication (PHYCOM) Journal (Elsevier) (2008-2017),
on several journal and conference executive committees. and the founding Editor-in-Chief Emeritus of the Nano
Communication Networks (NANOCOMNET) Journal
(Elsevier) (2010-2017). Dr. Akyildiz launched many
IEEE and ACM conferences. He is an IEEE Fel-
low (1995) and ACM Fellow (1996) and received nu-
merous awards from IEEE and ACM and other pro-
fessional organizations. His current research inter-
ests are in 6G Wireless Systems, Reconfigurable Intel-
ligent Surfaces, Terahertz Communications, Nanonet-
works, Internet of xThings (x=Underwater, Under-
ground, Space/CubeSats, Nano and BioNano). He
Ian F. Akyildiz received his BS, MS, and PhD de- graduated 45 PhD students and 30 of them are in
grees in Electrical and Computer Engineering from the academia in very prestigious academic positions world-
University of Erlangen-Nı̈¿œrnberg, Germany, in 1978, wide. He advised 13 Postdoctoral researchers. Accord-
1981 and 1984, respectively. Currently he serves as a ing to Google Scholar as of September 2020, his H-index
Consulting Chair Professor with the Computer Science is 124 and the total number of citations to his papers is
Department at the University of Cyprus since January 119+K. His rank in terms of h-index in the world is 46
2017. and in the USA 32.
He is also the Megagrant Leader with the Institute
for Information Transmission Problems at the Russian

© International Telecommunication Union, 2020 77

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

DESIGN AND ANALYSIS OF A RECONFIGURABLE INTELLIGENT META-SURFACE FOR VEHICULAR

NETWORKS

Mohammad Ojaroudi1 , Valeria Loscrí1 , Anna Maria Vegni2

1
Inria Lille - Nord Europe, 2 Roma Tre University, Rome, Italy

NOTE: Corresponding author: Valeria Loscrí, [email protected]

Abstract – In this work, a new paradigm for vehicular communications based on Reconfigurable Intelligent Meta-surfaces
(RIMs) is presented. By using the proposed RIM, we are able to manipulate electromagnetic waves in the half-space, since
the element is reflective. The unit cell consists of a U-shaped designed microstrip structure equipped with a pin diode and via
a hole. In this study, two different reflection modes are achieved for 1-bit data transferring in each state. By incorporating
these two different configurations together, the reflected phases in the proposed RIM surface can be controlled respectively in
0∘ and 180∘ . The proposed unit cell can provide a usable double negative functional characteristic around 5.3 GHz. The main
goal of this paper is the use of a multifunctional behavior RIM for vehicular communications to code the transmitted wave. A
novel phase distribution diagram is generated to propagate in each angle. Moreover, two major electromagnetic modulation
functions, beam forming and space coding have been demonstrated. Finally, we show how the RIM can be employed for
vehicular communications, acting as a coated access point along the street. We derive the instantaneous data rate at the re-
ceiver node, the outage probability and the channel capacity, as affected by different beam widths, distances and vehicle speed.

Keywords – Beam width, Instantaneous data rate, Reconfigurable meta-surfaces, V2X communications

1. INTRODUCTION of the propagating signal is not available. [22] and [23]

are among the first contributions of exploitation of RIM
In recent years, special attention has been paid to driver- in vehicular communications. In [22], the authors focus
less cars using automotive radars [7, 8]. A Vehicle-to- on the physical layer security based on the exploitation of
Everything (V2X) communication paradigm has emerged a RIM or RIS (Reconfigurable Intelligent Surface). They
as one of the most important enabling technologies for derive the average secrecy capacity of the system, by con-
vehicular networks. This has the potential of making sidering an ideal reconfigurable meta-surface. They do
streets and highways safer, the traffic more efficient and not give any detail about the design of a real meta-surface
less harmful to the environment [9]. An example is the with the needed characteristics in order to obtain the spe-
exchange of traffic information, as in the Self-Organizing cific behavior as described in the paper. Also in [23], the
Traffic Information System approach [12]. authors show the rich potentiality of the RIM integration
in a vehicular communication system, by analysing a hy-
One of the most important challenges arising in the V2X pothetic ideal controllable meta-surface. Anyway, no de-
paradigm is the high dynamic and the need of highly ef- tail about the features of a real meta-surface is provided in
ficient communication solutions in order to “follow” the the paper. In [17], we have proposed a unit cell configura-
rapid changing of communication nodes. Recently, an tion working at millimiter-waves in order to control the
innovative wireless communication paradigm based on phase shift from 0 to 180∘ . Specifically, we have investi-
the utilization of specific features of controllable meta- gated the unit cell design without evaluating the full struc-
surfaces has been investigated in [2, 3, 4, 5, 6, 15]. With ture and relate the capability to control the beam width
the advancement of technology in metamaterials, meta- based on the number of the unit cells of the full structure.
surfaces which are a thin two-dimensional structure of Indeed, the higher is the number of the unit cells of the full
metamaterials [1], due to their unique properties like the structure, the better is the capacity to control the beam
capability to provide abrupt phase shift, amplitude modu- width of the radiation pattern generated as the reflected
lation, and polarization conversion of the electromagnetic wave of an impinging signal.
(EM) wave, have found a variety of applications in various
fields of science, such as physics, engineering, biotechnol- As demonstrated in [11], it is possible to accurately model
ogy, and telecommunications [18]. the number of cells as sources of secondary radiation [16]
by applying the Huygens principle in the far-field limit. In
Reconfigurable Intelligent Meta-surfaces (RIMs) can be practice, the higher is the number of unit cells we consider
very effective for improving the performance of a V2X in the full structure, the better is the capability to control
communication system, especially when the Line-of-Sight the phase, but above all to control the beam width of the

© International Telecommunication Union, 2020 79

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

main beam for beam stearing objective. capacity, by demonstrating the great potentiality of
this type of structure.
In [18], it is demonstrated how this type of meta-surface
allows a digital control of EM waves, by associating two
The rest of the article is organized as follows. Section 2
coding elements with opposite reflection phases (i.e., 0∘
describes the specific scenario considered and details the
and 180∘ ) and considering them as digital bits (i.e., 0 and
proposed unit cell structure. Section 3 presents simu-
1 in the binary case). Reconfigurable meta-surface struc-
lated results for the proposed RIM unit cell in two ON and
tures can be applied to manipulate EM waves in a sim-
OFF states, expressed in terms of (i) reflection magnitude
ple and effective way, by changing the coding elements on
and phase, and (ii) effective permittivity and permeabil-
a 2D plane with predesigned coding sequences [19]. In
ity. In Section 4 we validate the full RIM structure, while in
order to independently control and create different cod-
Section 5 the performance of the proposed RIM has been
ing sequences a Field Programmable Gate Array (FPGA)
exploited for vehicular communications, by considering
is used. By changing the coding sequences stored in
that the RSU and the receiver vehicle are coated with the
the FPGA, many different functionalities can be switched
implemented 10 × 10 unit cells RIM. Instantaneous data
in real time, thereby leading to programmable meta-
rate, outage probability and channel capacity have been
surfaces.
obtained as validation results. In Section 6, we present a
In this paper, we formulate the specific features that a few considerations related to the obtained results. Finally,
meta-surface for vehicular communication applications conclusions are drawn at the end of the article.
needs to have in the frequency range of [5, 5.9] GHz. In
particular, we consider a tracking application with beam 2. UNIT-CELL DESIGN AND CONFIGURA-
steering, for which is of paramount importance to con- TION
trol the phase and to concentrate the power in the main
lobe of the reflected signal, as much as possible. In or- The envisioned vehicular communications paradigm
der to meet these specific requirements, we will design based on using RIM is shown in Fig. 1. In particular, we fo-
a unit cell and then derive a full structure constituted by cus on a RIM relay based scheme. An RSU (Road Side Unit)
the periodic repetitions of these unit cells, behaving as a is at the side of the road and is coated with the specific
reflector for a pair of transmitter receivers. We will focus meta-surface structure we will detail later. We assume
on the phase shift and the main beam width in order to that the RSU, as represented in Fig. 1, is re-transmitting
design a beam tracking system which is able to “follow” data to the receiver (i.e., the green vehicle). In order to
the mobile receiver node for improving the efficiency in maximize the SNR to the destination, the RIM RSU will be
terms of data rate and outage probability. Based on that, “beam tracking” the receiver. This specific behavior can
we will present a multifunctional reconfigurable meta- be realized by designing a unit cell with specific features
surface structure based on the radiation pattern modu- as explained to the follow.
lation of the reflection coefficient. Firstly, we will design
a reconfigurable U-shaped unit cell using a pin diode via The proposed unit cell structure is shown in Fig. 2 (a). In
a hole, which can provide 180∘ -phase difference between this structure we proposed a U-shaped microstrip struc-
ON and OFF states. Specifically, a 10 × 10 meta-surface ture which is able to be reconfigurable by a PIN diode. By
loaded with PIN diodes is designed for multifunctional be- creating a rectangular slot in the corner of the rectangu-
havior, such as coding and beam steering. The simulated lar patch a new path of surface current will be created.
results in both scenarios of unit cell and full structures Hence, we can control the input impedance of the unit cell
will show the effectiveness of the proposed structure for which is suitable for such an application of RIM. Moreover,
vehicular communications. the relevant equivalent circuit for this structure is shown
in Fig. 2 (b). Regarding the ON and OFF states of the PIN
Our main contributions can be summarized as follows: diode, we put two different equivalent circuits for this sec-
• We design a specific meta-atom working at 5.3 GHz tion. All the dimensions are summarized in Table 1.
for automotive applications and based on this unit
cell, we design a full structure and validate it to as-
sess its suitability for the vehicular application con- Table 1 – Parameters of the proposed unit cell
sidered;
Parameter Value [mm] Parameter Value [mm]
• We design a full structure and validate it to assess its 𝑊𝑆𝑢𝑏 10 𝐿𝐶 9
suitability for the vehicular application considered; 𝐿𝑆𝑢𝑏 10 𝑊𝐶 6
ℎ𝑆𝑢𝑏 1.6 𝑊𝐶1 2
• We consider the integration of the designed Recon- 𝐿𝑑 1 𝐿𝐶1 3
figurable Intelligent Meta-surface (RIM) in a vehicu- 𝑊𝑑 3 𝑅𝑣𝑖𝑎 0.25
lar system and we numerically evaluate the perfor-
mance; We compare the results of the system with
and without RIM in terms of outage probability and In meta-surface structures, beam steering can be consid-

80 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

RIM
becomes
𝑀 𝑁
𝐹 (𝜃, Φ) = ∑ ∑ 𝐴𝑚𝑛 𝑒𝑗/𝛼𝑚𝑛 𝑓𝑚𝑛 (𝜃𝑖 , Φ𝑖 )⋅
V2I 𝑚=1 𝑛=1 (2)
RIM RSU
I2V 𝑗/Φ𝑚𝑛 𝑗𝑘0𝑚𝑛 (𝜃,Φ)
V2V V2V Γ𝑚𝑛 𝑒 𝑓𝑚𝑛 (𝜃, Φ)𝑒 ,
where 𝐴𝑚𝑛 and 𝛼𝑚𝑛 are the amplitude and phase of the
wave incident to the (𝑚, 𝑛)-th unit cell, respectively, with
𝑚 = [1, 2, … , 𝑀 ] and 𝑛 = [1, 2, … , 𝑁 ]. In Eq. (2), Γ𝑚𝑛
Fig. 1 – Vehicular communications paradigm based on the use of and Φ𝑚𝑛 are the amplitude and phase of the response of
RIMs as relay node, for data transmission to a receiver node (green the (𝑚, 𝑛)-th unit cell, respectively; 𝑓𝑚𝑛 (𝜃, 𝜙) denotes the
vehicle), in case of Vehicle-to-Vehicle (black lines) and Vehicle-to- scattering diagram of the (𝑚, 𝑛)-th unit cell towards an
Infrastructure/Infrastructure-to-Vehicle (blue lines).
arbitrary direction of reflection, whereas 𝑓𝑚𝑛 (𝜃𝑖 , Φ𝑖 ) de-
notes the response of the (𝑚, 𝑛)-th unit cell at the direc-
tion of incidence determined by 𝜃𝑖 , Φ𝑖 and 𝑘0 = 2𝜋/𝜆0 is
the wave number. Finally, we introduce 𝜁𝑚𝑛 (𝜃, Φ), which
denotes the relative phase shift of the unit cells with re-
spect to the radiation pattern coordinates, given by
1 1
𝜁𝑚𝑛 (𝜃, Φ) = 𝐷𝑢 𝑠𝑖𝑛(𝜃) [(𝑚 − ) 𝑐𝑜𝑠(Φ) + (𝑛 − ) 𝑠𝑖𝑛(Φ)] ,
2 2
(3)
(a) with 𝐷𝑢 [m] as the unit cell size.
In order to make the model able to be calculated, we make
a further assumption in the point of plane incident wave
view, so that factors 𝐴𝑚𝑛 , 𝛼𝑚𝑛 , and 𝑓𝑚𝑛 (𝜃𝑖 , Φ𝑖 ) are con-
stants for all 𝑚 and 𝑛 indexes. In addition, we apply the
widespread assumption to the scattering pattern of the
unit cell, which is modeled over the positive semisphere
with the function 𝑐𝑜𝑠(𝜃), which is a widespread assump-
tion, [11]. Finally, and without loss of generality, we con-
sider the normal incidence i.e., (𝜃𝑖 = Φ𝑖 = 0). Then,
(b) Eq. (2) becomes [11]
𝑀 𝑁
Fig. 2 – (a) The configuration and the geometry of the proposed unit
cell with a U-shaped radiating patch, and (b) unit cells with PIN diode’s 𝐸(𝜃, Φ) = 𝑘 𝑐𝑜𝑠(𝜃) ∑ ∑ Γ𝑚𝑛 𝑒𝑗 [Φ𝑚𝑛 + 𝑘0 𝜁𝑚𝑛 (𝜃, Φ)],
equivalent circuit model for ON and OFF states. 𝑚=1 𝑛=1
(4)
ered as a particular case of wavefront manipulation that with 𝑘 as a constant.
occurs in the far field. Regarding the Huygens princi-
ple, the meta-surface structures can be considered as an In order to have anomalous reflection, the main objective
integrated EM radiator array [20]. Herein, in order to is controlling the phase shift of the unit cells Φ𝑚𝑛 . In par-
model the meta-antenna array, we consider a method that ticular, we manipulate the phase of the reflected wave-
has been validated in several works via extensive simula- form but not its amplitude. In this current version we do
tions [21]. Considering each unit cell as an element of the not focus on the control scheme for our system, since it is
planar array, the far field of the meta-surface can be ob- out of scope for this work. In reconfigurable meta-surface
tained as: generating different coding sequence for unit cells, we are
able to achieve desired functionalities such as beam steer-
ing and radiated wave modulation. In this regard, the am-
𝐹 (𝜃, Φ) = 𝑓𝐸 (𝜃, Φ) ⋅ 𝑓𝐴 (𝜃, Φ), (1) plitude Γ𝑚𝑛 and phase Φ𝑚𝑛 of the (𝑚, 𝑛)-th unit cell need
to be determined somehow which the entire response of
by considering infinite sphere, 𝜃 is the elevation angle, Φ the array matches with the required functionality. After
is the azimuth angle of an arbitrary direction in this coor- this step, by mapping the required Γ and Φ to the clos-
dination. est available unit cell states, the desired functionality will
be obtained. In the case of anomalous reflection for beam
Regarding the planar array, the pattern function of each steering, analytical methods provide high accuracy.
unit cell 𝑓𝐸 (𝜃, Φ) is the element factor and the pattern
function of full planar configuration 𝑓𝐴 (𝜃, Φ) is the array In this study, in order to obtain beam steering function-
factor. In far field region, we assume a planar wave covers ality, the phase gradient approach is used to determine
the entire meta-surface. Therefore, the radiated pattern the direction of reflection [13]. Considering Φ(𝑥, 𝑦) as the
will depend only on the array factor. In this case, the far phase profile which is imposed by reconfigurable meta-
field pattern for the meta-surface with 𝑁 × 𝑀 unit cells, surface, the virtual wave vector KΦ = Φ𝑥 x̂ + Φ𝑦 ŷ can be

© International Telecommunication Union, 2020 81

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

assigned to the meta-surface. In this context, the momen-

tum conservation law for wave vectors can be expressed
as
𝜕Φ𝑥
𝑘𝑖 𝑠𝑖𝑛(𝜃𝑖 ) 𝑐𝑜𝑠(Φ𝑖 ) + = 𝑘𝑟 𝑠𝑖𝑛(𝜃𝑟 ) 𝑐𝑜𝑠(Φ𝑟 ) (5)
𝜕𝑥

𝜕Φ𝑦
𝑘𝑖 𝑠𝑖𝑛(𝜃𝑖 ) 𝑐𝑜𝑠(Φ𝑖 ) + = 𝑘𝑟 𝑠𝑖𝑛(𝜃𝑟 ) 𝑠𝑖𝑛(Φ𝑟 ) (6) (a)
𝜕𝑦
where 𝜕Φ𝑥 /𝜕𝑥 and 𝜕Φ𝑦 /𝜕𝑦 describe the gradients along
the x-̂ and y-̂ directions, respectively. For simplicity we
consider the normal incident wave case i.e., (𝜃𝑖 = Φ𝑖 = 0)
in lossless medium scenario [14]. Assuming air as the
medium of the incident and reflected wave, we can sim-
plify the formulations above as
2𝜋𝜕𝑥 𝑐𝑜𝑠 Φ𝑟 𝑠𝑖𝑛 𝜃𝑟 2𝜋𝜕𝑦 𝑠𝑖𝑛 Φ𝑟 𝑠𝑖𝑛 𝜃𝑟
𝜕Φ𝑥 = , 𝜕Φ𝑦 = ,
𝜆0 𝜆0
(b)
(7)
which demonstrate the phase shift Φ𝑥 and Φ𝑦 that need to Fig. 3 – Simulated results for the proposed unit cell in two ON and OFF
be performed per unit of distance (i.e., 𝜕𝑥 and 𝜕𝑦) along situations, expressed in terms of (a) reflection magnitude and (b) reflec-
the x̂ and ŷ directions, respectively. Then, in Eq. (6) we set tion phases.
the unit cell size as 𝜕𝑥 = 𝜕𝑦 = 𝐷𝑢 , in order to obtain the In order to assess the double negative characteristics, 𝑆11
phase required at the (𝑚, 𝑛)-th unit cell as and 𝑆21 reflection and transmission coefficients are ex-
2𝜋𝐷𝑢 (𝑚 𝑐𝑜𝑠 Φ𝑟 𝑠𝑖𝑛 𝜃𝑟 + 𝑛 𝑠𝑖𝑛 Φ𝑟 𝑠𝑖𝑛 𝜃𝑟 ) tracted from the design in CST in both magnitude and an-
Φ𝑚𝑛 = . (8) gle (expressed in rad). Then, the effective permittivity
𝜆0
𝜀𝑒𝑓𝑓 and effective permeability 𝜇𝑒𝑓𝑓 are obtained respec-
tively as [20]:
For beam-steering functionality the required phase Φ𝑚𝑛
is calculated for all the unit cells, to assign radiated states 1
to each unit cell. Then, a closest neighbor mapping is done 𝑘𝑑 𝑐𝑜𝑠−1 [ 2𝑆1 (1 − 𝑆11
2 2
+ 𝑆21 )]
𝜀𝑒𝑓𝑓 = 21
, (9)
between the required phase and that provided by the dif- 2 2
√ (1+𝑆11 )2 −𝑆21
2
ferent unit cell states. (1−𝑆11 ) −𝑆21

and
3. SIMULATION RESULTS
In this section, we evaluate the performance of the unit 1 1 (1 + 𝑆11 )2 − 𝑆21
2
𝜇𝑒𝑓𝑓 = 𝑐𝑜𝑠−1 [ 2
(1 − 𝑆11 2
+ 𝑆21 )] √ ,
cell we designed. The simulation is realized by the means 𝑘𝑑 2𝑆21 2 2
(1 − 𝑆11 ) − 𝑆21
of a commercial software CST studio suite. The specific (10)
configuration considered in CST is the boundary condi-
where 𝑘 denotes the wave number of the incident wave
tion as unit cell in x-̂ and y-directions
̂ with open-add
and 𝑑 [mm] is the thickness of the unit cell. In this study,
space in z-direction.
̂ The reflectivity and reflection phases
the meta-surface is printed on an FR-4 substrate with
of unit cells are simulated using the frequency domain
thickness of 1.6 mm and is designed to increase the direc-
solver. The simulated reflection magnitude and phase of
tivity and bandwidth of the structure. Fig. 4 shows the ef-
the unit cell are shown in Fig. 3. It is obvious that at 5.3
fective permittivity and effective permeability related to
GHz, while the reflection magnitude is almost identical
the designed unit cell. It is clear that the proposed unit
between the ON and OFF state, the reflection phase be-
cell has double-negative material characteristics around
tween the two cases has a 180∘ change. The maximum
5.3 GHz for both ON and OFF states.
unit cell loss is around 2.5 dB for the OFF state. It is ev-
ident that the proposed 1-bit unit cell is suitable for the
multifunctional meta-surface such as coding and beam 4. FULL STRUCTURE DESCRIPTION AND
steering. As the phase change between ON/OFF states is
EVALUATION
relative, we can simply state that a unit cell with an ON
state corresponds to a 0∘ phase reflection, while one with In this section, in order to evaluate the performance of
an OFF state has a −180∘ phase reflection. the proposed unit cell for multifunctional applications, a

82 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Fig. 4 – The effective permittivity and permeability of the unit cell.

Fig. 6 – Reflection phase distribution and 3D pattern of the coding meta-

surface with Horn antenna as an EM source located at (0 cm, 0 cm, 100
cm). (a) Ideal reflection phase distribution; (b) 3D pattern in full struc-
ture; (c) the beam pattern based on ideal phase distribution; (d) the
beam pattern contour for coding application.

beam. Fig. 6 (d) demonstrates contour plot with corre-

sponding 3D scattering pattern.
(a)

4.2 Beam-Steering Meta-surface Construction

In this section, an 8 × 8 beam steering meta-surface is
modeled and simulated using CST Studio software to ver-
ify the beam steering capability of the RIM. In this sim-
ulation, the meta-surface is in the X-Y plane, and a horn
serves as an EM source, which is located at (50 mm, 0 cm,
(b) 0 cm) with a rotation of (45∘ , 0∘ ) with respect to the meta-
surface. Then, an ON/OFF pattern matrix for steering to
Fig. 5 – (a) Horn antenna as a transmitter and (b) the 1-bit 10 × 10
meta-surface simulated model. (120∘ , 0∘ ) is loaded to the PIN diode of each unit cell. Fi-
nally, the simulation results are exported and shown in
Horn antenna is employed as a main radiator at the trans- Fig. 7 (a) and (b). It is clearly observed that the cod-
mitting side, as shown in Fig. 5 (a). The redesigned Horn ing meta-surface is capable of steering the beam to the
antenna is given to excite the meta-surface structure at desired direction with 20∘ angular resolution in the full
5.3 GHz. In addition, the 1-bit 10 × 10 meta-surface simu- structure case.
lated model is shown in Fig. 5 (b).

5. EVALUATION IN VEHICULAR APPLICA-

4.1 Coding Meta-surface Construction TIONS
In this section we consider coding meta-surface based on The proposed RIM structure is now exploited for vehic-
the RIM structure, as the possibility to characterize the ular communications, as depicted in Fig. 1. Specifically,
states ON and OFF as matching the bits 0 and 1. Fig. 6 we consider the communication link in a vehicular con-
(a) shows a random coding meta-surface with a fixed ra- text, established from a source (red vehicle) and a receiver
tio and a different coding sequence, and Fig. 6 (b) shows node (green vehicle). We assume a highway scenario and
the simulated 3D radiation in full structure using CST. As the transmitter and receiver vehicles have constant speed
shown clearly, with coding sequence, the diffusion of the for the transmission time window. In order to character-
far-field pattern and the scattering amplitude at the nor- ize the impact of the controllable meta-surface in the com-
mal incident angle are apparently the same. According to munication system, we assume a simplified system with
the code MATLAB, a 3D far-field pattern can be obtained no interferences caused by others vehicles. In order to
with fixed ratio between 0 and 1 coding elements. The quantify the impact of the RIM on the performance sys-
simulated results are demonstrated as Fig. 6 (c). Once ra- tem, we will consider (i) the relation between different
tio fixed, the scattering amplitude at the normal incident beam width on the average rate, given a certain estimated
angle 0∘ is determined, and the efficient of coding meta- velocity, and (ii) the derivation of the outage probability
surface only depends on the uniformity of the scattering and the analysis of the communication system in terms of

© International Telecommunication Union, 2020 83

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Table 2 – Simulation Parameters

Parameter Value
𝐸𝐼𝑅𝑃 57 dBm
𝐵 75 MHz
𝑓𝑐 5.3 GHz
𝑟 [0, 200] m
(a) 𝑁𝐹 6 dB
𝑣 [25, 50, 75] km/h

receiver node. The term 𝒦 is defined as

10 𝑙𝑜𝑔10 (𝒦) = 𝐸𝐼𝑅𝑃𝑑𝐵𝑚 − 𝐸 + 10𝑛 𝑙𝑜𝑔10 (𝜆/4), (15)

where 𝐸𝐼𝑅𝑃 is the Equivalent Isotropic Radiated Power

(b) and 𝜆 [m] is the wavelength and 𝐸 represents the shad-
Fig. 7 – The simulation results of 3D gain total with the horn and meta- owing margin. In order to compute 𝒦, we consider the
surface and the 2D radiation pattern for 10 × 10 with 50 mm distance power associated with the reflected signal 𝑆21 as com-
and 25∘ rotation at Y-plane at 5.25 GHz with different phase distribution, puted in Eqs. (9) and (10).
(a) uniform distribution and (b) random distribution.
Notice that we assume that the transmitter node is trans-
both outage and capacity, in cases with and without RIM. mitting at an EIRP that for Europe is equal to 57 dBm. The
relay will not be able to retransmit at full power since
5.1 Beam width Impact Evaluation a percentage of the power associated to the impinging
A 10 × 10 unit cells RIM is coating both the RSU (acting wave will be dissipated as a reflected signal. In particu-
as relay node) and the receiver vehicle. In particular, we lar, the design of the unit cell has been optimized in order
compute the instantaneous data rate by assuming a per- to minimize this power loss i.e., a 2% of the total transmit-
fect alignment between the beam from the RSU and the ted power will be wasted as a reflected component of the
receiver node. The instantaneous data rate is derived by wave. Furthermore, without loss of generality, the speed
the Shannon capacity formula for the instantaneous rate of the receiver vehicle is considered constant during the
as: transmission time, since the interval time of reception is
𝑅(𝑡, 𝜃𝑏 ) = 𝐵 𝑙𝑜𝑔2 (1 + 𝑆𝑁 𝑅(𝑡, 𝜃𝑏 )), (11) short with respect to the variation of the vehicle speed.
Table 2 collects the parameters used in the numerical re-
where 𝐵 [Hz] is the bandwidth of the system, 𝑆𝑁 𝑅 [dB] sults.
is the Signal-to-Noise Ratio, 𝑡 [s] is the time instant, and
𝜃𝑏 is the beam width computed in degrees.
In Fig. 8 (a) we show the instantaneous and ideal rate,
We assume that our system is able to “instantaneously” i.e., the maximum achievable rate with the assumption of
switch from the (𝑖 − 1)-th beam to the 𝑖-th one, and the a perfect alignment and without error estimation of the
“transmitter” beam and the “receiver” one are perfectly velocity at the receiver by considering a narrow beam
aligned. Moreover, we consider a beam model where the width, i.e., 𝜃𝑏 = [0.1, 10]∘ . Data rate values range from
gain inside the beam is uniform and zero outside. In ≈ [15, 25] Gbps, where higher values are obtained for
Eq. (11), the expression of the 𝑆𝑁 𝑅 can be written as smaller beam widths, while a data rate decrease is ob-
served for larger beam widths.
𝑃𝑟𝑥 (𝑡, 𝜃𝑏 )
𝑆𝑁 𝑅(𝑡, 𝜃𝑏 ) = , (12)
𝑃𝑛𝑜𝑖𝑠𝑒
Furthermore, we have considered the impact of different
where 𝑃𝑛𝑜𝑖𝑠𝑒 [dB] represents the thermal noise including speeds on the achievable rate and we can observe that,
a Noise Figure i.e., 𝑁 𝐹 [dB], which can be expressed as as expected, the data rate decreases for increasing vehicle
speed. Finally, the effect of the distance on the data rate is
𝑃𝑛𝑜𝑖𝑠𝑒 = −174 + 10 𝑙𝑜𝑔10 𝐵 + 𝑁 𝐹 . (13) a decrease of performance when the distance increases.
In Eq. (12), the term 𝑃𝑟𝑥 is the received power, expressed In Fig. 8 (b) we show the impact of a wider beam width
in linear scale as [10]: for the same scenario. Of course, wider beam widths are
2 related with a smaller number of unit cells. Indeed, the
𝜋2 1 smaller is the number of unit cells, the lower is the fine
𝑃𝑟𝑥 (𝑡, 𝜃𝑏 ) = 𝒦 ( 2
) 𝑛/2
, (14)
𝜃𝑏 [(𝑣𝑡)2 + 𝑟2 ] control we can realize on the RIM. As expected, we ob-
serve a decreasing of the instantaneous data rate from a
where 𝑣 [m/s] is the vehicle speed, 𝑛 is the path loss ex- maximum of 23 Gbps in Fig. 8 (b) to a maximum of 15 Gbps
ponent, and 𝑟 [m] is the distance from the RSU and the in Fig. 8 (a).

84 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

(a) (b) (c)

Fig. 8 – Evaluation of the instantaneous data rate [Gbps] vs. the beam width and the vehicle speed, in case of (a) 𝜃𝑏 = [0.1, 10]∘ and (b) 𝜃𝑏 = [30, 120]∘ ,
both by considering different speeds i.e., 𝑣1 = 25, 𝑣2 = 50, and 𝑣3 = 75 [m/s]. (c) Instantaneous data rate [Gbps] vs. vehicle speed and distances, in
case of 𝜃𝑏 = [20, 70, 120]∘ .

1
Finally, in Fig. 8 (c) we analyze the impact of the speed RIM, = 20 °
b
i.e., from 10 to 100 km/h, with respect to specific values of RIM, = 90 °
b
beam width i.e., 𝜃𝑏 = [20, 70, 120]∘ . The behavior follows 0.8
no RIM

Outage Probability
the same trend as observed in Fig. 8 (a) and (b), where,
0.6
as expected, for higher beam width the data rate perfor-
mance is strongly reduced. Also, lower distances provide
0.4
higher performance.
0.2

5.2 Outage Probability

0
The outage probability is a crucial metric for vehicular 0 100 200 300 400 500
Distance [m]
communication [24, 25], and can be defined as the instan-
taneous mutual information rate that falls below a certain Fig. 9 – Outage Probability vs. the distance, in case of (i) a meta-surface
threshold 𝛽, i.e.: with beam width 𝜃𝑏 = 20∘ and 𝜃𝑏 = 90∘ , and (ii) without a meta-
surface.
𝑃 𝑟 [𝑆𝑁 𝑅 < 𝛽] = 1 − 𝑃 𝑟 , (16) meta-atoms needed in the metastructure for a high tun-
𝑜𝑢𝑡𝑎𝑔𝑒 𝑠𝑢𝑐𝑐𝑒𝑠𝑠
ing of the phase. In practice, the higher is the number of
where 𝑃 𝑟𝑠𝑢𝑐𝑐𝑒𝑠𝑠 is the success probability that SNR is unit cells of the full structure, the better is the control of
higher than 𝛽, and can be expressed as the whole system, but also the control logic to drive the
controllable meta-surface is more sophisticated/compli-
𝑃𝑛𝑜𝑖𝑠𝑒
𝑃𝑟 = 𝑃 𝑟[𝑆𝑁 𝑅 > 𝛽] = 𝑒𝑥𝑝 (−𝛽𝑑𝑛 ) , (17) cated, since it is demanding a higher precision and could
𝑠𝑢𝑐𝑐𝑒𝑠𝑠 𝑃𝑡𝑥 make the system more vulnerable to misalignment errors.
with 𝑃𝑡𝑥 as the transmitting power.
Based on the outage probability, given a specific target 6. DISCUSSION
rate as a Quality of Service (QoS) parameter, we can de-
In order to assess the performance of the system and to
rive the throughput of success delivery with constrained
evaluate the potential impact of the RIM in a vehicular
outage probability 𝜖, as [24]:
context, we have considered some assumptions to sim-
𝐶 = (1 − 𝜖)𝐵𝑙𝑜𝑔2 (1 + 𝑆𝑁 𝑅). (18) plify the analysis and give some useful insights for fu-
ture work. Firstly, a perfect alignment between the relay
In order to evaluate the impact of the presence of the coated with the RIM and the receiver has been considered.
meta-surface, in Fig. 9, we consider a target rate of 1 Gbps This assumption cannot be ensured above all in a con-
and derive the corresponding outage probability as com- text characterised with high speed. In order to account
puted in Eq. (16). In order to better appreciate the ef- for that, we need to introduce an error factor of the speed
fect of the meta-surface, we consider two different beam to account for the misalignment. A possible approach for
widths of 20∘ and 90∘ . As expected, the lower is the beam accounting the misalignment has been proposed in [26].
width, the lower is the outage probability and the system We believe that this misalignment could be translated as a
has better performance in terms of capacity (see Fig. 10). reduction of the received power and then in a decreasing
The better performance is related to a higher precision of of the SNR.
the system, that corresponds to an increased number of Another simplification introduced is the ideal configura-

© International Telecommunication Union, 2020 85

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

20
RIM, = 20 °
with RIM with two different configurations in terms of
b
RIM, = 90 ° beam width, and without RIM.
b
15 no RIM
As a future piece of work, a factor that we have not consid-
Capacity [Gbps]

ered in this context is the presence of other sources and

10 interferences. We trust the integration of the RIM in the
system can improve the reduction of the interference im-
pact by properly addressing it. In order to account for this
5
point, we plan to extend the analytical model with the in-
terference factor and design RIM-based approaches that
0 reduce the impact of the interference.
0 100 200 300 400 500
Distance [m]

Fig. 10 – Capacity of the system with a rate set to 1 Gbps, in case of (i) a ACKNOWLEDGMENT
meta-surface with beam width 𝜃𝑏 = 20∘ and 𝜃𝑏 = 90∘ , and (ii) without
a meta-surface.
This work is partially supported by the Exploratory Ac-
tion ETHICAM.
tion that is computed in real time and without error. In-
deed, in order to control the RIM and make it as a tunable
and intelligent meta-surface, we need to include an intel- REFERENCES
ligent logic, potentially based on a machine learning (ML) [1] A. Nemati, Q. Wang, M. Hong, and J. Teng, “Tunable
approach to generate the correct configuration of the RIM and reconfigurable meta-surfaces and metadevices,”
when an external change impacts on the system (e.g., the in Opto-Electronic Advances, 01, 180009, 2018.
vehicle moves far away from the RSU and a new beam
needs to be calculated). ML approaches for systems based [2] C. Liaskos, A. Tsioliaridou, S. Nie, A. Pitsillides, S.
on RIM is a topic that is gaining more and more interest in Ioannidis, and I. F. Akyildiz, “On the Network-layer
recent times. Anyway, since we considered a mobile sce- Modeling and Configuration of Programmable Wire-
nario where some crucial parameters can be easily pre- less Environments,” in IEEE/ACM Transactions on
dicted (e.g., the velocity of the vehicle and the direction), Networking, vol. 27, no. 4, pp. 1696-1713, Aug.
we trust that the computation of a new configuration can 2019.
be calculated in real time and this should not impact too
much on the ideal performance of the system. [3] C. Liaskos, A. Tsioliaridou, A. Pitsillides, S. Ioanni-
dis, and I. F. Akyildiz, “Using any Surface to Realize
a New Paradigm for Wireless Communications,” in
7. CONCLUSIONS Communications of the ACM, vol. 61 no. 11, pp. 30-
33, October 2018.
In this paper, we presented a multifunctional reconfig-
urable meta-surface based on the radiation pattern mod- [4] C. Liaskos, S. Nie, A. Tsioliaridou, A. Pitsillides, S.
ulation of the reflection coefficient. We designed a recon- Ioannidis, and I. F. Akyildiz, “A New Wireless Com-
figurable U-shaped unit cell using a pin diode via a hole, munication Paradigm through Software-controlled
that allows to obtain a 180∘ -phase difference between ON Meta-surfaces,” in IEEE Communications Magazine,
and OFF states. The proposed RIM structure has been de- vol. 56, no. 9, pp. 162-169, September 2018.
signed for multifunctional behavior such as coding and
beam steering, considering a 10 × 10 meta-surface loaded [5] L. La Spada, V. Loscrí, and A.M. Vegni, “Meta-Surface
with PIN diodes. Structure Design and Channel Modelling for THz
Band Communications,” in Proc. of INFOCOM 2019
Simulation results have been carried out in order to val- Workshop, 1st IEEE Workshop on Ultra-High Broad-
idate the proposed RIM in a vehicular scenario, for data band Terahertz Communication for 5G and Beyond
transmission via the V2X mode. The instantaneous data Networks (UBTCN 2019), April 29-May 2, Paris,
rate has been obtained, as a function of the beam width, France.
the vehicle speed and the distance. As expected, the lower [6] A. Taibi, A. Durant, V. Loscrí, A.M. Vegni, and L.
the beam width is, the higher the data rate is; addition- La Spada, “Controlling Light by Curvilinear Meta-
ally, the lower the vehicle speed is, the higher the data Surfaces,” in Proc. of ACM Nanocom 2019, Septem-
rate is. Same consideration occurs for the distance. In or- ber 25-27, 2019, Dublin, Ireland.
der to assess the effectiveness of the RIM integration in
the vehicular communication system, we have derived the [7] J. Dickmann et al., “Automotive radar the key tech-
outage probability for a specific target QoS in terms of an nology for autonomous driving: From detection and
instantaneous rate. We also have computed the capacity ranging to environmental understanding,” in Proc. of
based on the outage probability and we have compared 2016 IEEE Radar Conference (RadarConf), Philadel-
both the outage and the capacity for a system equipped phia, PA, 2016, pp. 1-6.

86 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[8] O. Schumann, J. Lombacher, M. Hahn, C. Wöhler and meta-surfaces,” in Nature Communication, 9, 4334
J. Dickmann, “Scene Understanding With Automo- (2018).
tive Radar,” in IEEE Transactions on Intelligent Ve-
hicles, vol. 5, no. 2, pp. 188-203, June 2020, doi: [19] L. Zhang, R. Y. Wu, G. D. Bai, H. T. Wu, Q. Ma, X. Q. Chen,
10.1109/TIV.2019.2955853. T. J. Cui, “Transmission, Reflection, Integrated Mul-
tifunctional Coding Meta-surface for Full,” in Space
[9] J. Wang, Y. Shao, Y. Ge, and R. Yu, “A Survey of Vehi- Controls of Electromagnetic Waves, in Adv. Funct.
cle to Everything (V2X) Testing,” in Sensors (Basel). Mater. 2018, 28, 1802205.
2019;19(2):334. Published 2019 Jan 15.
[20] C. A. Balanis, Antenna Theory: Analysis and Design,
[10] J. Kim and A. F. Molisch, “Enabling Gigabit ser- 3rd ed., Wiley, Ed., 2005.
vices for IEEE 802.11ad-capable high-speed train
networks,” in IEEE Radio and Wireless Symposium [21] E. Hosseininejad, K. Rouhi, M. Neshat, A. Cabellos-
(RWS), pp. 145-147, Jan. 2013. Aparicio, S. Abadal, and E. Alarcon, “Digital Meta-
surface Based on Graphene: An Application to Beam
[11] H. Yang, X. Cao, F. Yang, J. Gao, S. Xu, M. Li, X. Chen, Steering in Terahertz Plasmonic Antennas,” in IEEE
Y. Zhao, Y. Zheng, and S. Li, “A programmable meta- Transactions on Nanotechnology, vol. 18, no. 1, pp.
surface with dynamic polarization, scattering and 734–746, 2019.
focusing control,” in Scientific Reports, vol. 6, no.
[22] A. Makarf, K. M. Rabie, O. Kaiwartya, K. Adhikari,
35692, 2016.
X. Li, M. Quiroz-Castellanos, and R. Kharel, “Re-
[12] A. Khan, F. Ullah, Z. Kaleem, S. ur Rahman and Y. Z. configurable intelligent surfaces-enabled vehicular
Cho, “Sensor-based self-organized traffic control at networks: A physical layer security perspective,”
intersections,” in Proc. of 2017 International Con- arXiv:2004.11288, 2020.
ference on Information and Communication Tech-
[23] B. Masini, M., Silva, M. Cristiano, and A. Balador,
nology Convergence (ICTC), Jeju, 2017, pp. 634-638,
“The Use of Meta-Surfaces in Vehicular Networks,”
doi: 10.1109/ICTC.2017.8191056.
in Journal of Sensor and Actuator Networks, 2020,
[13] N. Yu, P. Genevet, M. a. Kats, F. Aieta, J.-P. Tetienne, vol. 9, no 1, p. 15.
F. Capasso, and Z. Gaburro, “Light Propagation with
[24] X. Wu, S. Sun, Y. Li, Z. Tan, W. Huang, and X.
Phase Discontinuities: Generalized Laws of Reflec-
Yao, “A Power Control Algorithm Based on
tion and Refraction,” in Science, vol. 334, no. October,
Outage Probability Awareness in Vehicular Ad
pp. 333–337, 2011.
Hoc Networks,” in Advances in Multimedia,
[14] S. Liu, T. Jun Cui, A. Noor, Z. Tao, H. Chi Zhang, G. vol. 2018, Article ID 8729645, 8 pages, 2018.
Dong Bai, Y. Yang, and X. Yang Zhou, “Negative re- https://doi.org/10.1155/2018/8729645
flection and negative surface wave conversion from
[25] B. E. Y. Belmekki, A. Hamza and B. Escrig, “On the
obliquely incident electromagnetic waves,” in Light:
Outage Probability of Vehicular Communications
Science and Applications, vol. 7, no. 5, pp. 18 008–18
at Intersections Over Nakagami-𝑚 Fading Chan-
011, 2018.
nels,” in Proc. of 2020 IEEE 91st Vehicular Technol-
[15] T. Hamidreza, A. Cabellos-Aparicio, J. Georgiou, and ogy Conference (VTC2020-Spring), Antwerp, Bel-
S. Abadal, “Error Analysis of Programmable Meta- gium, 2020, pp. 1-5, doi: 10.1109/VTC2020-
surfaces for Beam Steering,” in IEEE Journal on Spring48590.2020.9128618.
Emerging and Selected Topics in Circuits and Sys-
[26] V. Va, X. Zhang and R. W. Heath, “Beam Switch-
tems 10, no. 1 (2020): 62–74.
ing for Millimeter Wave Communication to Sup-
[16] H. Taghvaee et al., “Scalability Analysis of Pro- port High Speed Trains,” in Proc. of 2015 IEEE
grammable Meta-surfaces for Beam Steering,” in 82nd Vehicular Technology Conference (VTC2015-
IEEE Access, vol. 8, pp. 105320-105334, 2020, doi: Fall), Boston, MA, 2015, pp. 1-5, doi: 10.1109/VTC-
10.1109/ACCESS.2020.3000424. Fall.2015.7390855.

[17] C. Rizza, V. Loscrí, and M. Parchin, “A Millimeter-

Wave Reconfigurable Intelligent Meta-surface De-
sign for Vehicular Networks Applications,” in Proc.
of 2020 IEEE Vehicular Technology Conference, Oct
2020, Victoria, Canada.

[18] L. Zhang, X. Qing Chen, S. Liu, Q. Zhang, J. Zhao, J.

Yan Dai, G. Dong Bai, X. Wan, Q. Cheng, G. Castaldi,
V. Galdi and T. Jun Cui, “Space-time-coding digital

© International Telecommunication Union, 2020 87

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

AUTHORS MASCOT, etc.), Italian and French projects. She is on the

editorial board of IEEE COMST, Elsevier ComNet, JNCA,
IEEE Trans. on Nanobioscience. Since 2019, she is Sci-
entific International Delegate for Inria Lille-Nord Europe.

Mohammed Ojaroudi received a PhD degree at Shahid

Beheshti University, Tehran, Iran, in 2014. He worked, as
a visiting researcher at University of Tennessee, Knoxville,
USA from 2013 to 2014. From Oct. 2014, he was a
post-doctoral researcher at Middle East technical Univer-
sity (METU), Ankara, Turkey and from 2016 to 2017, he
was an academic lecturer in Ankara University, Ankara, Anna Maria Vegni (Senior member, IEEE) is a tenure-
Turkey. From 2017-2020, he worked as a scientist re- track Assistant Professor in the Department of Engineer-
searcher at XLIM research laboratory Limoges, France. ing at Roma Tre University (Italy), since March 2020.
From 2020 he has been working on a project in the frame She received the Ph.D. degree in Biomedical Engineering,
of the reconfigurable intelligent meta-surface to develop Electromagnetics and Telecommunications from the De-
short-range sensing for THz communication and imaging partment of Applied Electronics, Roma Tre University, in
applications at Inria, Lille, France. From 2012, he has March 2010. She received the 1st and 2nd level Laurea
been a member of the IEEE-APS, IEEE-MTT, IEEE-AWPL, Degree cum laude in Electronics Engineering at Roma Tre
MAP-IET, reviewers’ group. He is an author and coau- University, in July 2004, and 2006, respectively. In 2009,
thor of more than 150 journals and international confer- she was a visiting researcher in the Multimedia Commu-
ences papers. His research interests include analysis and nication Laboratory, directed by Prof. Thomas D.C. Little,
design of microstrip antennas, design and modeling of at the Department of Electrical and Computer Engineer-
microwave structures, microwave imaging systems, and ing, Boston University, Boston, MA. Her research activity
emerging electromagnetic technologies for THz commu- focused on vehicular networking supported by heteroge-
nications. neous wireless networks and optical wireless communi-
cations. She is a member of ACM and an IEEE Senior Mem-
ber. In March 2018, she got the Italian Habilitation (Abili-
tazione Scientifica Nazionale) for Associate Professorship
in Telecommunication Engineering. She is involved in the
organization of several IEEE and ACM international con-
ferences and is a member of the editorial board of IEEE
Communications Magazine, Ad Hoc Networks, Journal of
Networks and Computer Applications, Nanocomnet Else-
vier journals, IEEE JCN, ITU J-FET and ETT Wiley journal.

Valeria Loscrí is a permanent researcher of the FUN

Team at Inria Lille-Nord Europe since Oct. 2013. From
Dec. 2006 to Sept. 2013, she was Research Fellow in
the TITAN Lab of the University of Calabria, Italy. She re-
ceived her MSc and PhD degrees in Computer Science in
2003 and 2007, respectively, from the University of Cal-
abria and her HDR (Habilitation a diriger des recherches)
in 2018 from Université de Lille (France). Her research
interests focus on emerging technologies for new commu-
nication paradigms such as Visible Light Communication
and TeraHertz bandwidth and cooperation and coexis-
tence of wireless heterogeneous devices. She has been in-
volved in the activity of several European Projects (H2020
CyberSANE, FP7 EU project VITAL, the FP6 EU project

88 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

A BLUEPRINT FOR EFFECTIVE PANDEMIC MITIGATION

Rahul Singh1 , Wenbo Ren2 , Fang Liu3 , Dong Xuan4 , Zhiqiang Lin5 , Ness B. Shroff6
1
Department of ECE, Indian Institute of Science, Bangalore, 2 CSE Department, The Ohio State University, 3 ECE
Department, The Ohio State University, 4 CSE Department, The Ohio State University, 5 CSE Department, The Ohio
State University, 6 ECE and CSE Departments, The Ohio State University,
NOTE: Corresponding author: Ness B. Shroff, [email protected]

Abstract – Traditional methods for mitigating pandemics employ a dual strategy of contact tracing plus testing
combined with quarantining and isolation. The contact tracing aspect is usually done via manual (human) contact
tracers, which are labor-intensive and expensive. In many large-scale pandemics (e.g., COVID-19), testing capacity is
resource limited, and current myopic testing strategies are resource wasteful. To address these challenges, in this work,
we provide a blueprint on how to contain the spread of a pandemic by leveraging wireless technologies and advances
in sequential learning for efficiently using testing resources in order to mitigate the spread of a large-scale pandemic.
We study how different wireless technologies could be leveraged to improve contact tracing and reduce the probabilities
of detection and false alarms. The idea is to integrate different streams of data in order to create a susceptibility graph
whose nodes correspond to an individual and whose links correspond to spreading probabilities. We then show how
to develop efficient sequential learning based algorithms in order to minimize the spread of the virus infection. In
particular, we show that current contact tracing plus testing strategies that are aimed at identifying (and testing)
individuals with the highest probability of infection are inefficient. Rather, we argue that in a resource constrained
testing environment, it is instead better to test those individuals whose expected impact on virus spread is the highest.
We rigorously formulate the resource constrained testing problem as a sequential learning problem and provide efficient
algorithms to solve it. We also provide numerical results that show the efficacy of our testing strategy.
Keywords – contact tracing, COVID-19, selective testing

1. INTRODUCTION approach, relying on a human being’s memory. Such

an approach cannot scale to large and rapidly moved
The outbreak of COVID-19 has unfolded an unprece- populations today. Meanwhile, manual tracing may
dented worldwide health, economical, and social crisis. result in delays, which could limit its utility. There-
Today, COVID-19 has spread to 188 countries, infected fore, recently numerous digital contact tracing systems
nearly 30 million people globally, and resulted in close to have been developed and deployed across the globe, by
one million deaths. The International Monetary Fund using a wide variety of sources to track “encounters”
(IMF) has predicted that the global economy will shrink including CCTV footage, records of credit card trans-
by 3% this year, the worst decline since the Great De- actions [1], locations measured using cellular networks
pression of the 1930s [2]. Today, millions of workers have or WiFi hotspots [26], locations via GPS, and crypto-
been laid off, and the tourism or hospitality industry has graphic tokens exchanged via Bluetooth Low Energy
been hurt particularly hard. (BLE) or acoustic channels [15]. For a recent survey
The COVID-19 outbreak and the mixed successes that of works on contact tracing and privacy-aware contact
nations have had in controlling the virus has under- tracing, see [9, 11, 10, 23, 24, 28, 7, 19]. Also see [14, 25]
scored the need for the development of technological for some interesting recent works in this area.
tools for pandemic mitigation. This paper provides a
blueprint on how technologies should be used in con- However, currently contact tracing system appears to be
junction with smart testing techniques in order to con- uncoordinated individual efforts focused on individual
tain a pandemic such as COVID-19. The first step for technologies and one type of data stream (e.g., camera
based systems, phone apps, etc.). Further, a number of
pandemic mitigation is to identify or trace the close con- these solutions suffer from a variety of technological lim-
tacts who might have been exposed to the disease from itations including a lack of coverage, privacy and secu-
a contagious individual. Contact tracing, an old tech- rity concerns, high missed detection and/or false alarm
nique, has been used as effective tools to battle pan- rates. For instance, increasingly Bluetooth-based con-
demics for many years, and some countries do use ag- tact tracing has gained mainstream use particularly with
gressive contact tracing to successfully contain COVID- Apple/Google’s support. However, our recent analy-
19. However, traditionally, contact tracing is a manual sis [29] with the released COVID-19 contact tracing apps

© International Telecommunication Union, 2020 89

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

shows that most Bluetooth-based contact tracing apps

use just the received signal strength indicator (RSSI )
of the Bluetooth for distance measurements. Unfor-
tunately, in practice, numerous factors can affect the
RSSI that can make the distance measurement inaccu-
rate, such as the power of the antenna used for broad- 1 Advertising
casting (i.e., the TxPower) and the obstacles blocking
transmission paths. Moreover, Bluetooth-based proxim- 2 Connection
ity tracing can also raise false positives because of the
potential misinterpretation of various scenarios. For ex-
ample, a proximity tracing system may interpret two 3 Ultrasound Signal
users have contact even if they are separated by a solid
wall, where the risk of infection is much lower than the 4 Ultrasound Signal

risk indicated by the measured distance.

5 Ultrasound Data
Therefore, we would like to propose an improved ap- Exchange

proach by combining as many data sources as possible

in an integrated way, with the key objective of mini- Disconnection
6

mizing false positives and false negatives in the contact

tracing and meanwhile protecting user’s privacy. The
Fig. 1 – A Simplified Protocol for Improved Data Exchange.
contact tracing data we can collect includes (1) multi-
ple channels including both Bluetooth and ultrasound 2. IMPROVED DATA COLLECTION
(using both microphones and speakers available in the
smartphone), and multiple sources including (2) WiFi There are two fundamental objectives that a good digital
and (3) cellular networks if they are available. We show contact tracing system must satisfy: (1) it should be ef-
how we can use improved methodology to collect data fective in tracking an individual (e.g., few false positives
that is privacy aware, transparent, and integrated in and missed detections), and (2) it should protect the pri-
Section 2. vacy of users. For example, a CCTV footage would be
highly effective if (1) were the sole objective, however it
Similarly, while testing followed by quarantining/iso- does not meet (2) since it is too much privacy-invasive.
lation is a powerful tool against a pandemic such as Therefore, we must look for effective and privacy-aware
COVID-19, testing capacity remains an issue, especially digital contact tracing techniques.
in hard hit areas where testing results could take multi-
ple days, even up to a week, to arrive. While traditional Since the outbreak of Covid-19, numerous techniques
approaches have focused on testing individuals who ex- based on Bluetooth, WiFi, and cellular networks have
hibit symptoms or have come in contact with other in- been developed for contact tracing. Each technique has
fected individuals, these approaches miss out many po- its own pros and cons. For instance, Bluetooth based
tential areas of outbreak where asymptomatic or pre- solutions can achieve reliable communication and a low
symptomatic super-spreaders seed the virus, which gets energy operation, but these suffer from a high rate of
detected only after it has already spread significantly. false positives due to a long communication range. WiFi
Thus, testing capacity needs to be used judiciously to based solutions do not require installation of apps on
prevent widespread outbreaks. In Section 4, we argue mobile phones, and they rely heavily on access point
that the current myopic approach to testing focuses on (AP) deployment, and its coverage. Therefore, in this
identifying individuals with the highest probability of paper, we aim to present an integrated approach that
being infected, which does not help minimize the over- improves the accuracy of a Bluetooth-based approach
all number of infected individuals. with additional channels, and combines WiFi and cel-
lular information if they are available. Furthermore by
Organization. The rest of this paper is organized as
utilizing clever algorithms that are provably optimal, it
follows: In Section 2, we describe an improved and inte-
aims to increase the efficiency with which infected indi-
grated methodology to collect contact tracing data. In
viduals are contained early without infecting too many
Section 3, we describe techniques that enable us to effi-
of their “neighbors”.
ciently integrate the data collected from various streams.
This allows us to reduce the “error probabilities” asso- Improving Bluetooth-based Contact Tracing
ciated with false alarms or missed detection of diseases, with Ultrasound Signals. Since Bluetooth-signals
and generate a dynamic susceptibility graph. In Sec- can penetrate obstacles such as solid walls, and also
tion 4, we address the practical problem of testing un- have a long transmission range, we would like to leverage
der constraints on resources. We perform simulations to other sensors in a smartphone to improve its proximity
show necessities of contact tracing and building a con- accuracy. In particular, we can use the inaudible ultra-
tact graph in Section 5, and conclude in Section 6. sound generated from the speaker and recorded by the

90 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

microphone for distance measurement. This feature can culate the distance by using Equation (1), where 𝐶 is
be used very easily. Ultrasound-based distance measure- the sound speed, and 𝐷𝐴,𝐴 or 𝐷𝐵,𝐵 is the distance from
ment is promising, and it has been shown to achieve a one phone’s speaker to its microphone respectively1 . Fi-
centimeter-level accuracy, e.g. BeepBeep [16]. It can nally, in Step », two devices disconnect Bluetooth when
help to improve the performance of mobile contact trac- the transmissions of other types of data for contact trac-
ing systems. According to the CDC guidelines [6], for ing finish.
the purpose of Covid transmission, it can be assumed Integrated Contact Tracing with WiFi.
that two individuals are in close contact (and hence
Besides the above mobile phone based contact tracing
likely to spread infection) if the distance between them
approaches, we can also use WiFi logs. Different from
is within 6 feet. Clearly, with an accuracy of a few
the above approaches, the WiFi based solution does not
centimeters, ultrasound is a very reliable and accurate
require app installation on mobile phones, and relies
technique for the purpose of mobile contact tracing.
upon widely deployed WiFi access points. It also does
Note that ultrasound cannot penetrate solid walls, and not require active involvement of mobile phones for ex-
hence it can also help us to rule out those “false con- changing the required information. The basic principle
tacts” that have been declared by the Bluetooth con- is described as follows: WiFi networks log all the associ-
nections occurring across walls.Using ultrasound for dis- ations and disconnections of devices connected to access
tance measurement requires a pair of devices to ex- points. We can analyze these WiFi logs to know where
change data between each other [16]. Fortunately, and when the devices (and hence their users) are close
Bluetooth-based mobile contact tracing systems have al- to each other. This provides information about the con-
ready implemented such an information exchange mech- tacts of device users.
anism, and hence developers can easily extend these sys- One of the advantages of such a WiFi-based solution is
tems in order to support ultrasound distance measure- that the WiFi log data is always available as long as
ment. This means that we can conveniently deploy both WiFi networks are active. Such networks allow both a)
ultrasound and Bluetooth technologies. reactive and b) proactive techniques for contact tracing.
An illustration of how we can integrate ultrasound and Let’s take the example of a university campus in order
Bluetooth for improved contact tracing is presented in to illustrate this. In reactive contact tracing, once a stu-
Fig. 1. The scheme is composed of six steps for the im- dent is confirmed to be infected, the university health
proved data exchange. In particular, when two smart- administration can use the WiFi MAC addresses of the
phones encounter, they first use Bluetooth to discover student’s mobile phone and his/her other computing de-
each other and initiate connections. At the beginning, vices such as laptop and Apple watch etc. to search
in Step ¶, one smartphone keeps advertising BLE pack- in the WiFi logs of campus networks, thereby deter-
ets to nearby devices. Meanwhile, another smartphone mining the locations where the student visited during a
scans its environment for any possible advertising pack- certain time period, and also his/her contacts at these
ets from its nearby devices, and initiates the connection locations. In proactive contact tracing, the university
in Step ·. After a connection has been established, two health administration can proactively analyze WiFi logs
devices emit ultrasound signals in turns (Step ¸ and ¹). to identify potential high risk users such as super spread-
When both devices have received signals from the other ers, and hot-spots (such as big gathering) in the cam-
device, they record the information that is required by pus. The university health administration can proac-
algorithms in order to measure distances, and then ex- tively pull WiFi logs and determine if the number of
change this information amongst each other via Blue- students in a gathering exceeds the limit that social dis-
tooth in Step º. For example, when using the algorithm tancing allows, and take appropriate measures.
from BeepBeep [16] for distance measurement, in Step The WiFi based contact tracing technique described
¸, alongside emitting an ultrasound signal, the phone above has its own limitations, for example the AP as-
(𝑃𝐴 ) will record the timestamp 𝑇𝐴1 when it senses the sociation logs can generate false alarms. In order to
signal sent by itself, and the other phone (𝑃𝐵 ) will also overcome these, we might consider using Received Sig-
store the timestamp 𝑇𝐵1 when it receives such a sig- nal Strength Indicator (RSSI) and Channel State Infor-
nal. Similarly, in Step ¹, phone 𝑃𝐵 records the times- mation (CSI) to reduce the errors. WiFi based solu-
tamp for signal emitting as 𝑇𝐵2 and phone 𝑃𝐴 stores the tion cannot be applied in areas without WiFi connec-
timestamp 𝑇𝐴2 when receiving signal. Next, in Step º, tivity. We should consider enhancing contact tracing by
phone 𝑃𝐴 has to send both 𝑇𝐴1 and 𝑇𝐴2 to phone 𝑃𝐵 , integrating multiple solutions such as Bluetooth, WiFi
and phone 𝑃𝐵 also needs to share its two timestamps and acoustic symbiotically, where one helps or replaces
(i.e., 𝑇𝐵1 and 𝑇𝐵2 ) with 𝑃𝐴 . the other depending on user preferences, environmental
dynamics, and resource availability. For example, in a
𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒(𝑃𝐴 , 𝑃𝐵 ) = WiFi-AP dense area such as a campus academic build-
𝐶 (1) ing, the WiFi-based solution can play a dominant role,
× ((𝑇𝐴2 − 𝑇𝐴1 ) − (𝑇𝐵2 − 𝑇𝐵1 )) + 𝐷𝐵,𝐵 + 𝐷𝐴,𝐴 while the mobile app running Bluetooth and acoustic
2
When each phone has these four timestamps, it can cal- 1 For more details, please refer to BeepBeep [16]

© International Telecommunication Union, 2020 91

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

based contact tracing can be automatically switched off A similar approached has been proposed and adopted
for better energy savings. Similarly, if the density of in ROBERT [3].
WiFi AP is less, then Bluetooth and acoustic based so-
Moreover, privacy concerns might arise from using the
lutions could be activated. They can help each other in
cell tower information for locating users because the
a crowded and occluded (e.g., by walls and other types
identifier of a user’s phone needs to be accessed. In order
of obstacles) environment for accurate and reliable con-
to mitigate these, we can also link such identifiers with
tact tracing. Note that Bluetooth gives relative loca-
pseudonyms. Similar approaches can also be applied
tion information whereas WiFi gives absolute location
in WiFi positioning. Therefore, the entry of data for
information. There might be a mismatch in the loca-
upload involves self pseudonym, encounter pseudonym,
tions as identified by WiFi data modality in conjunction
timestamp, Bluetooth proximity, ultrasound proximity.
with Bluetooth data. In order to resolve this issue, we
can model the location of the user using probabilistic After processing each upload data entry, the output of
techniques, and then use filtering techniques in order to this improved data collection procedure is data entries
derive a more accurate location. Readers interested in that involve pseudonyms of two encounters, the times-
more details about these techniques can refer to Kalman tamp, the adjusted proximity, and the infection risk. In
filtering and related topics in [17, 13]. particular, the adjusted proximity is the weighted av-
erage from combining proximity measured from differ-
Integrated Contact Tracing with Cellular Net-
ent sources (i.e., Bluetooth, ultrasound, WiFi, and cell
work. The location of a smartphone can also be iden- tower), and the infection risk can be obtained by using
tified from its communication with nearby cell towers. environment detecting heuristics. For example, when
Since a phone has to connect with cell towers in order to there is no proximity measurement from ultrasound and
send and receive data through cellular network, it con- the WiFi proximity indicates encounters are in different
stantly searches nearby cell towers and initiates connec-
rooms, the infection risk can be adjusted to a low level.
tions during movements. In each established connection,
Besides the above privacy-preserving data collection
a cell tower not only knows which phone is trying (each methods, we can also apply tools from the field of differ-
phone has a unique identifier) to connect at which time, ential privacy [8]. These utilize different kinds of data
but can also calculate the distance from itself to such a processing and noise injection methods, thereby making
phone (e.g., using the time elapsed between a ping com- it difficult for any party to determine whether or not a
mand, and the corresponding reply). As such, having
particular individual is in the original data records and
access to the locations of cell towers, as well as the dis-
providing privacy protection to the users. Such a guar-
tance of a phone from each of the involved towers, we antee on the privacy would encourage more users to join
can use a “triangulation” technique to pinpoint the lo- the system.
cation of a phone. However, in practice, such techniques
often can only locate a smartphone in an area instead
of an exact position.Moreover, using this technique for 3. DATA INTEGRATION AND SUS-
location tracking could raise privacy concerns, in that, CEPTIBILITY GRAPH
it requires access to the identifier of each phone that Here, the goal will be to create a “susceptibility graph”
may disclose user identify as well. Therefore, when only that describes compactly the different ways in which
having the corresponding permissions, cellular network disease is likely to spread. We begin by introducing
can be used for contact tracing, and meanwhile the user this graph, and then also describe how to construct this
identity must also need to be protected. graph by integrating the data from multiple sources.
How to collect the encounter records. Even though The graph would be time-variant.
there are distributed models for contact tracing which
allow each user to individually control whether or not 3.1 Graph Structure
to disclose its own encounter records, we advocate a A basic version of the graph would contain the following
centralized model in which each individual user’s con- components, and the designer is free to make reasonable
tact is collected by a central agency, and then stored modifications on it.
at a central backend. This is bound to raise pri-
vacy concerns, and hence we need to introduce privacy- • Nodes. Each node represents an individual that
preserving mechanisms.To this end, we will generate could be potentially infected. Individuals that are
pseudonyms for each user periodically and the linkage isolated will be removed from the graph. Also, we
between a pseudonym and the real user is only resolved can remove individuals who have recovered from
at the trust authority. The authority is only allowed the virus from the graph. However, since recovered
to link pseudonyms to real users when the pseudonyms individuals lose their antibodies for most viruses
belong to (𝑖) infected individuals that are confirmed by (including COVID-19), re-infections are possible af-
healthcare authorities or (𝑖𝑖) individuals who have close ter a period of time, so they would have to be re-
contact with infected ones. As such, the privacy of indi- introduced into the graph after some time. We use
viduals who have no risk of infection will be preserved. 𝒩 to denote the set of nodes (individuals).

92 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

• Node infection state. For each individual 𝑖 and needs to be investigated. If time is discretized, then
time 𝑡, we use 𝑋𝑖 (𝑡) to denote its infection state, the duration of a discrete time-slot could be any-
where 𝑋𝑖 (𝑡) = −1 means that this individual does where from several minutes to one day. Using time
not have the disease, 𝑋𝑖 (𝑡) = 0 denotes that it has slots can help reduce the storage and computation
the virus but cannot spread the virus, and 𝑋𝑖 (𝑡) = resources required.
1 denotes that it has the virus and is able to spread
the virus. For individual 𝑖 and time 𝑡, we use 𝑈𝑖 (𝑡) 3.2 Graph Construction
to denote its test results at time 𝑡. 𝑈𝑖 (𝑡) = 0 means
individual 𝑖 does not take a test at time 𝑡, 𝑈𝑖 (𝑡) = As discussed earlier, the graph 𝒢 consists of a set of
−1 means it is tested as negative at time 𝑡, and nodes and a set of edges, where each node holds a base
𝑈𝑖 (𝑡) = 1 means it is tested as positive at time 𝑡. infection probability and each edge holds a spreading
probability. The nodes, edges, and the probabilities all
• Edges. For every two nodes 𝑖 and 𝑗, If persons need to be deduced from the data, and the data can be
𝑖, 𝑗 have direct contact, then there is an undirected multi-sourced, for example wifi access logs of all users,
edge (𝑖, 𝑗) between them. We are free to choose the CCTV cameras, or Bluetooth scanning based contact
way in which we define “contact”: for example if tracing. More details on how to construct such a graph
these people are staying less than 6 feet apart for are as follows:
at least a certain duration of time co-occurrence in
• Individual identification. Identifying the indi-
a narrow space (e.g., a room and a bus), or par-
viduals and avoiding duplication are necessary for
ticipating in the same event. The contact informa-
the success of graph construction. How to do these
tion can be deduced by the techniques introduced
may depend on the data collection methods. For
in Section 2. We use ℰ to denote the set of undi-
instance, in the university WiFi logging system, an
rected edges and that (𝑖, 𝑗) is in ℰ means there is a
individual has and only has one access ID, and thus,
contact between 𝑖 and 𝑗.
this ID can be used to identify an individual. How-
• Base infection probabilities. Given the fact ever, in general WiFi systems, an individual may
that we cannot test every individual, each untested have multiple devices, and removing the duplication
individual has a base probability of being infected. is significant in this case. One method is restricting
This probability can be helpful for some tasks like the tracking to one type of device such as mobile
finding a suspected infected individual. For in- phones. When using the Bluetooth contact tracing,
stance, a person who contacted 500 people yester- we can use the IDs of the mobile phones to iden-
day could be more likely to be infected than some- tify the individuals, which is also applicable when
one who was in contact with a confirmed positive using Bluetooth contact tracing and WiFi logging
person; and we can use the base infection proba- simultaneously.
bilities to deduce this probability. The simulations • Edge detection. If there is a possible contact
in Section 5 also indicate that take the base infec- between two individuals, then an edge should be
tion probabilities into account can find and isolate generated to connect these two individuals. The
more infected people. The base infection probabil- contact can have multiple types. For instance, the
ity could be time-varying (e.g. abrupt changes due contact can be staying less than 6 feet, co-occurring
to certain events), and we use 𝑝𝑏 (𝑡) to denote the in the same room at some time period, or connected
base infection probability at time 𝑡. to the same access point during some time period.
This information can be deduced from the collected
• Spreading probabilities. In case two individu-
data.
als 𝑖, 𝑗 have been in contact, and one of them, say
user 𝑖, was a positive case, then there is a chance • Base infection probabilities. The base infec-
that individual 𝑗 got infected by the contact. This tion probability can be deduced from the positive
chance may also be time-variant. We let the spread- rate per test or the number of confirmed posi-
ing probability be denoted as 𝑝𝑖→𝑗,𝑠 (𝑡), which is the tive cases per randomly tested individuals. For in-
probability that 𝑗 got infection from a contact with stance, a university randomly tested 1,000 students
𝑖. The calculation of the probability will be dis- and found 20 positive cases, then we can assume
cussed later. that each student of the university has 2% proba-
bility to be positive. If we do not have this infor-
• Time. The time can either be continuous or dis- mation, we can use the number of newly detected
crete. Continuous time better fits the reality, but infections in a period with a multiplier as the esti-
such an assumption also needs more storage and mate.
computation power to process the graph. Besides,
given the fact that there are delays, or the occur- • Spreading probabilities (link probabilities).
rence time of events or contacts are not known pre- Deducing the spreading probabilities is a relatively
cisely, how to construct an accurate timely graph harder task, which can be divided into two steps.

© International Telecommunication Union, 2020 93

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

The first step is to infer the type or level of con- the pandemic. Note that this objective is quite differ-
tacts between two individuals. Have they stayed ent from focusing on testing individuals with the highest
closer than 6 feet or stayed in the same room for a probability of infection, which is what current systems
while? The second step is to deduce the link proba- try to do. Rather our focus must be on testing individu-
bility. Accurate characterizations of the link prob- als that have the highest expected impact on viral spread.
abilities could come from exposure data studies to Consider the following example as an illustration.
the virus. However, a reasonable model would be Example: Assume that two individuals 𝑖 and 𝑗 are in-
to use a concave function of time to estimate the fected with probabilities 0.1 and 0.3, respectively. How-
link probability. ever, assume that the expected number of individuals
that 𝑖 encounters is 50 times larger than the expected
• Testing results. For an individual, if this individ- number of individuals that 𝑗 comes in contact with. In
ual has taken a virus test and got the result, then this case, it makes more sense to prioritize testing indi-
we know whether this individual has the disease or vidual 𝑖 over individual 𝑗. This is another reason why
not (with a certain confidence). we should test healthcare workers more often, because
of their frequent contact with a large number of individ-
3.3 Data Integration uals. Based on this key insight, our goal will be to:
In real-world scenarios, there are multiple data sources.
For example, different contact tracking data sources as • Develop learning based approaches that result in
described earlier (Bluetooth or ultrasound contact trac- smart testing capabilities which balance the ex-
ing data, WiFi logs, GPS, etc.) could be integrated to ploration and exploitation subject to testing con-
greatly improve the quality of contact tracing. The in- straints. Isolate individuals who have been tested
tegration could be done by using filtering techniques, positive and quarantine their contacts.
in which we compute the probability of an edge condi-
tioned on the (multi-source) information available to us. • Our model will also incorporate practical issues
We would typically rely upon generative models of the such as inaccurate estimates, testing errors, pool
data in order to compute these conditional probabilities. testing, etc.
With multiple data sources, we need to deal with incon-
sistent data. For instance, Bluetooth gives relative loca- • Develop efficient rules of thumb that can be eas-
tion information whereas WiFi gives absolute location ily implemented in practice. This could mean test-
information, and the information of two sources may ing asymptomatic individuals who have not encoun-
be inconsistent. We can deal with this issue by assum- tered a confirmed infected person, but have made
ing that the data collections of the sources are random a large number of contacts.
and independent and assign probability distributions to
them. Probabilistic description allows “soft recovery” 4.1 Suspicious Infection Inference
of data after we use filtering algorithms. Bayesian up-
dates can be used to merge or pool information from One significant task is to find the most likely infected
various source. We can use Kalman filtering or some individuals from the partial observations, i.e., the test
other filtering algorithm. results of some individuals. To do this, one way is to
interpret the probability that a person is infected given
Such an integration can yield us the following kind of
the partial observations, such as (“noisy”) contact graph
improvements:
or test results of a few individuals from the graph etc.
• Reduced inaccuracies and better estimates of the These algorithms could be based upon the susceptibil-
link probabilities. Consider for example the case ity graph constructed by using the methods stated in
when people could have social contact by virtue of Section 3.
being located in a crowded facility such as students
in the same classroom or people in the same flight. 4.1.1 Partial Observed Markov Decision Pro-
However, data sources such as building information, cess (POMDP)
WIFI access might be noisy. In this case, one could
combine GPS data (collected from probably smart- We formulate the problem of sequential testing for
phone usage) in order to yield an accurate estimate COVID-19 as a Markov Decision Process (MDP). Pop-
of social contacts. ulation is composed of 𝑁 individuals, and the state
evolves at discrete times 𝑡 ∈ [1, 𝑇 ]. Let 𝑋𝑖 (𝑡) ∈ {0, 1}
4. TESTING UNDER RESOURCE denote the hidden state of individual 𝑖 at 𝑡, where
𝑋𝑖 (𝑡) = 0 means that 𝑖 is free of disease at 𝑡 and
CONSTRAINTS
𝑋𝑖 (𝑡) = 1 indicates that 𝑖 is infected. We use the vector
The goal here is to leverage the information contained in 𝑋(𝑡) ∶= (𝑋1 (𝑡), 𝑋2 (𝑡), … , 𝑋𝑁 (𝑡)) ∈ {0, 1}𝑁 to represent
the susceptibility graph in order to sequentially choose the state of the entire system. Let 𝒳 ∶= {0, 1}𝑁 de-
individuals for testing so as to minimize the spread of note the state-space of the network. Note that the state

94 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

vector 𝑋(𝑡) is never fully revealed to the learner2 . crucial because not many testing kits are available dur-
Test and Quarantine: At each time 𝑡 ∈ [1, 𝑇 ], the ing epidemics. An alternative, somewhat equivalent and
learner has a unit budget to choose an individual 𝑖 ∈ simpler objective is to remove the capacity constraints
[1, 𝑁 ] in order to “sample” (test for infection). Sam- altogether and include a cost for using testing kits,
pling an individual 𝑖 at 𝑡 reveals the state 𝑋𝑖 (𝑡). We
𝑇
let 𝑈 (𝑡) ∈ [0, 𝑁 ] denote the sampling decision at time
min 𝔼𝜋 (∑ ‖𝑋(𝑡)‖1 + 𝜆1 (𝑈 (𝑡) ≠ 0)) , (7)
𝑡. In case no one is sampled at 𝑡, we let 𝑈 (𝑡) = 0. We 𝜋
𝑡=1
let 𝑌𝑖 (𝑡) denote the test result at time 𝑡; 𝑌𝑖 (𝑡) = +1
means the person tested positive, 𝑌𝑖 (𝑡) = −1 means the where 𝜆 > 0. We now briefly discuss how to solve the
test was negative, and 𝑌𝑖 (𝑡) = 0 means that the individ- above discussed MDP. Theoretical results on the ex-
ual was not tested at time 𝑡. The vector comprising of istence of optimal policies, and methods to solve con-
observations 𝑌𝑖 (𝑡) is denoted 𝑌 (𝑡). strained MDPs, or POMDPs can be found in [4, 21, 22].
If sampled individuals are found to be infected, then In case the parameters describing the environment are
they are isolated, i.e., kept out of the population, and unknown, we can use machine learning techniques de-
hence cannot spread the disease to their neighbors. We veloped in [20].
let 𝑄(𝑡) denote the set of those individuals who are iso- Belief State MDP: We introduce a belief state, which is
lated at time 𝑡. a posterior distribution over the state space 𝒳. This
State Transition: Let us now look at the controlled transforms the POMDP to a continuous-state MDP
transition probabilities of the controlled Markov process on the belief state. We denote the belief state by
𝑋(𝑡). We first introduce some notations. For 𝑥, 𝑦 ∈ 𝒳, ℐ(𝑡) = {ℐ(𝑡, 𝑥)}𝑥∈𝒳 , where ℐ(𝑡, 𝑥) ∶= ℙ (𝑋(𝑡) = 𝑥|ℱ𝑡 ).
define By Bayes’ Rule, the terms ℐ𝑡 (𝑥) are computed recur-
sively as
𝑁
Δ1 (𝑥, 𝑦) = 1 {∑ |𝑥𝑖 − 𝑦𝑖 | = 1} and (2) ℐ𝑡+1 (𝑥) = ∑ ℐ𝑡 (𝑦)ℙ (𝑌𝑈(𝑡) |𝑋(𝑡) = 𝑦) 𝑃𝑡 (𝑦, 𝑥), (8)
𝑖=1
𝑦∈𝒳
𝑖 if 𝑥𝑖 ≠ 𝑦𝑖 and Δ1 (𝑥, 𝑦) = 1,
Δ2 (𝑥, 𝑦) = { (3) where the state transition probabilities 𝑃𝑡 (𝑦, 𝑥) are as
∅ otherwise . discussed in (4).
Optimal Policy: The optimal sampling policy can be
Clearly, Δ1 (𝑦, 𝑥) assumes value 1 only if 𝑥 and 𝑦 differ obtained by solving the following set of non-linear Dy-
in a single position; since disease can spread to only one namic Programming equations [12],
more person during two consecutive times, this function
is 0 if 𝑥 cannot evolve to 𝑦 in one single time-step. Δ2 𝑉𝑡 (ℐ𝑡 ) = ∑ ‖𝑥‖1 ℐ𝑡 (𝑥)
provides us with the node that “transitioned” to dis- 𝑥∈𝒳
eased state when the system evolved in a unit step from + min (𝔼 𝑉𝑡+1 (ℐ𝑡+1 ) + 𝜆1{𝑢 ≠ 0}) , (9)
𝑥 to 𝑦. Thus, the single-step transition probabilities are 𝑢∈[0,𝑁]

given as 𝑉𝑇 (ℐ) = ∑ ‖𝑥‖1 ℐ(𝑥), ∀ℐ ∈ Δ(𝒳), (10)

𝑥∈𝒳
𝑃𝑡 (𝑥, 𝑦) = Δ1 (𝑥, 𝑦)𝑝 ∑ 𝑤𝑡′ (𝑖, Δ2 (𝑦, 𝑥)). (4)
where Δ(𝒳) denotes simplex on 𝒳 and ℐ𝑡 denotes rep-
𝑖∈𝒱′𝑡
resentative belief state at time 𝑡. Optimal sampling ac-
Objective: Let ℱ𝑡 ∶= ∪𝑡𝑠=1 (𝑈 (𝑠), 𝑌 (𝑠), ℓ(𝑠)) be the ob- tion at time 𝑡 in state ℐ𝑡 corresponds to a minimizer
servation history of the learner [18]. Then, the policy of the r.h.s. in the above equation. However, equa-
𝜋 is a sampling decision at 𝑡 on the basis of ℱ𝑡−1 , i.e., tions (9), (10) are computationally intractable because
𝜋 ∶ ℱ𝑡−1 ↦ 𝑈 (𝑡), 𝑡 ∈ [1, 𝑇 ]. Our goal is to find a policy the number of required computations is 𝑂(2𝑁 ). Thus,
that solves the following problem, we propose tractable provably approximate solutions
next.
𝑇
min 𝔼𝜋 (∑ ‖𝑋(𝑡)‖1 ) , (5) 4.1.2 Hidden Markov Model
𝜋
𝑡=1
𝑇 Since we might not observe the susceptibility graph 𝒢,
s.t. 𝔼𝜋 (∑ 1 (𝑈 (𝑡) ≠ 0)) ≤ 𝐶, (6) we can model it as a hidden Markov process [17]. As-
𝑡=1 sume that the system evolves at discrete time-instants
𝑡 = 1, 2, … , 𝑇. Each slot could be of a duration equal
where ‖⋅‖1 denotes the 𝐿1 norm and 𝐶 is the total testing to half hour, one hour, or one day. For each time slot
capacity. The instantaneous cost ‖𝑋(𝑡)‖1 encourages the 𝑡, we use 𝑋𝑖 (𝑡) to denote the infection state of the 𝑖-th
policy to keep the infections as low as possible in an as individual. In this section we slightly tweak the binary-
early as possible manner. The capacity constraints are valued state model that was described earlier, and allow
2 So this problem is a partially observable MDP (POMDP), which it to assume 4 values. This allows us to design an algo-
is non-trivial to solve in general case. rithm that is more accurate. Thus, we let 𝑋𝑖 (𝑡) = −1 if

© International Telecommunication Union, 2020 95

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

individual 𝑖 is not infected by the virus at time 𝑡, and 𝑖=1 as follows,

{𝑈𝑖 (𝑡)}𝑁
𝑋𝑖 (𝑡) = 0 if individual 𝑖 is infected by the virus but can-
not spread the virus, i.e., is in incubation, 𝑋𝑖 (𝑡) = +1 ℙ{𝑈𝑖 (𝑡) = 1 ∣ ℱ𝑡 }
if individual 𝑖 is infected by the virus and can spread = 1𝑈𝑖 (𝑡)≠0 ( ∑ 1𝑋𝑖 (𝑡)=𝑠 ⋅ 𝑝𝑠→1 ), (15)
the virus, and finally 𝑋𝑖 (𝑡) = −2 if they have recovered 𝑠=−1,0,1
from the virus or have been isolated already. We call
ℙ{𝑈𝑖 (𝑡) = 1 ∣ ℱ𝑡 }
𝑋𝑖 (𝑡) = 0 “inactive” and 𝑋𝑖 (𝑡) = 1 active.
The rest of the discussion in this section makes the fol- = 1𝑈𝑖 (𝑡)≠0 ( ∑ 1𝑋𝑖 (𝑡)=𝑠 ⋅ 𝑝𝑠→−1 ), (16)
lowing simplifying assumptions: 𝑠=−1,0,1

where whether 𝑈𝑖 (𝑡) = 0 or not is determined by the

• States 𝑋𝑖 (𝑡) and decisions 𝑈𝑖 (𝑡) do not change tracing or testing algorithms.
within a slot. Clearly, 𝑋(𝑡) = {𝑋𝑖 (𝑡)}𝑖∈𝒩 is a Markov process, and if
we are provided with the values of 𝑈𝑖 (0), 𝑈𝑖 (1), 𝑈𝑖 (2), …,
• Upon becoming an active spreader, an individual then our goal is to find the most likely values of 𝑋(𝑡).
can spread the disease only after the current time- In order to do this, we might use Markov chain Monte
slot ends. This might seem to be restrictive, but is Carlo (MCMC) algorithms such as Gibbs Sampling.
justifiable since our modeling procedure already in- Readers may refer to [27] for a review of MCMC al-
troduces “noise” due to erroneous tracing and test- gorithms.
ing.
4.1.3 Graph Embedding
• The spreading probability, denoted as 𝑝𝑠 , is a con-
stant that is independent of other parameters such Computing the infection probabilities of individuals di-
as the values of the states, the number of days one rectly will be computationally cumbersome, and we can
has been infected, etc. utilize graph embedding [5] techniques in order to find
suspicious infected individuals. These techniques map
• After becoming an inactive infected person, at each the nodes of a graph to points in ℝ𝑑 , where 𝑑 is a natural
time slot the individual becomes active with a prob- number. If the graph embedding algorithm is properly
ability equal to 𝑝0,1 . chosen, then if two points are close in the space ℝ𝑑 then
they are also close in the susceptibility graph, so that the
probability that the virus spreads from one individual to
• For an active infected person, for every time slot, the other is high. Note that in the graph each node may
this individual has a constant probability to get re- have up to |𝒩| edges, but in the embedded graph, each
moved. We use 𝑝1,−2 to denote this probability. node only has 𝑑 coordinates. Since the number of edges
might be much more than 𝑑, performing computations
• For an individual at state 𝑋𝑖 (𝑡) = 𝑎, it has a with the embedded coordinates is much more efficient
constant probability 𝑝𝑎→𝑏 to be tested to be state than directly working with the original graph.
𝑈𝑖 (𝑡) = 𝑏.
5. SIMULATIONS
Let ℱ𝑡 be the filtration generated by (𝑋𝑖 (𝑠), 𝑌𝑖 (𝑠), 𝑖 ∈
𝒩, 𝑠 ≤ 𝑡). With the above assumptions in place, we can In this section, we use a simulation to indicate the ne-
write the “dynamics” or transition probabilities govern- cessities of contact tracing and building a contact graph.
ing 𝑋(𝑡) = {𝑋𝑖 (𝑡))}𝑁
𝑖=1 as follows,
Contact tracing is an essential technique for finding po-
tential infectious people. A commonly employed naive
ℙ{𝑋𝑖 (𝑡 + 1) = 2 ∣ ℱ𝑡 } contact tracing technique is to trace and test only those
who have had contact with a confirmed positive person.
= 1𝑋𝑖 (𝑡)=−2 + 1𝑋𝑖 (𝑡)=1 ⋅ 𝑝1,−2 , (11)
We call this simple and intuitively appealing contact
ℙ{𝑋𝑖 (𝑡 + 1) = 1 ∣ ℱ𝑡 } tracing policy as Policy 1. However, this method may
= 1𝑋𝑖 (𝑡)=1 (1 − 𝑝1,−2 ) + 1𝑋𝑖 (𝑡)=0 ⋅ 𝑝0,1 , (12) not be optimal, especially under circumstances when a
sizeable proportion of the population is infected. To
ℙ{𝑋𝑖 (𝑡 + 1) = 0 ∣ ℱ𝑡 }
see why this might be the case, consider the scenario
= 1𝑋𝑖 (𝑡)=0 ⋅ (1 − 𝑝0,1 ) when two people are waiting to get tested. The first
person had a close contact with a confirmed infected
+ 1𝑋𝑖 (𝑡)=−1 (1 − ∏ (1 − 𝑝𝑠 (𝑡))), (13)
𝑗∶(𝑖,𝑗)∈ℰ,𝑋𝑗 (𝑡)=1
person, while the second person did not have any such
close contact with a confirmed infected person; but did
ℙ{𝑋𝑖 (𝑡 + 1) = −1 ∣ ℱ𝑡 } closely contact 500+ untested people (for example this
= 1 − ℙ{𝑋𝑖 (𝑡 + 1) = 1, 0, or − 2 ∣ ℱ𝑡 }, (14) person works in a supermarket). Policy 1 will suggest to
us to test the first person; however, when a significant
and the dynamics or transition probabilities of 𝑈 (𝑡) = proportion of the population (e.g., 1%) are positive, in

96 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

expectation, the second person would have been in close • During any particular day, any two people in the
contact with more than one infected people. Thus, tak- city meet each other with a probability of 0.01.
ing the number of contacts with all people into con- Thus, on average, a person meets around 10 people
sideration could significantly improve the tracing. The per day in expectation. When two people meet, and
simulation in this section also confirms our claim. one of them is infected while the other is not, the
These two policies can be mathematically described as virus will spread with a probability of 3/(14 × 10);
follows. hence an infected person spreads the virus to an
average of 𝑅 = 3 people before being removed.
• Policy 1: Fix a time frame of a duration say 2 weeks,
and call the time duration composed of the previous • Each day the community chooses 20 people to quar-
two weeks as the “tracing window”. At any given antine by using its policy. If quarantined persons
time we only take into account those contacts that are found to be infected, then they will be isolated
occurred during the tracing window. Let 𝑝𝑠 be the until they are removed. Otherwise, they will be
probability that the virus spreads from an infected quarantined for 14 days, and then will be back to
person to a healthy person during a contact. We the normal schedule.
assume that 𝑝𝑠 is constant and known. For any per-
• We assume that the community as a whole knows
son 𝑠, given that person 𝑠 contacted 𝑚 confirmed
all the contacts between all of its people, and when-
infected persons during the tracing window, we use
ever a person is removed the community gets to
ℙ ({𝑠 got infected}) = 1 − (1 − 𝑝𝑠 )𝑚 (17) know this information at the beginning of the next
day. Also, we assume that the spreading probabil-
to measure the risk that person 𝑠 is infected. We ity 𝑝𝑠 is known to the community. We note that,
then choose to test those persons who have the with the knowledge of 𝑝𝑠 and assuming the value of
highest probabilities of being infected. 𝑝𝑏 , the community is able to compute Eq. (17) and
Eq. (18).
• Policy 2: It additionally utilizes the contact graph,
and checks the number of contacts of each person We perform simulations for 150 consecutive days, and
𝑠 ∈ 𝒮. Hence, if 𝑛 is the number of contacts of 𝑠 in record the cumulative infections in the population for
the tracing window, we let the following 5 policies and parameters:
• No contact tracing of any sort is utilized.
ℙ({𝑠 got infected}) = 1 − (1 − 𝑝𝑠 )𝑚 (1 − 𝑝𝑏 𝑝𝑠 )𝑛
(18) • Policy 1 (Eq. (17)).

in order to measure the risk that person 𝑠 is in- • Policy 2 with 𝑝𝑏 = 0.02, where 0.02 is a well tuned
fected. Over here, 𝑝𝑏 is the so-called base infection value.
probability, which can either be a constant or de-
pend on the proportion of confirmed cases of the • Policy 2 with 𝑝𝑏 = 0.2, where 0.2 is an example of
population (i.e., adaptive). Note that we are as- a not well tuned value of 𝑝𝑏 .
suming that the infection status of these 𝑛 people • Policy 2 with adaptive 𝑝𝑏 = 𝑁rr /1000, where “rr”
are unknown. denotes “recently removed” and 𝑁rr means the
number of people removed in the tracing window
Our simulations results are depicted in Fig. 2, and
(i.e., the last two weeks).
clearly show the superior performance of Policy 2 as
compared to that of Policy 1. More details on the sim- Our simulation results are summarized in Fig. 2. We
ulation setup are as follows: explicitly state the number of total infections in Table 1.
• The population size is 1000 people, and a single tracing policy parameter 𝑝𝑏 total infections
person (that is chosen uniformly at random from no tracing — 987
the population) is infected by the virus at day zero. Policy 1 — 617
Policy 2 0.02 540
• Regarding the transmission capability of the virus, Policy 2 0.2 669
we assume that a person will be able to spread the Policy 2 adaptive 569
virus 1 day after getting infected. Moreover, a per-
son remains infected for at least 7 days. After this Table 1 – Total number of infections of the virus under different
duration, on each day the person will change his tracing policies.
state (to either isolated due to its symptoms, recov- We summarize our findings as follows.
ered, or deceased) with a probability of 1/7. Thus,
in expectation, the virus lasts for 14 days. A per- • Contact tracing and quarantine facilities are essen-
son whose state has changed to removed, will not tial in order to control the spread of virus. Without
spread the virus or get infected. these, the total number of infections are around 987,

© International Telecommunication Union, 2020 97

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

of detection and false alarms. The idea is to use possibly

disparate wireless data streams for data collection, then
1000

900
integrate this data to improve coverage, reduce probabil-
800 ities of errors and false alarms and create a susceptibility
num of cumulative infections

700 graph that could be used for intelligent testing. Based

600 on this susceptibility graph, we show how to develop
500
efficient sequential learning based algorithms in order
to minimize the spread of the virus infection. Another
contribution is that we develop provably optimal algo-
400

300
no tracing rithmic solutions that rely upon the theory of partially
observable Markov decision processes. In particular, we
policy 1
200
policy 2 with p b = 0.02

100 policy 2 with p b = 0.2 show that current contact tracing plus testing strategies
policy 2 with adaptive p b
that are aimed at identifying (and testing) individuals
0 50 100 150
with the highest probability of infection are inefficient.
Instead, we find that it is better to test those individuals
days

Fig. 2 – Simulation of the spreading of the virus under different whose expected impact on virus spread is the highest.
tracing policies. We formulate the testing problem as a Partially Observ-
which is approximately the entire population. Even able Markov Decision Process whose goal is to minimize
a simple contact tracing technique such as Policy 1, the expected spread of the virus subject to testing ca-
reduces the total infections to 617 (which is a 37% pacity constraints. We provide efficient algorithmic so-
reduction). Thus, it is worth the effort to establish lutions to this problem and show via numerical results
a system that can track the contacts of the people that our solution substantially reduces the spread of the
in order to better detect the infections. virus.

• Compared to Policy 1 that only considers the con- ACKNOWLEDGEMENT

tacts with confirmed infected people, Policy 2 also
takes contacts with untested (but probable to be This work was supported in part by NSF grants CNS-
infected) people into account, and hence it has a 1618520 and CNS-2028547.
better performance. This is clearly demonstrated
in its superior performance in all of the three ex- REFERENCES
periments, for example when 𝑝𝑏 = 0.02, we get a [1] Covid-19 contact tracing: a briefing. https://ww
gain of around 12%. w.bmj.com/content/369/bmj.m1859. (Accessed
• When 𝑝𝑏 is tuned properly (e.g., 𝑝𝑏 = 0.02), Pol- on 07/02/2020).
icy 2 performs better. However, the tuning effort
[2] The global economy is expected to shrink by 3%
is substantial, and might be deemed infeasible in
this year. https://www.economist.com/graphi
practice. For instance, when 𝑝𝑏 = 0.2, the perfor-
c-detail/2020/04/14/the-global-economy-is-
mance of Policy 2 is worse than Policy 1. Thus,
expected-to-shrink-by-3-this-year.
Policy 2 with adaptive values of 𝑝𝑏 is a good option
in practice. [3] Robert – robust and privacy-preserving proximity
• If we are to use only contact tracing and quarantine tracing protocol. https://github.com/ROBERT-
facilities, our performance is not very good. Even if proximity-tracing. (Accessed on 09/17/2020).
tracing is possible for 2% of the population per day, [4] Eitan Altman. Constrained Markov decision pro-
the majority of the population will get infected af- cesses, volume 7. CRC Press, 1999.
ter a few months. Hence, it is necessary to combine
contact tracing and quarantines with other policies, [5] Hongyun Cai, Vincent W Zheng, and Kevin Chen-
e.g., avoiding contacts to reduce the number of con- Chuan Chang. A comprehensive survey of graph
tacts, and also wearing masks to reduce the virus embedding: Problems, techniques, and applica-
spreading probability. tions. IEEE Transactions on Knowledge and Data
Engineering, 30(9):1616–1637, 2018.
6. CONCLUSION
[6] CDC. How to protect yourself & others. https:
In this paper, we have provided a detailed blueprint on //www.cdc.gov/coronavirus/2019-ncov/preven
how to contain the spread of a pandemic by integrating t-getting-sick/prevention.html.
the use of various wireless technologies with sequential
learning based techniques. In particular, we show how [7] Aaqib Bashir Dar, Auqib Hamid Lone, Saniya Za-
different wireless technologies could be leveraged to im- hoor, Afshan Amin Khan, and Roohie Naaz. Ap-
prove contact tracing efforts and reduce the probabilities plicability of mobile contact tracing in fighting pan-

98 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

demic (covid-19): Issues, challenges and solutions. [19] Viktoriia Shubina, Sylvia Holcer, Michael Gould,
Computer Science Review, page 100307, 2020. and Elena Simona Lohan. Survey of decentral-
ized solutions with mobile devices for user location
[8] Cynthia Dwork, Aaron Roth, et al. The algorith- tracking, proximity detection, and contact tracing
mic foundations of differential privacy. Foundations in the covid-19 era. Data, 5(4):87, 2020.
and Trends in Theoretical Computer Science, 9(3-
4):211–407, 2014. [20] Rahul Singh, Abhishek Gupta, and Ness B Shroff.
Learning in Markov decision processes under con-
[9] Luca Ferretti, Chris Wymant, Michelle Kendall, straints. arXiv preprint arXiv:2002.12435, 2020.
Lele Zhao, Anel Nurtay, Lucie Abeler-Dörner,
Michael Parker, David Bonsall, and Christophe [21] Richard D Smallwood and Edward J Sondik. The
Fraser. Quantifying sars-cov-2 transmission sug- optimal control of partially observable Markov pro-
gests epidemic control with digital contact tracing. cesses over a finite horizon. Operations research,
Science, 368(6491), 2020. 21(5):1071–1088, 1973.
[22] Edward J Sondik. The optimal control of par-
[10] Stephen M Kissler, Christine Tedijanto, Edward
tially observable markov processes over the infinite
Goldstein, Yonatan H Grad, and Marc Lipsitch.
horizon: Discounted costs. Operations research,
Projecting the transmission dynamics of sars-cov-
26(2):282–304, 1978.
2 through the postpandemic period. Science,
368(6493):860–868, 2020. [23] Qiang Tang. Another look at privacy-preserving
automated contact tracing. arXiv preprint
[11] Mirjam E Kretzschmar, Ganna Rozhnova, Mar- arXiv:2010.13462, 2020.
tin CJ Bootsma, Michiel van Boven, Janneke HHM
van de Wijgert, and Marc JM Bonten. Impact of [24] Qiang Tang. Privacy-preserving contact trac-
delays on effectiveness of contact tracing strategies ing: current solutions and open questions. arXiv
for covid-19: a modelling study. The Lancet Public preprint arXiv:2004.06818, 2020.
Health, 5(8):e452–e459, 2020.
[25] Amee Trivedi and Deepak Vasisht. Digital contact
[12] Vikram Krishnamurthy. Partially Observed Markov tracing: technologies, shortcomings, and the path
Decision Processes. Cambridge University Press, forward. ACM SIGCOMM Computer Communica-
2016. tion Review, 50(4):75–81, 2020.

[13] Panqanamala Ramana Kumar and Pravin Varaiya. [26] Amee Trivedi, Camellia Zakaria, Rajesh Balan, and
Stochastic systems: Estimation, identification, and Prashant Shenoy. Wifitrace: Network-based con-
adaptive control. SIAM, 2015. tact tracing for infectious diseases using passive wifi
sensing, 2020.
[14] Douglas J Leith and Stephen Farrell. Coronavirus
contact tracing: Evaluating the potential of using [27] Don Van Ravenzwaaij, Pete Cassey, and Scott D
Bluetooth received signal strength for proximity de- Brown. A simple introduction to markov chain
tection. 2020. monte–carlo sampling. Psychonomic bulletin & re-
view, 25(1):143–154, 2018.
[15] Yuxiang Luo, Cheng Zhang, Yunqi Zhang,
[28] Haohuang Wen, Qingchuan Zhao, Zhiqiang Lin,
Chaoshun Zuo, Dong Xuan, Zhiqiang Lin, Adam C.
Dong Xuan, and Ness Shroff. A study of the privacy
Champion, and Ness Shroff. Acoustic-turf:
of covid-19 contact tracing apps. In International
Acoustic-based privacy-preserving covid-19 contact
Conference on Security and Privacy in Communi-
tracing, 2020.
cation Networks, 2020.
[16] Chunyi Peng, Guobin Shen, Yongguang Zhang, [29] Qingchuan Zhao, Haohuang Wen, Zhiqiang Lin,
Yanlin Li, and Kun Tan. Beepbeep: a high ac- Dong Xuan, and Ness Shroff. On the accuracy
curacy acoustic ranging system using cots mobile of measured proximity of Bluetooth-based contact
devices. In Proceedings of the 5th international tracing apps. In International Conference on Se-
conference on Embedded networked sensor systems, curity and Privacy in Communication Networks,
pages 1–14, 2007. 2020.
[17] Martin L Puterman. Markov decision processes:
discrete stochastic dynamic programming. John Wi-
ley & Sons, 2014.

[18] Albert N Shiryaev. Optimal stopping rules, vol-

ume 8. Springer Science & Business Media, 2007.

© International Telecommunication Union, 2020 99

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

AUTHORS machine learning conferences, such as ICML, NeurIPS,

ICLR, and AAAI.
Rahul Singh received his
B. Tech. degree in Electri-
cal Engineering from Indian
Institute of Technology, Kan- Dong Xuan received his B.S.
pur, India, in 2009, M.Sc. and M.S. degrees in Electronic
degree in Electrical Engineer- Engineering from Shanghai
ing from University of Notre Jiao Tong University, China,
Dame, South Bend, IN, in and his Ph.D. degree in Com-
2011, and his Ph.D. degree in puter Engineering from Texas
Electrical and Computer Engi- A&M University. In 2001, he
neering from the Department joined the Department of Com-
of Electrical and Computer puter Science and Engineering
Engineering Texas A&M Uni- at The Ohio State University, where he is currently
versity, College Station, TX, in 2015. Currently he is a full professor. He was on the faculty of Electronic
an Assistant Professor at the Department of Electrical Engineering at Shanghai Jiao Tong University from
Communication Engineering, the Indian Institute 1993 to 1998 and a visiting scholar in Computer Science
of Science. His research interests include stochastic at City University of Hong Kong from 1997 to 1998.
control, machine learning, applied probability, networks He was a research assistant/associate in the Real-Time
and large-scale complex cyber physical systems. He Systems group in the Department of Computer Science
has earlier worked as a Postdoctoral Researcher at the at Texas A&M University from 1998 to 2001. His
Laboratory for Information Decision Systems (LIDS), research interests include computer networking and
Massachusetts Institute of Technology, at Intel, Santa mobile systems. Dr. Xuan has served as an editor for
Clara as a Machine Learning Engineer, and as a Data IEEE TPDS and ACM ToSN, and has been a TPC
Scientist at Encored Technologies. member for a number of IEEE and ACM flagship
conferences such as IEEE INFOCOM, ICDCS, ICNP,
and ACM MobiHoc. He is a recipient of the National
Wenbo Ren received his Science Foundation CAREER Award and the College
B.S. degree in Electronic In- of Engineering Lumley Research Award at The Ohio
formation Engineering from State University.
University of Science and
Technology of China in 2016.
Since then, he has been a
Ph.D. student in Computer Zhiqiang Lin received his
Science and Engineering at Ph.D. degree in Computer Sci-
The Ohio State University. ence from Purdue University in
His research interests include 2011. He was a faculty mem-
machine learning theory and algorithms, with a focus ber in the Computer Science
on ranking, active learning, sequential decision making, Department at University of
recommender systems, etc. He has made publications Texas at Dallas between 2011
or served as reviewers for top-tier machine learning and 2017. Since 2018, he has
conferences including ICML, NeurIPS, and AISTATS. been a faculty member in the Department of Computer
Science and Engineering at Ohio State University. His
research interests are systems and software security,
Fang Liu received his B.S. with an emphasis on binary analysis and vulnerability
degree in Information Engi- discovery. Dr. Lin is currently an associate editor of
neering from Shanghai Jiao ACM TOPS, IEEE TDSC, and IEEE TMC. He has
Tong University, China, and his also served as a TPC member for numerous systems
Ph.D. degree in Electrical and security conferences including IEEE S&P, ACM CCS,
Computer Engineering from USENIX Security, and NDSS. Dr. Lin received the
The Ohio State University. In AFOSR Young Investigator award (2014) and the
2020, he joined Facebook Inc. NSF CAREER award (2015). He is also a recipient
as a research scientist. His of VMware Faculty Research Award (2012) and the
research interests include machine learning theory and Outstanding Junior Faculty Research Award at UT
systems, with a focus on sequential decision making Dallas (2013).
under uncertainty. Dr. Liu has served as a program
committee member or reviewer for many top-ranked

100 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Ness B. Shroff received his

Ph.D. degree in electrical engi-
neering from Columbia Univer-
sity in 1994. He joined Purdue
University immediately there-
after as an Assistant Professor
with the School of Electrical
and Computer Engineering. At
Purdue, he became a Full Pro-
fessor of ECE and the direc-
tor of a university-wide center
on wireless systems and appli-
cations in 2004. In 2007, he joined The Ohio State
University, where he holds the Ohio Eminent Scholar
Endowed Chair in networking and communications, in
the departments of ECE and CSE. He holds or has
held visiting (chaired) professor positions at Tsinghua
University, Beijing, China, Shanghai Jiaotong Univer-
sity, Shanghai, China, and IIT Bombay, Mumbai, In-
dia. He has received numerous best paper awards for
his research and was listed in Thomson Reuters’ on The
World’s Most Influential Scientific Minds, and has been
noted as a Highly Cited Researcher by Thomson Reuters
in 2014 and 2015. He has served on numerous edito-
rial boards and chaired various major conferences and
workshops. He currently serves as the steering commit-
tee chair for ACM Mobihoc, and Editor in Chief of the
IEEE/ACM Transactions on Networking. He received
the IEEE INFOCOM Achievement Award for seminal
contributions to scheduling and resource allocation in
wireless networks.

© International Telecommunication Union, 2020 101

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

MACHINE LEARNING-ASSISTED CROSS-SLICE RADIO RESOURCE OPTIMIZATION:

IMPLEMENTATION FRAMEWORK AND ALGORITHMIC SOLUTION
Ramon Ferrús, Jordi Pérez-Romero, Oriol Sallent, Irene Vilà, Ramon Agustí
Dept. of Signal Theory and Communications, Universitat Politècnica de Catalunya (UPC), c/ Jordi Girona, 1-3, Barcelona, Spain

NOTE: Corresponding author: Ramon Ferrús ([email protected])

Abstract – Network slicing is a central feature in 5G and beyond systems to allow operators to customize
their networks for different applications and customers. With network slicing, different logical networks, i.e.
network slices, with specific functional and performance requirements can be created over the same
physical network. A key challenge associated with the exploitation of the network slicing feature is how to
efficiently allocate underlying network resources, especially radio resources, to cope with the spatio-
temporal traffic variability while ensuring that network slices can be provisioned and assured within the
boundaries of Service Level Agreements / Service Level Specifications (SLAs/SLSs) with customers. In this
field, the use of artificial intelligence, and, specifically, Machine Learning (ML) techniques, has arisen as a
promising approach to cater for the complexity of resource allocation optimization among network slices.
This paper tackles the description of a feasible implementation framework for deploying ML-assisted
solutions for cross-slice radio resource optimization that builds upon the work conducted by 3GPP and O-
RAN Alliance. On this basis, the paper also describes and evaluates an ML-assisted solution that uses a Multi-
Agent Reinforcement Learning (MARL) approach based on the Deep Q-Network (DQN) technique and fits
within the presented implementation framework.

Keywords – 5G, cross-slice resource optimization, deep learning, machine learning, network slicing

resources efficiently [7]-[14]. Remarkably, the

1. INTRODUCTION automation of the life-cycle management of
Network slicing allows operators to customize their network slices in the RAN requires two main
networks for different applications and customers functionalities: slice admission control and cross-
[1], [2]. Slices can differ in functionality (e.g. air slice resource optimization.
interface capabilities, mobility tracking features), in
Slice admission control is needed to decide on the
performance requirements (e.g. latency, availability,
acceptance or rejection of a new RAN slice creation
reliability and data rates), or they can serve only
request with specific coverage, functional (i.e.
specific users (e.g. public safety users, corporate
features) and performance (e.g. service quality,
customers, or industrial users). A network slice can
capacity) requirements. Under Network as a Service
provide the functionality of a complete network,
(NaaS) business models such as neutral host
including radio access network and core network
services, the slice requirements will be determined
functions. Support for network slicing has been
by the Service Level Agreement (SLA) / Service
introduced by the 3rd Generation Partnership
Level Specifications (SLS) established between the
Project (3GPP) as part of the first release of the Fifth
service provider (e.g. the operator of a RAN
Generation (5G) system specifications (Release 15),
infrastructure installed in a venue) and the
with multiple enhancements still to follow in future
customer (e.g. a Mobile Network Operator - MNO).
releases, as reflected by different study items in
The fulfillment of the RAN slice requirements may
progress, such as [3]-[6].
result in the need to guarantee the availability of a
The creation and management of network slices is certain amount of radio resources to the new slice,
especially challenging in the Radio Access Network defined in terms of, e.g. number of Resource Blocks
(RAN), where multiple slices can be delivered over (RBs) per cell, percentage of cell capacity, etc.
the same radio channel and the system shall Therefore, the slice admission control shall estimate
guarantee that the allocation and distribution of the the amount of radio resources required by the new
radio resources within the radio channel is done so slice and decide whether this can be enforced given the
that specific requirements per slice can be fulfilled deployed network capacity and the amount of
(e.g. guaranteed capacity) while using radio resources consumed by the already admitted slices.

© International Telecommunication Union, 2020 103

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Once multiple slices have been activated in the RAN, paid to the practical implementation aspects of
the cross-slice resource optimization shall ensure these solutions, as it will be further discussed in
that the slice requirements are satisfied over time Section 2. In this respect, departing from 3GPP and
and RAN resources are efficiently utilized. This may O-RAN Alliance specifications, a first contribution of
imply a dynamic modification of the slice this paper is the delineation of the functional
configurations (e.g. specifying the amount of radio framework and information models to be accounted
resources assigned to each slice at each cell, when targeting a practical realization of ML-
adjustment of slice-aware scheduling settings, assisted cross-slice radio resource optimization
configuration of rate limiters, bandwidth parts, solutions for 5G and beyond systems. More
mobility load balancing parameters, access control specifically, the focus is put here on the
priorities, etc.) during its lifetime in order to deal identification of the specific functional components
with the dynamics of the traffic load of the slice and enabling the deployment of ML-based solutions for
with the random propagation effects that lead to RAN management along with the set of information
non-deterministic mapping between radio resource models that have been defined to represent SLAs,
consumption and performance requirements. network slice instances’ characteristics and slicing-
Cross-slice resource optimization has been related configuration parameters of 5G base
identified by 3GPP as a use case in the context of stations. On this basis, a second contribution of this
Self-Organizing Network (SON) feasibility studies paper is the formulation and assessment of a
[15], addressing not only the dynamic allocation of plausible ML-assisted cross-slice radio resource
radio resources to slices but also the distribution of optimization solution that fits within the delineated
other resources such as storage and computing for implementation framework. The solution makes
virtualized implementations. use of Multi-Agent Reinforcement Learning (MARL)
based on the Deep Q-Network (DQN) technique.
The decision-making logic for cross-slice resource
Illustrative performance results of the proposed
optimization needs to deal with a lot of
solution are provided by means of simulations.
uncertainties and random processes associated
with the variability in traffic generation, device The rest of the paper is organized as follows.
mobility and radio channel conditions, so it is highly Section 2 presents an overview of related works in
difficult to have an accurate a priori statistical order to position the paper in relation to the state-
knowledge of the network resource utilization and of-the-art. Section 3 describes the implementation
delivered performance. For this reason, model-free framework, which is particularized to the proposed
Machine Learning (ML)-based methods, which do ML-assisted cross-slice optimization solution in
not rely on predefined models but are able to learn Section 4 and Section 5 presents some illustrative
and/or predict the particular network dynamics as proof-of-concept results. Finally, our concluding
well as to operate under goal-oriented policies, remarks are wrapped up in Section 6.
become adequate solutions to the problem [16].
Besides, the complexity of the problem with a huge
2. RELATED WORK
number of variables and conditions (e.g. particular Artificial Intelligence (AI) and more specifically ML
device capabilities, pending traffic, link channel techniques have been applied in the literature for
conditions, resource consumption, etc.) also pushes both slice admission control and cross-slice radio
for the introduction of these sorts of methods. As a resource allocation. In the area of slice admission
result, the system can be in a large number of control, [17] studied an optimal algorithm using
possible states in which the cross-slice resource Semi-Markov Decision Processes (SMDP) and then
allocation needs to determine the optimum capacity proposed an adaptive algorithm based on Q-
sharing among slices. In this case, among the learning. Then, other works have considered deep
possible ML techniques, deep reinforcement Q-learning [18] along with variants for enhancing
learning (RL) schemes become particularly relevant the training process, such as deep dueling neural
because they provide faster convergence under networks [19]. ML tools have also been used for
large state/action spaces in comparison with enhancing the slice admission control with traffic
classical reinforcement learning. prediction, such as in [20], [21], which use Holt-
Winters prediction, or [22], which uses a
While there is a significant amount of work
combination of Long Short Term Memory (LSTM)
addressing the cross-slice resource optimization
and dense neural networks for predicting the
problem from an algorithmic and performance
resource usage.
assessment perspective, less attention has been

104 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

In the field of cross-slice optimization, different Another important novelty comes from the
approaches exist exploiting several ML tools. specification of the SLA terms for a RAN slice to be
Q-learning was used in [23] to design a slicing used by the ML-based solution. This paper takes as
controller that decides which resource units are a reference the attributes defined in the GSMA
allocated to each slice based on requirements at the Generic Slice Template considered by 3GPP to
user level. Q-learning complemented with a genetic specify the SLA to be fulfilled for a RAN slice across
algorithm was considered in [24] for scaling down a geographical area covering multiple cells in terms
allocated resources to slices for congestion control of, e.g. the total amount of capacity to be provided
purposes. In [25] deep deterministic policy gradient to each slice. Instead, other approaches such as
(DDPG) is used to allocate resource blocks to [28]-[32] just consider the SLA specified in terms of
different tenants in a cloud RAN environment. In the QoS parameters defined at the user level, but
turn, game theory with exponential learning is without enforcing any aggregate capacity per slice.
proposed in [26] to divide the network resources
Finally, another difference with respect to previous
(i.e. bandwidth) among slices using OpenFlow,
works comes from the algorithmic solution
being a general approach not particularized to the
considered in the proposed framework, which
specificities of radio resource allocation. Recently,
consists of a multi-agent DQN with one agent per
deep Q learning has become a quite popular tool for
slice that learns the capacity to be allocated to each
allocating radio resources to slices, as reflected by
slice in each cell. In contrast to single agent
works [27]-[33] that include different variants of
solutions like those of [30], [31], which jointly
this technique and address the problem from
consider all the tenants when making decisions, the
different perspectives, such as the joint allocation of
multi-agent approach has advantages such as better
computational resources and radio resources to
scalability as it allows easily adding/removing
users in [27], the allocation of aggregate capacity
slices in the scenario simply by adding/removing
per slice to multiple cells in [28], [29], the allocation
the corresponding agent. Moreover, while some
of resources to slices on a single cell basis in [30],
multi-agent approaches have already been
[31], [32], or the allocation of per-cell resources to
considered in [28], [29][32], the one considered
the different slices jointly considering multiple cells
here has the advantage that an agent learns the
in [33]. Finally, other works have proposed the use
policy for assigning capacity to be provided to the
of traffic forecasting for cross-slice resource
slice in each cell, in contrast to [32], which
allocation, applying techniques such as LSTM neural
considered the capacity in a single cell, or [28], [29],
networks [34], deep convolutional neural networks
which provided the aggregated capacity over all the
[35], Generative Adversarial Networks (GANs) [36],
cells.
or deep neural networks [37].
This paper introduces several novelties with
3. ML-ENABLED CROSS-SLICE
respect to previous works. First of all, this paper MANAGEMENT FRAMEWORK
presents a functional framework aligned with 3.1 O-RAN framework for ML-assisted
current 3GPP and O-RAN specifications for solutions
implementing ML-assisted cross-slice radio
resource optimization and particularizes it to a As part of the specification of new interfaces and
specific algorithmic solution coming from our functionality for an open and intelligent RAN, the O-
previous work [33]. Instead, the above-mentioned RAN Alliance is working on the definition of a
works have put the focus on algorithm development framework for the deployment of ML-assisted
but without going into detail of the mapping on solutions within the RAN (i.e. solutions that rely on
existing architectures from standardization bodies. the use of ML models such as supervised learning,
For example, some works just consider a slicing reinforcement learning, etc.) [38].
controller (e.g. [23]) or a network slicing module A representation of the overall RAN functional
(e.g. [28], [29]) but without providing details of how architecture being defined by O-RAN is illustrated
this would be mapped on practical architectures. in Fig. 1 [39]. This constitutes a disaggregated RAN,
Only in [24] an architectural framework for slice compliant with 3GPP specifications, where the
management and orchestration that is aligned with radio protocol stack is split and distributed between
3GPP is presented, but without providing specific different RAN nodes. In particular, the O-RAN Radio
details on the algorithm implementation. Unit (O-RU) hosts the RF processing and the lower
part of the PHY layer functionality (e.g. i/FFT

© International Telecommunication Union, 2020 105

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

processing), the O-RAN Distributed Unit (O-DU) is perform policy management, ML model
in charge of the High-PHY layer processing (e.g. management (described below in more detail) and
modulation, channel coding), Medium Access delivery of enriched information for near-RT RIC
Control (MAC) and Radio Link Control (RLC), the O- operation (e.g. RAN data analytics that could be
RAN Central Unit - Control Plane (O-CU-CP) hosts exploited by the near-RT RIC). Furthermore,
the upper layers of the control plane radio protocol complementing the A1 interface, the interactions
stack, i.e. Radio Resource Control (RRC) and control between the SMO and the underlying RAN nodes
plane of Packet Data Convergence Protocol (PDCP), also rely on the adoption of other standardized
and the O-RAN Central Unit - User Plane (O-CU-UP) interfaces named as O1 and O2 in Fig. 1. In
handles the upper layers of the user plane protocol particular, O1 refers to the set of service-based
stack, i.e. Service Data Adaptation Protocol (SDAP) management interfaces being standardized by
and user plane of PDCP layers. Then, sitting on top 3GPP for configuration, performance and fault
of these RAN nodes handling the distributed radio management of the RAN functionality [41]. In turn,
protocol stack, there is the near-real-time RAN the O2 interface supports the management of the
Intelligent Controller (near-RT RIC), which serves cloud infrastructure and resources allowing the
as the brain of the RAN by coping with the different execution of virtualized RAN functions.
Radio Resource Management (RRM) functions
Building upon such a RAN reference architecture,
needed for overall RAN operation, such as radio
Fig. 2 shows the main components and relations
connection, mobility, Quality of Service (QoS) and
being delineated under O-RAN for the training and
interference management. With respect to the
deployment of ML-assisted solutions within the
interfaces between these RAN nodes, E1, F1-c and
SMO layer and/or within the RAN nodes themselves.
F1-u interfaces are specified by 3GPP while Open
fronthaul and E2 are being specified by the O-RAN ML Training Host
Alliance. Off-line data for
model training ML Training
Service management and orchestration (SMO) (and evaluation)

Non-RT RIC
ML model deployment

A1 O1 ML Inference Host
Online data for
O2 model inference
Near-RT RIC
ML Inference

O-CU-CP Outputs
E2 (RRC, PDCP) E1 O-CU-UP Data Collection
and pre-
Actor
(SDAP, PDCP)
processing
F1-c F1-u
ML-Assisted
O-DU: RLC/MAC/PHY-high Solution

Open FrontHaul

O-RU: PHY-low / RF Data: Actions:

Management and operational -Internal actions within the actor
data (e.g. performance (i.e. subject of action is the actor itself)
O-Cloud
measurements, number of -Configuration management over O1
connected devices, alarms) -Policy management over A1
Fig. 1 – O-RAN functional architecture from RAN nodes and UEs -Control actions over E2
Moving at the management plane, O-RAN defines
the Service Management and Orchestration (SMO) Fig. 2 – Components and relations for ML-assisted solutions
layer, which actually represents the Operations within O-RAN
Support Systems (OSS) of the MNO for the RAN As shown in Fig. 2, a variety of management and
domain. As part of the SMO layer, O-RAN basically operational data is collected from the different RAN
defines the role of a non-real-time RAN Intelligent nodes and User Equipment (UE) devices. Such data,
Controller (non-RT RIC) entity for the interaction properly preprocessed, is used to feed the two key
with the near-RT RIC via the A1 interface, which is components of the ML processing workflow,
also being standardized by the O-RAN Alliance. denoted as the ML training host and the ML
Through the A1 interface [40], the non-RT RIC can inference host. The ML training host represents the

106 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

runtime environment where offline training of the into other network components via the different
ML model takes places. This refers to the training of specified interfaces. For example, management
model before being executed within the network. In configuration actions from an actor within the SMO
addition to the data collected from the real network, layer on any RAN node can be conducted via the O1
offline training may also rely on synthesized data interface, control actions on O-CU/O-RU from an
which can accurately reproduce the behavior of the actor within the near-RT RIC can go over the E2
real network environment. The training may interface and policy management configuration
include an evaluation stage to assess the actions between the non-RT RIC and the near-RT
performance of the model and validate that it is RIC can be communicated over the A1 interface.
ready and reliable to be deployed in the live
network environment. Offline training is necessary
3.2 Information models for network slice
to obtain supervised learning models (e.g. deep management
neural networks, support vector machines, etc.) as With regard to the management of network slicing
well as reinforcement learning models (e.g. Q- in 5G networks, 3GPP specifications include
learning, multi-armed bandit learning, deep RL). information model definitions, referred to as
The training host component is likely to be part of Network Resource Models (NRMs), for the
the SMO layer. characterization of network slices [42] together
The ML inference host represents the runtime with a set of management services (MnS) for
environment where the (previously trained and network slice life-cycle management (e.g. network
validated) ML model is executed and fed with online slice provisioning MnS for network slice creation,
data to produce the outputs that will be used in the modification and termination, performance
network operation. Multiple ML inference hosts can monitoring services per slice, etc.) [43]. In addition,
be in place, whose location depends on aspects such work is being conducted at 3GPP level to support
as the purpose and type of ML models being SLA/SLS management [44], as well as closed loop
executed, its computation complexity, the assurance solutions that allow a service provider to
availability and the quantity of data used and the continuously deliver the expected level of
response time requirements (real-time or non-real- communication service quality in a 5G network [45].
time) of the ML application. Hence, ML inference Fig. 3 provides an overview of the different types of
hosts can be placed within the SMO layer but also information models and their relations that are
within the RAN nodes (i.e. near-RT RIC, O-CU, O-DU). relevant for network slice management. The main
In turn, the actor represents the network entity (i.e. idea behind the overall flow of the information
UE, O-DU, O-DU, Near-RT RIC and Non-RT RIC) that models, as illustrated in Fig. 3, is that a network slice
hosts the decision-making function that consumes is conceived as a “product” offered by a Network
the outputs of the ML inference host and takes Slice Provider (NSP) to a Network Slice Customer
actions. It is worth noting that the distinction (NSC). In this respect, the GSMA Generic Slice
between the ML inference host and the actor obeys Template (GST) is used as the SLA information
the fact that these components may or may not be associated with the network slice product for the
co-located as part of the same network entity. An communication between the NSC and NSP through,
example of non-co-location could be the case of a e.g. a Business Support Systems (BSS) product
mobility prediction model executed in an inference order management Application Programming
host within the non-RT RIC that produces outputs Interface (API).
(e.g. mobility patterns) that are retrieved and The GSMA GST provides a standardized list of
consumed by the near-RT RIC (i.e. the actor in this attributes (e.g. performance related, function
case) for enhanced RRM (e.g. handover decisions related, etc.) that can be used to characterize
based on mobility patterns). In contrast, an example different types of network slices [46]. GST is generic
of co-location could be an RRM algorithm for and is not tied to any type of network slice or to any
mobility management that embeds a reinforcement agreement between an NSC and an NSP. A Network
learning model and is executed within the near-RT Slice Type (NEST) is a GST filled with (ranges of)
RIC, which in this case serves as both the inference values. There are two kinds of NESTs: Standardized
host and the actor. The actions decided by the actor NESTs (S-NEST), i.e. NESTs with values established
can be handled either internally within the actor by standards organizations, working groups, fora,
(e.g. RL-based RRM algorithm for mobility etc. such as, e.g. 3GPP, GSMA, 5GAA, 5G-ACIA, etc.;
management within the near-RT RIC) or enforced and Private NESTs (P-NEST), i.e. NESTs with values

© International Telecommunication Union, 2020 107

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

decided by the NSP. Among the attributes included NetworkSlice and NetworkSliceSubnet, specified in
in the GST there are: the 3GPP information model definitions for
“network slice NRM” [42]. The NetworkSlice IOC and
• Attributes that specify the area where the
the NetworkSliceSubnet IOC represent, respectively,
terminals can access a particular network slice
the properties of a Network Slice Instance (NSI) and
and the spectrum used.
a Network Slice Subnet Instance (NSSI) in a 5G
• Attributes that define the services and network. It is worth clarifying at this point that the
capabilities supported in the slice (e.g. “support realization of an NSI may be tied to the realization
for non-IP traffic”, “MMTel support”, “NB-IoT of several NSSIs, that is, an NSI that is composed of,
support”, “Positioning support”, etc.) e.g. a RAN NSSI, a 5G Core (5GC) NSSI and a
Transport Network NSSI. However, depending on
• Attributes that establish the capacity and
the NSP’s product offering, it is also possible the
performance of the slice, including guaranteed
realization of an NSI composed of a single domain
and maximum data rates per slice and per UE,
such as, an NSI consisting of a single, RAN-only NSSI.
as well as maximum number of concurrent
Within the NetworkSlice IOC and
sessions and terminals in the slice.
NetworkSubnetSlice IOC, the attributes that are
• Attributes that define the terminal mobility defined to encode the network slice related
conditions and density of terminals. requirements that should be supported by the NSI
and the NSSI are named, respectively, ServiceProfile
• Attributes that define management and
and SliceProfile. Such attributes are compound data
operational aspects and features of the slice (e.g.
types that include attributes directly inherited from
performance monitoring indicators,
the GSMA GST template, as well as additional
performance prediction indicators, user
attributes to capture more specific requirements
management openness).
derived from the service performance
• Attributes that define the isolation level of the requirements defined in [47], [48] . In particular, let
slice with regard to other slices (e.g. physical us introduce here three attributes included in the
isolation, logical isolation). ServiceProfile that are are directly inherited from
the GSMA GST and used in the algorithm presented
Network Slice SLA
GSMA
in Section 4:
GST, P-NEST/S-NEST
• dlThptPerSlice: It defines the achievable
Network Slice Consumer aggregate downlink data rate of the network
BSS Product Order Management API
slice.
Network Slice Provider
NSP’s OSS • dlThptPerUe: It defines the average data rate
3GPP Network Slice NRM delivered by the network slice per UE.
NetworkSlice
Transport
(NSI)
Network NSSI • termDensity: It specifies the maximum user
3GPP Network Slice density over the coverage area of the network
3GPP Network Slice NRM
NRM
NetworkSubnetSlice slice.
NetworkSubnetSlice
(5GC NSSI)
(RAN NSSI)
And last but not least, as also captured in Fig. 3,
3GPP also provides information model definitions
for “New Radio NRM” [42], where different RAN
3GPP New Radio NRM
RRMPolicyManagedEntity
management parameters are defined to configure
the behaviour of the RAN nodes with regard to the
operation of the established network slices. In this
O1 interface respect, a more detailed, though still simplified,
RAN Domain
5GC TN view of the classes and attributes of the “New Radio
(O-RU, O-DU, O-CU,
Near-RT RIC)
Domain Domain NRM” model that allow for the characterization of
the RRM policies for configuring the way that
Fig. 3 – Information models for network slice management resources are allocated to the slices is provided in
Fig. 4. Specifically, the RRMPolicyManagedEntity
Internally, within an NSP’s OSS, the managerial
proxy class represents the different RAN managed
representation of the network slice is realized with
components (e.g. cell resources managed at DU, cell
two Information Object Classes (IOC), named
resources managed at CU functions, DU functions,

108 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

etc.) that are subject to the RRM policies and the The sum of the rRMPolicyDedicatedRatio values
RRMPolicy_ IOC represents the properties of an shall be less or equal than 100.
abstract RRMPolicy that defines two attributes: the
• rRMPolicyMinRatio, defines the minimum
resourceType attribute, used to define the type of
resource usage quota for the associated
resource (e.g. PRB, RRC connected users, etc.) and
network slice(s), including prioritized
the rRMPolicyMemberList attribute, used to indicate
resources and dedicated resources, which
the associated network slice or group of network
means the resources quota that need to be
slices that is subject to this policy. The associated
guaranteed for use. The sum of the
network slices are specified here in terms of slice
‘rRMPolicyMinRatio’ values shall be less or
identifiers such as the Single Network Slice
equal than 100.
Assistance Information (S-NSSAI) and Public Land
Mobile Network Identifier (PLMNid). • rRMPolicyMaxRatio, defines the maximum
resource usage quota for the associated
RRMPolicyManagedEntity network slice(s), including shared resources,
<<proxy class>> prioritized resources and dedicated resources.
Proxy class that represents an The sum of the ‘rRMPolicyMaxRatio’ values can
1
<<IOC>>NRCellCU, or an be greater than 100.
<<names>> <<IOC>>NRCellDU or an
4. ML-ASSISTED CROSS-SLICE
<<IOC>>GNBCUUPFunction, or an
<<IOC>>GNBCUCPFunction, or an
OPTIMIZATION SOLUTION
*
<<IOC>>GNBDUFunction. This section describes an ML-assisted solution for
cross-slice optimization based on the O-RAN
RRMPolicy_
framework and network slicing information models
Abstract
presented in the previous section. The solution is
<InformationObjectClass>>
conceived to be deployed as part of the RAN SMO.
<<inherits>> The functional model and components of the
resourceType
rRMPolicyMemberList
solution are illustrated in Fig. 5. The cross-slice
radio resource optimization problem considered
here consists of dynamically adjusting the amount
of downlink radio resources assigned to each RAN
RRMPolicyRatio
slice in each of the cells where the RAN slice is
<InformationObjectClass>>
accessible in order to account for the spatio-
temporal traffic variations across the cells. The
rRMPolicyMaxRatio solution is designed to operate with N cells and K
rRMPolicyMinRatio RAN slices and keep track of the traffic variations in
rRMPolicyDedicatedRatio periods (time steps) of t minutes. This is achieved
through the dynamic configuration of the
Fig. 4 – Simplified representation of the classes and attributes
for configuration of RAN functions subject to RRM policies for rRMPolicyDedicatedRatio attribute of each cell on a
slice management. per RAN slice basis. This configuration is conducted
via the management provisioning services offered
On this basis, the RRMPolicyRatio IOC represents a
by the O1 interfaces, as seen in Fig. 5. In particular,
particular realization of an RRMPolicy. Specifically,
since the rRMPolicyDedicatedRatio attribute
it establishes a resource model for resource
establishes the resource usage quota assigned to
distribution among slices based on three resource
the RAN slice defined in terms of the fraction of
categories: shared resources (shared among slices
Physical Resource Blocks (PRBs) that can be used
with no specific guarantees per slice), prioritized
by this slice, the attribute is configured in the O-DU
resources (guaranteed for use by associated slices
unit, so that the MAC layer can take this resource
but still usable for other slices when free), and
usage quota into account when allocating PRBs to
dedicated resources (only used for the associated
the users of the RAN slice in the cell.
slices). Accordingly, the following attributes are
included in the RRMPolicyRatio IOC: For determining the rRMPolicyDedicatedRatio for
the n-th cell and the k-th slice, denoted as (k,n), the
• rRMPolicyDedicatedRatio, defines the dedicated
resource usage quota for the associated
network slice(s), including dedicated resources.

© International Telecommunication Union, 2020 109

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

GSMA
P-NEST/S-NEST

SMO RAN NSI ServiceProfile / NSSI SliceProfile:

dlThptPerSlice, dlThptPerUe, termDensity

ML Training Host ML Inference Host & Actor

Trainer of RAN Cross-Slice Management Policies RAN Cross-Slice Manager

DQN Agent Slice 1 Policy Slice 1:
Evaluation NN
r(1)
parameters (1) (1)
Training Evaluation Target
Experience a(1)
s(1) NN NN
Required Environment Dataset Obtain
Real Network Data

capacity state
Synthethic Data /

per tenant a(1) Resource

/cell usage quota
O-CU O-CU
Policy Slice K: computation
Evaluation NN
O-DU O-DU r(K) parameters (K) (K)
DQN Agent Slice K a(K)
O-RU O-RU
s(K) Evaluation Target Obtain
Cell N Experience
Cell 1 NN NN state
Dataset
a(K)
(k,n)

Performance RRMPolicyRatio
O1 measurements O1 rRMPolicyDedicatedRatio (k,n)
per rRMPolicyMemberList (S-NSSAI)

Management interfaces O-CU O-CU

Interactions between components
O-DU O-DU
Internal interactions inside a component
Cell 1 O-RU Cell N O-RU

Fig. 5 – Deep Q Network-based cross-slice optimization solution

proposed ML-based solution relies on Multi-Agent process, as illustrated in Fig. 5.

Reinforcement Learning (MARL) based on Deep Q-
The solution considered here assumes that the SLA
Network (DQN) whose mathematical details can be
specification of the RAN slice requirements is done
found in [33]. An important advantage of this multi-
based on three ServiceProfile parameters explained
agent scheme is that it uses slice-specific DQN
in Section 3, namely dlThptPerSlice, dlThptPerUe
agents and action selection policies for the training
and termDensity, which are directly derived from
and inference processes and, therefore, slices can be
the GSMA GST template and used as inputs for the
easily added/removed in the scenario just by
different solution components described in more
adding or removing the corresponding agent and
detail in the following subsections. The specific
action selection policy. Specifically, as seen in Fig. 5,
values of these parameters for the slice k are
as part of the solution there is a Resource Usage
denoted as dlThptPerSlice(k), dlThptPerUe(k) and
Quota Computation module that determines the
termDensity(k) 1.
values of (k,n) based on the outputs obtained
through the execution of K action selection policies 4.1 RAN cross-slice manager
(k), each one associated to one slice. Each one of This component includes the inference part of the
these policies is specified through a deep neural DQN model and the functions needed to configure
network (NN) defined by a vector of parameters k the RAN nodes through the O1 interface (i.e. it takes
that have been previously learnt during the training the ML inference and actor roles of the O-RAN

1 For the interested reader, the parameters of the algorithm

described in [33] are related to the considered Service Profile
attributes as follows: Scenario Aggregated Guaranteed Bit Rate
(SAGBR)=dlThptPerSlice; Maximum Cell Bit Rate (MCBR) of
slice k in cell n: MCBR(k,n)=dlThptPerUe(k)  termDensity(k) 
cell n service area.

110 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

framework). The computation of (k,n) and followed by the different slices in order to maximize
potential update of the rRMPolicyDedicatedRatio the cumulative reward are learnt offline by the ML
attribute is done every t minutes. The training host who provides them to the ML
determination of the resource usage quota (k,n) is inference host. Further details about this training
realized through the following functions: process and the reward formulation are given in
Section 4.2.
• Per-slice action selection policies
• Resource usage quota computation
The action selection policy of slice k gets the
network state s(k) observed for this slice at the time This function computes the value of the resource
when the policy is executed and determines the usage quota σ(k,n) to be allocated to each slice and
action a(k) to be applied for this slice. The action a(k) cell for the next time step by applying the
is composed of N per-cell actions that take one out increase/maintain/decrease actions provided by
of three possible values corresponding to: increase the action selection policies of all the slices and
the resource usage quota (k,n) for slice k in cell n configures the resulting σ(k,n) values in the O-DU
in an amount of Δ for the next time step, maintain through the rRMPolicyDedicatedRatio attribute. To
the same resource usage quota or decrease it in an make the configuration on a per-slice basis, an
amount of Δ. rRMPolicyMemberList is specified for each RAN slice,
being composed of a single member with the S-
In turn, the state s(k) includes N different per-cell NSSAI and PLMNid of the RAN slice. Then, the
components, each one given by the triple rRMPolicyDedicatedRatio is configured per
<ρ(k,n),σ(k,n), σava(n)> where ρ(k,n) is the fraction rRMPolicyMemberList in each cell.
of PRBs occupied by the slice k in cell n, σ(k,n) is the
current resource usage quota allocated to the slice When applying the actions, this function ensures
and σava(n) is the total amount of resource usage that the maximum cell bit rate value associated to
quota in the cell not allocated to any slice. While the the termDensity and dlThptPerUe parameters is not
values of σ(k,n) and σava(n) are directly available at exceeded. Moreover, since the action selection
the RAN cross-slice manager, the value of ρ(k,n) is policies for the different slices operate
obtained from the different cells through the independently, this function also checks that the
performance management (PM) services offered by aggregated resource usage quota for all the slices in
the O1 interface. In particular, using the a cell after applying the actions does not exceed 1 in
performance measurements defined in [49], it order not to exceed the cell capacity. If this happens,
corresponds to the ratio between the “DL PRB used it applies first the actions of the slices involving a
for data traffic”, which measures the number of reduction or maintenance of the resource usage
PRBs used in average for data traffic in a given slice quota and the remaining capacity is distributed
and cell, and the “DL total available PRB”, which among the slices that have increase actions. This
measures the number of available PRBs in the cell. distribution is proportional to their dlThptPerSlice
Both measurements are collected from the gNB-DU values, as long as their current throughput is not
every time step, so their average is performed along already higher than the dlThptPerSlice. For doing
the time step duration t. this adjustment, the measured throughput per slice
across all the cells in the last time step is needed. It
Following the DQN approach, the action selection can be obtained from the PM services of the O1
policy (k) of the k-th slice seeks to maximize a interface using the “Downstream throughput for
cumulative reward that captures the desired Single Network Slice Instance” Key Performance
optimization target to be achieved. In particular, the Indicator of [50].
action selection policy (k) for a given state s(k) is
defined as argmaxa(k) Qk (s(k),a(k),θk ) , where 4.2 Trainer of RAN cross-slice management
policies
Qk(s(k),a(k),k) is the output of a deep NN for the
input state s(k) and the output action a(k), This component constitutes the training part of the
providing the maximum expected cumulative DQN model intended to learn the NN parameters k
reward starting at s(k) and triggering a(k). The that determine the per-slice action selection
internal structure of the NN is specified by the policies to be used by the RAN cross-slice manager.
vector of parameters k that contains the weights of
The training process makes use of a multi-agent
the different neuron connections. The optimum
DQN approach in which each DQN agent learns the
values of k that determine the policies to be
optimum policy of a different RAN slice by

© International Telecommunication Union, 2020 111

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

continuously interacting with a training where φ1, φ2 and φ3 are the weights of each
environment and updating the NN parameters as a component.
result of these interactions. The training
The first and second components in (1) correspond,
environment considered here is a network
respectively, to the SLA satisfaction ratio γSLA(k) of
simulator that mimics the behavior of the real
the slice k and the aggregate for the rest of slices
network when varying the offered load of the
k’≠k. Specifically, γSLA(k) is the ratio between the
different slices in the different cells and when
aggregate throughput obtained by the slice across
modifying the resource usage quota allocated to
all cells T(k) and the minimum between the
each slice as a result of the actions made by the DQN
aggregate offered load A(k) and the
agents. In this respect, the simulator is fed by
dlThptPerSlice(k) term of the SLA and is computed
training data consisting of multiple time patterns of
as:
the required capacity (i.e. offered load) of the slices
in the different cells. This data can be either built 𝛾𝑆𝐿𝐴 (𝑘) = 𝑚𝑖𝑛 (
𝑇(𝑘)
, 1) (2)
synthetically or extracted from real network 𝑚𝑖𝑛(dlThptPerSlice(𝑘),𝐴(𝑘))
measurements. The training is assumed to be where A(k) is the aggregate across all the cells of the
executed in a training host, located at the SMO, with per-cell offered load O(k,n) of slice k bounded by the
the necessary libraries, supporting tools and limit established by the TermDensity(k) and
computational capabilities for training the DQN dlThptPerUe(k) parameters of the SLA in the service
models and running the simulator. area S(n) of each cell n, that is:
For carrying out the training process, each DQN 𝑛=1 𝑚𝑖𝑛(𝑂(𝑘, 𝑛), dlThptPerUe(𝑘) ·
𝐴(𝑘) = ∑𝑁
agent is composed by three different elements: (i) TermDensity(𝑘) · 𝑆(𝑛)) (3)
The evaluation NN, which corresponds to the
Qk(s(k),a(k),k) being learnt that will eventually The third component of the reward is the capacity
determine the policy to be applied at the ML utilization factor, γu(k), which aims at minimizing
inference host. (ii) The target NN, which is another the over-provisioning of capacity and is defined as
NN with the same structure as the evaluation NN the ratio between the aggregate throughput T(k)
but with weights k-. It is used for obtaining the so- obtained by the slice and the total capacity allocated
called Time Difference (TD) target required for to the slice across all cells, that is:
updating the evaluation NN. (iii) The experience 𝑇(𝑘)
𝛾𝑢 (𝑘) = ∑𝑁 (4)
data set (ED), which stores the experiences of the 𝑛=1 𝐶(𝑛)·𝜎(𝑘,𝑛)
agent resulting from the interactions with the where C(n) is the capacity of cell n.
training environment as explained in the following.
The reward r(k) is provided by the training
The interactions between the DQN agent and the environment to the DQN agent at the end of each
training environment occur in time steps of time step and, correspondingly, the T(k) and A(k)
(simulated time) duration t. In each time step the values correspond to average values during the
DQN agent of the k-th slice observes the state s(k) in time step.
the training environment and selects an action a(k).
Action selection is based on an ε-Greedy policy that, As a result of the interactions between the training
with probability 1-ε, chooses the action that environment and the DQN agent, each experience of
maximizes the output of the evaluation NN, and, the ED is represented by a tuple that includes the
with probability ε, chooses a random action. As a state observed at the beginning of a given time step,
result of applying the selected action, the training the selected action, the obtained reward as a result
environment generates a reward value r(k) that of this action and the new state observed at the end
assesses how good the action was from the of the time step duration.
perspective of the desired behavior. In particular, in The experiences stored in the ED are used by the
the considered approach the reward captures both DQN agent to progressively update the values of the
the SLA satisfaction and the capacity utilization. In weights k and  k- in the evaluation and target NNs,
this way, the reward for slice k is defined as the respectively. For each time step, the update of the
weighted product of three terms given by: weights k of the evaluation NN is performed by
𝜑2
𝑟(𝑘) = 𝛾𝑆𝐿𝐴 (𝑘)𝜑1 · (
1
∑𝐾 · 𝛾𝑢 (𝑘)𝜑3 (1) randomly selecting a mini batch of experiences of
𝑘 ′ =1 𝛾𝑆𝐿𝐴 (𝑘′))
𝐾−1
𝑘′≠𝑘 the ED and updating the weights of the evaluation
NN k according to the mini-batch gradient descent

112 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

method. Moreover, the weights k- of the target NN is worth mentioning that the focus of the results is
are updated with the weights of the evaluation NN put on the temporal variations of the offered load so,
every M time steps. The reader is referred to [33] from the spatial perspective, it is assumed, for
for details on the mathematical formulation of this simplicity, that the aggregate offered load is
process. homogeneously distributed across the cells.
The training process stops after a sufficient number Table 1 – SLA parameters
of time steps that ensures the convergence of the
GSM GST
process. At this point, the ML training host is ready MNO1 MNO2
Attributes
to provide the evaluation NN parameters k so that
dlThptPerSlice 6 Gb/s 4 Gb/s
the model can be applied on the real network using
termDensity 1000 UEs/km2 500 UEs/km2
the ML inference host.
dlThptPerUe 50 Mb/s 100 Mb/s
5. ILLUSTRATIVE SCENARIO AND Table 2 – Cell configuration
EVALUATION
Parameter Value
To illustrate the behavior the proposed cross-slice
optimization framework and ML-assisted solution, Number of cells 5
let us consider a scenario with a localized RAN Cell radius 100m
deployment run by an infrastructure provider, Cell bandwidth
100 MHz (273 PRBs with 30 kHz
serving as an NSP, which offers a RAN slice product subcarrier spacing)
to a pair of MNOs, which in this case act as NSCs. Average spectral
5.1 b/s/Hz
efficiency
This could be the case of a dense urban deployment
MIMO configuration Spatial multiplexing with 4 layers
of small cells in a business district operated under a
Total cell capacity 2 Gb/s
neutral host model. Let us assume that the MNOs
use the RAN slices to offer enhanced Mobile Offered load MNO1 Offered load MNO2
BroadBand (eMBB) services to their customers so dlThptPerSlice MNO1 dlThptPerSlice MNO2

that key parameters to include in the SLA are the 8

7
number of UEs expected to be served in the area, 6

given in terms of the maximum terminal density, the 5

Gb/s

4
throughput guaranteed in the whole service area 3
2
per slice and the expected UE experienced data 1
rates. These SLA parameters are summarized in 0
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400
Table 1. On the other hand, let us assume a RAN Time(min)

deployment consisting of 5 small cells, which Fig. 6 – Offered load pattern of each MNO in Case 1.
provide an aggregated capacity of 10 Gb/s in an
area of 0.15 km2. The characteristics of this Offered load MNO1 Offered load MNO2

deployment are captured in Table 2. Under such 8

dlThptPerSlice MNO1 dlThptPerSlice MNO2

settings, note that the dlThptPerSlice values of 7

MNO1 and MNO2 SLAs actually account for the 60% 6

5
and 40% of the total capacity, respectively.
Gb/s

4
3

Two different cases of offered load patterns of the 2

1
MNOs throughout the day are considered for 0

evaluating the performance of the learnt policies. 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400
Time(min)
Case 1 shown in Fig. 6 corresponds to a situation in
Fig. 7 – Offered load pattern of each MNO in Case 2.
which the offered loads of the two MNOs exhibit a
certain complementarity during the time period The DQN-based cross-slicing solution has been
comprised between 900 and 1300 min, implemented in Python by using the library TF-
approximately, in which MNO2 exhibits a large load Agents [51]. Table 3 shows the parameters of the
while the load of MNO1 is kept at low values. DQN model (see [33] for details on these
Instead, Case 2 shown in Fig. 7 reflects a situation in parameters). To obtain the values of these
which the offered load of the two MNOs is more parameters a prior analysis of the model behavior
correlated. In both Fig. 6 and Fig. 7 the offered load with different combinations of parameters has been
corresponds to a period of one day and is conducted. The model has been trained using a data
represented as the average in intervals of 15 min. It set composed of 140 synthetically generated

© International Telecommunication Union, 2020 113

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

offered load patterns of the two MNOs in the values, from which the SLA satisfaction ratio and
different cells during one day. They capture capacity utilization metrics are determined.
different load levels and situations of
To illustrate the operation of the considered cross-
complementarity among MNOs, in order that the
slicing solution, Fig. 8 and Fig. 9 plot the evolution
DQN agents can visit multiple states during the
of the rRMPolicyDedicatedRatio parameter in %
training process.
configured by the algorithm for each slice in one of
Table 3 – DQN model parameters the cells for Cases 1 and 2, respectively. As a
reference, the evolution of the offered load pattern
Parameter Value
of each MNO, measured in % of the total scenario
Initial collect steps 5000 capacity is also shown in the plots.
Number of training steps 106
Offered load MNO1 Offered Load MNO2
Experience Data set maximum length 107
rRMPolicyDedicatedRatio MNO1 rRMPolicyDedicatedRatio MNO2
Mini-batch size 256
80
Learning rate 0.0001 70
Time steps between updates of the 60
1
target NN weights (M) 50
Discount factor 0.9 40

%
30
ɛ value (ɛ-Greedy) 0.1 20
2 layers of 100 10
Neural network nodes
nodes 0
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400
Resource quota increase () 0.1
Time(min)
Time step duration (t) 1 min
Reward weights (φ1, φ2 , φ3) (0.3, 0.2, 0.5) Fig. 8 – Evolution of the rRMPolicyDedicatedRatio for each
MNO in one cell for Case 1.
The training has been conducted with a system level
network simulator that considers the offered load Offered load MNO1 Offered Load MNO2

patterns of the different slices and cells as input. In rRMPolicyDedicatedRatio MNO1 rRMPolicyDedicatedRatio MNO2

every time step the DQN agents select the actions 80

that determine the rRMPolicyDedicatedRatio 70

60
assigned to each slice in each cell. Then, the number 50
of PRBs that are utilized by the slice is the minimum 40
%

between the assigned PRBs in accordance with 30

20
rRMPolicyDedicatedRatio and the required PRBs, 10
which are determined by the offered load and the 0

spectral efficiency. Then, the throughput achieved 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400
Time(min)
by each slice is obtained using the number of
utilized PRBs and the spectral efficiency. From this, Fig. 9 – Evolution of the rRMPolicyDedicatedRatio for each
the SLA satisfaction ratio from (2), the capacity MNO in one cell for Case 2.
utilization from (4) and the reward from (1) are Focusing on Fig. 8, it can be observed that in general
computed. The reward, together with the selected the algorithm is able to modify the amount of
action and the actual and previous states are stored resources assigned to each slice through the
in the experience data set and the weights of the rRMPolicyDedicatedRatio parameter following the
evaluation and target NNs are updated. This process offered load fluctuations, so that the algorithm
is repeated until reaching the number of training provides each slice with the resources it needs to
steps indicated in Table 3. At the end, the resulting support its load. Going into further details, different
weights of the evaluation NN determine the trained situations can be identified during the time
policy to be used during the ML inference stage. evolution of Fig. 8.
Once the training has been completed, the ML Initially at time t=0 min the
inference stage assesses the obtained policy using rRMPolicyDedicatedRatio is set to 60% and 40% for
the same system level network simulator of the slice 1 and slice 2. These values correspond to the
training, but now taking as input the offered load fractions of resources associated to the
patterns of Fig. 6 and Fig. 7 split equally among the dlThptPerSlice values established in the SLA. Then,
different cells. The trained policy is executed every as time increases, an initial transient period of
time step to obtain the rRMPolicyDedicatedRatio

114 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

around 100 min is observed in which the Table 4 presents some indicators to quantitatively
rRMPolicyDedicatedRatio is progressively adjusted assess the performance of the cross-slicing
to fit the actual resource demand. After this period, approach. Specifically, the average SLA satisfaction
the rRMPolicyDedicatedRatio approximately obtained for each MNO is presented. This is
follows the offered load of each MNO in the cell, as measured as the time average of equation (2) over
long as the aggregate load for all cells is below the the whole simulation time of 1440 min (excluding
dlThptPerSlice values. the initial transient period of 100 min) and provides
the percentage of time in which the
A particular situation occurs between t=350 min
rRMPolicyDedicatedRatio has allocated enough
and t=450 min, and between t=600 and 700 min,
resources to support the offered load while this load
approximately. In these periods the offered load of
was below the dlThptPerSlice value established in
MNO1 exceeds its dlThptPerSlice value of the SLA
the SLA. It is observed that the algorithm achieves
(60%). At the same time, the offered load of MNO2
high SLA satisfaction, above 97% in all the cases. To
during this period is still below its corresponding
further quantify the variability of the SLA
dlThptPerSlice value (40%) and the cell has enough
satisfaction, the 5th percentile of this indicator is
resources to satisfy the demands of both MNOs.
also shown in the table. The large obtained values
Therefore, for the sake of better resource utilisation,
around 91% reflect that the achieved SLA
the algorithm allows the rRMPolicyDedicatedRatio
satisfaction is very good most of the time.
of the slice of MNO1 to exceed the 60%, providing in
this way all the required capacity. An equivalent Similarly, to account for the degree of resource
situation occurs between time t=900 min and over-provisioning when allocating resources to
t=1300 min, but now it is the offered load of MNO2 each slice (i.e. for the situations in which the
that clearly exceeds its dlThptPerSlice while the load rRMPolicyDedicatedRatio allocated to a slice
of MNO1 is substantially lower than its includes more resources than actually required by
dlThptPerSlice. Again, since the cell has enough the MNO), Table 4 also shows the average capacity
capacity to satisfy the demands of both MNOs, the utilization of the assigned resources. This is
rRMPolicyDedicatedRatio of the slice of MNO2 is measured as the time average of the ratio between
increased beyond the dlThptPerSlice value of 40% the throughput of a slice and the amount of
to support all the load of this MNO. allocated capacity to this slice from equation (4).
The average is measured along the whole
In turn, the period between t=450 min and t=600 min
simulation time of 1440 min excluding the initial
corresponds to the case in which both MNOs are
transient period of 100 min. The corresponding 5th
demanding capacity beyond their dlThptPerSlice values
percentile is also indicated in Table 4. It is observed
and the cell does not have sufficient resources to support
that the algorithm achieves high utilization, being
all the demand (i.e. the sum of the offered loads of MNO1
the average above 94% and the 5th percentile
and MNO2 exceeds 100%). For this reason, it is observed
approximately above 80% for both slices. This
in Fig. 8 that the algorithm sets the
reflects that the algorithm is able to properly adjust
rRMPolicyDedicatedRatio in accordance with the
the resource allocation to the actual needs and thus
dlThptPerSlice values, i.e. 60% for MNO1 and 40% for
to reduce over-provisioning situations.
MNO2.
Table 4 – Performance indicators
Focusing now on Case 2, which reflects a larger
correlation between the offered load of both MNOs, Performance Case 1 Case 2
the results of Fig. 9 reveal a similar behavior than indicator MNO1 MNO2 MNO1 MNO2
the one discussed in Fig. 8, and the algorithm is able
to allocate to each slice the necessary amount of SLA satisfaction
98.31% 97.69% 97.38% 97.14%
resources to support their offered load. This occurs (average)
even when the offered load of a slice is above the SLA satisfaction
92.27% 90.97% 91.95% 91.75%
(perc. 5)
dlThptPerSlice limit as long as there are sufficient
Capacity
resources in the cell. In turn, when the load of both utilization 96.44% 94.06% 96.43% 94.61%
MNOs exceeds the dlThptPerSlice value and there (average)
are not enough resources in the cell (e.g. between Capacity
t=850 and t=1150 min in Fig. 9) the algorithm sets utilization (perc. 90.18% 81.79% 90.33% 79.28%
the rRMPolicyDedicatedRatio in accordance with 5)
the dlThptPerSlice values.

© International Telecommunication Union, 2020 115

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

6. CONCLUDING REMARKS [3] RP-201612, “Study on enhancement of RAN

Slicing for NR”, 3GPP TSG RAN Meeting #89e,
ML-assisted solutions with the ability to learn
September, 2020.
particular network dynamics and operate under
goal-oriented policies arise as a feasible approach to [4] SP-190931, “Feasibility Study on
tackling the complexity of the cross-slice radio Enhancement of Network Slicing Phase 2”,
resource allocation problem. Beyond the 3GPP TSG-SA Meeting #85, September, 2019.
algorithmic dimension, bringing into the equation [5] SP-200766, “Study on network slice
the information models and the architectural management enhancement”, 3GPP TSG-SA
context for the implementation of these ML-assisted Meeting #89e, September, 2020.
solutions is also key to further progress towards
their practical realization. [6] SP-200571, “Feasibility Study on Enhanced
Access to and Support of Network Slice”, TSG
In this respect, this paper has outlined the building SA Meeting #88e, July, 2020.
blocks of the architectural framework being
established under the O-RAN Alliance for the [7] R. Ferrús, O. Sallent, J. Pérez-Romero, R.
deployment of ML-assisted solutions in the RAN Agustí, “On 5G Radio Access Network Slicing:
along with the different types of information Radio Interface Protocol Features and
models developed by 3GPP for network slice Configuration”, IEEE Communications
management, from service characterization at the Magazine, Vol. 56, No. 5, May, 2018, pp.184-
SLA level down to the specific management 192.
attributes that can be used to configure how radio [8] K. Samdanis, X. Costa-Perez and V.
resources are allocated to the slices within the RAN Sciancalepore, "From network sharing to
nodes. multi-tenancy: The 5G network slice broker,"
Building upon this architectural framework and IEEE Communications Magazine, vol. 54, no. 7,
associated information models, the paper has July, 2016, pp. 32-39.
presented a plausible realization of an ML-assisted [9] O. Sallent, J. Pérez-Romero, R. Ferrús, R.
cross-slice radio resource optimization solution Agustí, “On Radio Access Network Slicing
based on the use of multi-agent DQN techniques. from a Radio Resource Management
The presented solution is shown to be a feasible Perspective”, IEEE Wireless Communications,
approach to dynamically adjust the resource Vol. 24. No.5, October, 2017,pp. 166-174.
allocation of the slices to the traffic variations in
order to fulfill an SLA and achieve high resource [10] A. S. D. Alfoudi, S. H. S. Newaz, A. Otebolaku, G.
utilization efficiency. M. Lee and R. Pereira, "An Efficient Resource
Management Mechanism for Network Slicing
ACKNOWLEDGEMENT in a LTE Network”, IEEE Access, Vol. 7, July,
This work has been supported by the Spanish 2019, pp. 89441-89457.
Research Council and FEDER funds under SONAR [11] P. L. Vo, M. N. H. Nguyen, T. A. Le and N. H.
5G grant (ref. TEC2017-82651-R) and by the Tran, "Slicing the Edge: Resource Allocation
Secretariat for Universities and Research of the for RAN Network Slicing," in IEEE Wireless
Ministry of Business and Knowledge of the Communications Letters, Vol. 7, No. 6,
Government of Catalonia under grant 2019FI_B1 December, 2018, pp. 970-973.
00102.
[12] I. Vilà, J. Perez-Romero, O. Sallent, A. Umbert,
REFERENCES “Characterization of Radio Access Network
[1] P. Rost, A. Banchs, I. Berberana, M. Breitbach, Slicing Scenarios with 5G QoS Provisioning”,
M. Doll, H. Droste, C. Mannweiler, M.A. Puente, IEEE Access, Vol. 6, March, 2020, pp. 51414-
K. Samdanis, B. Sayadi, “Mobile network 41430.
architecture evolution toward 5G”, IEEE [13] D. Marabissi, R. Fantacci, "Highly Flexible RAN
Communications Magazine, Vol. 54, No. 5, Slicing Approach to Manage Isolation,
May, 2016, pp. 84-91. Priority, Efficiency”, IEEE Access, Vol. 7, July,
[2] 3GPP TS 23.501 v16.6.0 “System Architecture 2019, pp. 97130-97142.
for the 5G System; Stage 2 (Release 16)”,
September, 2020.

116 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[14] R. Ferrus, O. Sallent, J. Perez-Romero, R. [24] B. Han, A. De Domenico, G. Dandachi, A.

Agustí, “On the automation of RAN slicing Drosou, D. Tzovaras, R. Querio, F. Moggio, O.
provisioning: solution framework and Bulakci, H.D. Schotten, “Admission and
applicability examples”, EURASIP Journal on Congestion Control for 5G Network Slicing”,
Wireless Communications and Networking, IEEE Conf. on Standards for Communications
June, 2019 and Networking (CSCN), 2018.
[15] 3GPP TR 28.861 v16.0.0 “Study on the Self- [25] X. Foukas, M.K.Marina, K. Kontovasilis, “Iris:
Organizing Networks (SON) for 5G networks Deep Reinforcement Learning Driven Shared
(Release 16)”, December, 2019. Spectrum Access Architecture for Neutral-
Host Small Cells”, IEEE Journal on Selected
[16] X. Shen, J. Gao, W. Wu, K. Lyu, M. Li, W. Zhuang,
Areas in Communications, Vol. 37, No. 8,
X. Li, J. Rao, “AI-Assisted Network-Slicing
August, 2019, pp. 1820-1837.
Based Next-Generation Wireless Networks”,
IEEE Open Journal of Vehicular Technology, [26] S. D’Oro, L. Galluccio, P. Mertikopoulos, G.
Vol. 1, January, 2020, pp. 45-66. Morabito, S. Palazzo, “Auction-based resource
allocation in OpenFlow multi-tenant
[17] D. Bega, M. Gramaglia, A. Banchs, V.
networks”, Computer Networks, Vol. 115,
Sciancalepore, K. Samdanis, X. Costa-Perez,
March, 2017, pp. 29-41.
“Optimising 5G infrastructure markets: The
Business of Network Slicing”, IEEE INFOCOM, [27] X. Chen, Z. Zhao, C. Wu, M. Bennis, H. Liu, Y. Ji,
2017. H. Zhang, “Multi-Tenant Cross-Slice Resource
Orchestration: A Deep Reinforcement
[18] D. Bega, M. Gramaglia, A. Banchs, V.
Learning Approach”, IEEE Journal on Selected
Sciancalepore, K. Samdanis, X. Costa-Perez, “A
Areas in Communications, Vol. 37, No. 10,
Machine Learning approach to 5G
October, 2019, pp. 2377-2392.
Infrastructure Market optimization”, IEEE
Transactions on Mobile Computing, Vol. 19, [28] G. Sun, Z. T. Gebrekidan, G. O. Boateng, D.
No. 3, March, 2020, pp. 498-512. Ayepah-Mensah, W. Jiang, "Dynamic
Reservation and Deep Reinforcement
[19] N. V. Huynh, D. T. Hoang, D. N. Nguyen, E.
Learning Based Autonomous Resource Slicing
Dutkiewicz, “Optimal and Fast Real-Time
for Virtualized Radio Access Networks”, IEEE
Resource Slicing with Deep Dueling Neural
Access, Vol. 7, April, 2019, pp. 45758-45772.
Networks”, IEEE Journal on Selected Areas in
Communications, Vol. 37, No. 6, June, 2019, pp. [29] G. Sun, K. Xiong, G.O. Boateng, D. Ayepah-
1455-1470. Mensah, G. Liu, W. Jiang, "Autonomous
Resource Provisioning and Resource
[20] V. Sciancalepore, K. Samdanis, X. Costa-Perez,
Customization for Mixed Traffics in
D. Bega, M. Gramaglia, A. Banchs, “Mobile
Virtualized Radio Access Network”, IEEE
Traffic Forecasting for Maximizing 5G
Systems Journal, Vol. 13, No. 3, September,
Network Slicing Resource Utilization”, IEEE
2019, pp. 2454-2465.
INFOCOM, 2017.
[30] R. Li, Z. Zhao, Q. Sun, C-L. I, C. Yang, X. Chen, M.
[21] V. Sciancalepore, X. Costa-Perez, A. Banchs,
Zhao, H. Zhang, "Deep Reinforcement
“RL-NSB: Reinforcement Learning-Based 5G
Learning for Resource Management in
Network Slice Broker”, IEEE/ACM
Network Slicing”, IEEE Access, Vol. 6,
Transactions on Networking, Vol. 27, No.4,
November, 2018, pp. 74429-74441.
August, 2019, pp. 1543-1557.
[31] C. Qi, Y. Hua, R. Li, Z. Zhao, H. Zhang, "Deep
[22] M. Toscano, F. Grunwald, M. Richart, J.
Reinforcement Learning With Discrete
Baliosian, E. Grampin, A. Castro, “Machine
Normalized Advantage Functions for
Learning Aided Network Slicing”, 21st Int.
Resource Management in Network Slicing”,
Conf. on Transparent Optical Networks
IEEE Communications Letters, Vol. 23, No. 8,
(ICTON), 2019.
August, 2019, pp. 1337-1341.
[23] A. Aijaz, “Hap-SliceR: A Radio Resource Slicing
[32] Y. Abiko, T. Saito, D. Ikeda, K. Ohta, T. Mizuno,
Framework for 5G Networks with Haptic
H. Mineno, “Flexible Resource Block
Communications”, IEEE Systems Journal, Vol.
Allocation to Multiple Slices for Radio Access
12, No. 3, September, 2018, pp. 2285-2296.

© International Telecommunication Union, 2020 117

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Network Slicing Using Deep Reinforcement [43] 3GPP TS 28.531 V16.6.0, “Management and
Learning”, IEEE Access, Vol. 8, April, 2020, pp. orchestration; Provisioning; (Release 16)”,
68183-68198. July, 2020.
[33] I. Vilà, J. Pérez-Romero, O. Sallent, A. Umbert, [44] SP-200190, “New WID Enhancement on
“A Novel Approach for Dynamic Capacity Management Aspects of 5G Service-Level
Sharing in Multi-tenant Scenarios”, IEEE Agreement”, 3GPP TSG-SA Meeting #87e,
International Symposium on Personal, Indoor March, 2020.
and Mobile Radio Communications (PIMRC),
[45] SP-200196, “New WID on Enhanced Closed
September, 2020.
loop SLS Assurance”, 3GPP TSG-SA Meeting
[34] M. Yan, G. Feng, J. Zhou, Y. Sun and Y. Liang, #87e, March, 2020.
"Intelligent Resource Scheduling for 5G Radio
[46] GSMA NG.116 - Generic Network Slice
Access Network Slicing”, IEEE Transactions on
Template Version 2.0 (2019-10-16).
Vehicular Technology, Vol. 68, No. 8, August,
2019, pp. 7691-7703. [47] 3GPP TS 22.261 v17.3.0, “Service
Requirements for the 5G System; Stage 1
[35] D. Bega, M. Gramaglia, M. Fiore, A. Banchs, X.
(Release 17)”, July, 2020.
Costa-Perez, “DeepCog: Optimizing Resource
Provisioning in Network Slicing with AI- [48] 3GPP TS 22.104 v17.3.0, “Service
based Capacity Forecasting”, IEEE Journal on requirements for cyber-physical control
Selected Areas in Communications, Vol. 38, applications in vertical domains; Stage 1
No.2, February, 2020, pp. 361-376. (Release 17)”, July, 2020.
[36] R. Giu, J. zhang, “GANSlicing: A GAN-Based [49] 3GPP TS 28.552 v16.6.0, “Management and
Software Defined Mobile Network Slicing Orchestration; 5G performance
Scheme for IoT Applications”, International measurements (Release 16)”, July, 2020.
Conference on Communications (ICC), 2019. [50] 3GPP TS 28.554 v16.5.0, “Management and
[37] S. Khatibi, A. Jano, “Elastic Slice-Aware Radio Orchestration; 5G end to end Key
Resource Management with AI-Traffic Performance Indicators (KPI) (Release 16)”,
Prediction”, European Conference on July, 2020.
Networks and Communications (EuCNC), [51] S. Guadarrama, et. al. “TF-Agents: A library for
2019. Reinforcement learning in TensorFlow.”,
[38] O-RAN Alliance, “Operator Defined Next 2018. Available at:
Generation RAN Architecture and Interfaces”, https://github.com/tensorflow/agents
https://www.o-ran.org, Accessed December,
2020.
[39] O-RAN Alliance, “O-RAN Use Cases and
Deployment Scenarios: Towards Open and
Smart RAN”, white paper, February, 2020.
[40] O-RAN Alliance, “O-RAN A1 interface: General
Aspects and Principles Version 1.0”, October,
2019.
[41] 3GPP TS 28.530 v16.0.0, “Management and
Orchestration; Concepts, use cases and
requirements (Release 16)”, September,
2019.
[42] 3GPP TS 28.541 V16.5.0, “Management and
orchestration; 5G Network Resource Model
(NRM); Stage 2 and stage 3 (Release 16)”,
June, 2020.

118 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

AUTHORS Oriol Sallent is a Professor at

the Universitat Politècnica de
Ramon Ferrús received the
Catalunya (UPC) in Barcelona.
degrees of Telecommunications
He has participated in a wide
Engineering (B.S. plus M.S.) and
range of European and national
Ph.D. from the Universitat
projects, with diverse
Politècnica de Catalunya (UPC),
responsibilities as Principal
Barcelona, Spain, in 1996 and
Investigator, Coordinator and
2000, respectively. He is
Workpackage Leader. He regularly serves as a
currently a tenured Associate
consultant for a number of private companies. He
Professor with the Department
has been involved in the organization of many
of Signal Theory and Communications at UPC. His
different scientific activities, such as Conferences,
research interests include system design, functional
Workshops, Special Issues in renowed international
architectures, protocols, resource optimization and
journals, etc. He has contributed to standardization
network and service management in wireless
bodies such as 3GPP, IEEE and ETSI. He is co-author
communications. He has participated in 10+
of 13 books and has published 250+ papers, mostly
research projects within the 6th, 7th and H2020
in high-impact IEEE journals and renowed
Framework Programmes of the European
international conferences. His research interests
Commission, taking the responsibility of WP leader
include 5G RAN (Radio Access Network) planning
in H2020 VITAL and FP7 ISITEP projects. He has
and management, artificial intelligence-based radio
also participated in numerous national research
resource management, virtualization of wireless
projects and technology transfer projects for public
networks, cognitive management in cognitive radio
and private companies. He has participated in ETSI
networks and dynamic spectrum access and
standardisation activities. He is co-author of one
management among others.
book on mobile and one book on mobile broadband
public safety communications. He has co-authored Irene Vilà received her B.E.
over 120 papers mostly in IEEE journals and degree in Telecommunication
conferences, with a h-index of 23 in Google Scholar. Systems Engineering and her M.E.
degree in Telecommunication
Jordi Pérez-Romero. is a
Engineering from the Universitat
professor in the Dept. of Signal
Politècnica de Catalunya (UPC),
Theory and Communications of
Barcelona, in 2015 and 2017,
the Universitat Politècnica de
respectively. In 2018, she joined
Catalunya (UPC) in Barcelona,
the Mobile Communication
Spain, where he received a
Research Group (GRCM) of the Department of
degree in telecommunications
Signal Theory and Communications (TSC) at UPC
engineering in 1997 and a Ph.D.
where she is currently a PhD student, supported
degree in 2001. He has been working in the field of
with an FI AGAUR grant by the Government of
wireless communication systems, with a particular
Catalunya. Her current research interests include
focus on radio resource management, cognitive
Radio Access Network (RAN) Slicing, network
radio networks and network optimization. He has
virtualization and the application of artificial
been involved in different European projects with
intelligence and, particularly, machine learning to
different responsibilities, such as researcher, work
radio resource management.
package leader, and Project Lead, has participated
in different projects for private companies and has Ramón Agustí received a
contributed to the 3GPP and ETSI standardization degree of Engineer of
bodies. He has published more than 250 papers in Telecommunications from the
international journals and conferences, three books Universidad Politécnica de
and has contributed to seven book chapters. He has Madrid, Spain, in 1973, and a
an h-index of 31 in Google Scholar. He serves as an Ph.D. degree from the
Associate Editor for IEEE Vehicular Technology Universitat Politècnica de
Magazine and EURASIP Journal on Wireless Catalunya (UPC), Spain, 1978.
Communications Networks. He became Full Professor of the
Department of Signal Theory and Communications
(UPC) in 1987. After graduation he was working in

© International Telecommunication Union, 2020 119

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

the field of digital communications with particular

emphasis on transmission and development
aspects in fixed digital radio, both radio relay and
mobile communications. For the last fifteen years
he has been mainly concerned with aspects related
to radio resource management in mobile
communications. He has published about two
hundred papers in these areas and co-authored
three books. He participated in the European
program COST 231 and in the COST 259 as Spanish
representative delegate. He has also participated in
the RACE, ACTS and IST European research
programs as well as in many private and public
funded projects. He received the Catalonia Engineer
of the year prize in 1998 and the Narcis Monturiol
Medal issued by the Government of Catalonia in
2002 for his research contributions to the mobile
communications field. He is a Member of the
Spanish Engineering Academy.

120 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

6G VISION: AN ULTRA-FLEXIBLE PERSPECTIVE

Ahmet Yazar1 , Seda Doğ an Tusha1 , Huseyin Arslan1,2

1
Department of Electrical and Electronics Engineering, Istanbul Medipol University, Istanbul, 34810, Turkey
2
Department of Electrical Engineering, University of South Florida, Tampa, FL, 33620, USA

NOTE: Corresponding author: Ahmet Yazar ([email protected])

Abstract – The upcoming sixth generation (6G) communications systems are expected to support an unprecedented variety
of applications, pervading every aspect of human life. It is clearly not possible to fulfill the service requirements without
actualizing a plethora of flexible options pertaining to the key enabler technologies themselves. At that point, this work
presents an overview of the potential 6G key enablers from the flexibility perspective, categorizes them, and provides a general
framework to incorporate them in the future networks. Furthermore, the role of artificial intelligence and integrated sensing
and communications as key enablers of the presented framework is also discussed.
Keywords – 6G, adaptive, arti icial intelligence, cognitive radio, dynamic, lexibility, sensing.

1. INTRODUCTION
Following the successful standardization of the Fifth
Generation (5G) networks worldwide, academia and
industry have started to turn their attention to the next
generation of wireless communications networks [1].
At present, there are more than 100 research papers
regarding the Sixth Generation (6G) of wireless com-
munications. There is no doubt that new visions and
perspectives will continue to be developed in the coming
Fig. 1 – Flexibility terms.
years. However, despite all these efforts, current litera-
ture lacks gathering the distinctive features of 6G under (OFDM) have provided lexibility in terms of multiplex-
a single broad umbrella. ing, rate adaptation and interference management via
the exploitation of different spreading factors and the
The evolution of cellular communications through dif- multidimensional resource utilization, respectively.
ferent generations from the Radio Access Technology
(RAT) perspective is shown in Table 1. The number of The introduction of various services with rich re-
capabilities for newer cellular generations increases as quirement sets under 5G has revealed the need for a
a result of the need to meet diversi ied requirements. lexible network that can simultaneously meet diverse
Flexibility1 , where it is de ined as the capability of making requirements. 5G has given a start for lexible wireless
suitable choices out of available options depending on communications by the accommodation of different
the internal and external changes, of the communications technologies. To exemplify, the coexistence of multi-
systems eventually evolves with an increasing number numerology in a single frame has been adopted during
of new options. In this context, Fig. 2 provides a concise standardization meetings. In a given network, achieving
lexibility analysis for different generations of cellular lexibility is mainly dependent on three capabilities
communications considering the features in Table 1. 1) awareness, 2) availability of a rich set of technology
options, and 3) adaptation & optimization. On this basis,
The Second Generation (2G) systems have paved the way although the lexibility perspective has been broadened
for lexibility in communications systems by means of in 5G systems with respect to previous generations, the
multiple frequency reuse options, adaptive equalization, existing technology options are not enough to reach all
and dynamic channel allocation. The Third Generation the goals of 5G networks. Additionally, it is expected
(3G) and the Fourth Generation (4G) cellular systems that 6G networks will put further pressure on service
have incorporated voice communications with data com- providers due to emerging applications and use cases
munications. Additionally, Code Division Multiplexing corresponding to new sets of requirements. Therefore,
(CDMA) and Orthogonal Frequency Division Multiplexing 6G systems need to extend the current lexibility by
(1) exploring the awareness for the different aspects of
1 The other terms used interchangeably for lexibility are shown in Fig. 1. the whole communications network and environment

© International Telecommunication Union, 2020 121

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Features/ Generations 1G 2G 3G 4G 5G
EDGE 8PSK
GSM GMSK
CDMA 2000 QPSK, OQPSK
BPSK
BPSK, QPSK,
Modulation W-CDMA QPSK, OQPSK QPSK
FM NR 16QAM, 64QAM,
Options 16 QAM
256QAM
GPRS GMSK UMTS QPSK 64 QAM

HSDPA QPSK, 16 QAM

Convolutional Coding Turbo Coding Turbo Coding LDPC

Coding
Block Coding
Options Block Coding Convolutional Coding Convolutional Coding
Polar Coding

Modulation and Coding

Fixed Limited MCS Options Medium MCS Options High MCS Options
Scheme (MCS) Options
Fixed Lattice Fixed Lattice Fixed Lattice Adaptive Lattice
Adaptive
Fixed Shape (GMSK) Fixed Shape (RC) Windowing and Filtering
Waveform Windowing and Filtering
Options Uplink SC-FDE Uplink OFDM, SC-FDE
Fixed Type (The same for uplink Fixed Type (The same for uplink
and downlink) and downlink)
Downlink OFDM Downlink OFDM
Uplink SC-FDMA Uplink SC-FDMA
Multiple Accessing
FDMA TDMA CDMA
Options Downlink OFDMA Downlink OFDMA

800 MHz 600 MHz microWave 600 MHz – 6 GHz

Carrier Frequency AMPS 800 MHz 900 MHz
microWave microWave GSM microWave – microWave -
Options 2.1 GHz 2.5 GHz mmWave 6 GHz – 300 GHz
NMT 450 MHz 1800 MHz

Architecture Options SISO SISO SISO MIMO mMIMO

Frequency Reuse – 1 Frequency Reuse – 1

Cell Planning Frequency Reuse – 7 Frequency Reuse – 3, 4, 7, 12 Frequency Reuse – 1 Fractional Frequency Reuse Fractional Frequency Reuse
Soft Frequency Reuse Soft Frequency Reuse

ICIC COMP
User-Cell Association
Mobile-assisted Hand-off Soft Hand-off CRAN
Options Attempt to COMP
Small Cell

Freq. Multi-User Diversity

Multi-User
Freq.
Freq. Frequency Hopping Freq. FHSS Diversity Space MIMO

Diversity Space MIMO Space Precoding

Options Space Precoding

Space Beamforming
Time Path Diversity Time DSSS
Space Beamforming Space COMP

Receiver Types Multi-tap TDE Rake Receiver A Single Tap FDE A Single Tap FDE

GSM 200KHz (8 slots) CDMA 1.25 MHz BWP

1.25 MHz to 20 MHz
Carrier Aggregation
Bandwidth DAMPS 30KHz (3 slots) WCDMA 5MHz
AMPS 30 kHz
Options Multi-numerology
100 MHz with Carrier
PDC Aggregation
25KHz (3 slots) TD-SCDMA 1.6MHz License Assisted Access (LAA)

Table 1 – Increasing number of features for cellular generations.

using different sensing mechanisms including Arti icial is seen that the lexibility perspective of the key enablers
Intelligence (AI), (2) enriching technology options, and is not considered as a distinguishing feature for 6G
(3) providing optimum utilization of available options systems in the literature.
considering the awareness with practical sensing capa-
bilities. In light of the aforementioned discussions, 6G networks
require the redesign of cellular communications to
The work scopes of 6G publications in the literature are provide extreme lexibility in all of its building blocks.
summarized in Table 2. A majority of the current 6G Correspondingly, this paper elaborates the example lex-
related studies attempt to identify the future applications ibility aspects of potential 6G key enablers and provides
and their key requirements [1,3–39]. Moreover, potential a unique categorization of the related technologies and
service types and application groups for 6G are analyzed concepts. Moreover, a novel framework is proposed to
in [1, 4–14, 24–28] together with the prospective key gather the said enablers under an umbrella of a single
requirements of 6G networks. Several works are focused ultra- lexible framework for 6G.
on the key enabler technologies and concepts under
6G studies in detail [1, 3] or in general [4–13, 15–21]. The rest of the paper is organized as follows: Section 2
Furthermore, speci ic technologies and concepts are gives a brief overview for the initial forecasts on 6G to ex-
also being pushed for 6G as described in [24–83]. These plain the background for the necessity of a lexible per-
studies are revisited in the next two sections, however, it spective without examining all potential applications, re-

122 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Flexibility

Infancy Stage of Evolved Stage of

Infancy Stage of Evolved Stage of Infancy Stage of Evolved Stage of
Flexible Flexible
Cellular Cellular High Speed High Speed
Communications Communications
Communications Communications Data Rates Data Rates
Platform Platform

Only Essential Rules Only Essential Rules

Standard Defined - Standard Defined - Standard Defined - Standard Defined -
Transmitter Standard Defined - Standard Defined -
No Flexibility Low Flexibility Medium Flexibility High Flexibility
Very High Flexibility Ultra Flexibility

Channel Random Random Random Random Random Controlled

Intentionally Created Intentionally Created

Interference Avoided Avoided Controlled Controlled
and Controlled and Controlled

Receiver No Flexibility Low Flexibility Medium Flexibility High Flexibility Very High Flexibility Ultra Flexibility

Generations
1G 2G 3G 4G 5G 6G
Fig. 2 – Flexibility analysis of the previous generations and 6G communications.

quirements, and service types. Flexibility discussions on sory Virtual Reality (VR) and Augmented Reality (AR),
the potential 6G key enablers are provided under a unique holographic conferencing, virtual education, virtual
categorization rather than giving different details of these tourism, smart city, smart home, smart clothes, disaster
enablers in Section 3. The scope of Section 3 is limited and emergency management, and work-from-anywhere.
to example lexibility aspects for the potential 6G key en- This list can be longer with more applications in the up-
ablers. Next, a framework is proposed to increase inter- coming years. Most of the aforementioned applications
operability of the 6G enablers in Section 4. Finally, con- were originally envisioned for 5G, however, they could
clusions are drawn with several open issues in Section 5. not be practically realized. Therefore, it makes sense to
address them irst while developing the 6G networks.
2. A BRIEF OVERVIEW: FORECASTS ON 6G
General wireless communications requirements for the
Identi ication of the future applications, requirements given application examples can be de ined as: high data
and possible service types is one of the primary objectives rate, high throughput, high capacity, high reliability,
of the initial 6G research studies. Fig. 3 illustrates the low latency, high mobility, high security, low complex-
basic relationship between these components. Mapping ity, high connectivity, long battery life, low cost, wide
the potential future applications to the several require- coverage, and more. The importance and priority of
ments with different priorities is accepted as a irst step the requirements may change under different cases.
in general. Next, these requirements are grouped under Moreover, higher levels of performances need to be
the service types in a reasonable manner. At the inal obtained in next generation systems while meeting the
stage, service types have unique requirement sets for the related requirements.
related application groups. In 5G systems, applications
are considered under three service types including Since the requirement diversity is continuously increas-
enhanced Mobile BroadBand (eMBB), Ultra-Reliable and ing, more sophisticated service types are expected for
Low-Latency Communications (URLLC), and massive 6G. Candidate service types are constituted by grouping
Machine-Type Communications (mMTC) [84]. Among applications with similar requirements. Examples2
these, eMBB applications prioritize high throughput, ca- can be given as Big Communications (BigCom), se-
pacity and spectral ef iciency; mMTC prioritizes energy cure uRLLC (SuRLLC), Three-Dimensional Integrated
ef iciency and massive connectivity while URLLC requires Communications (3D-InteCom), Unconventional Data
high reliability and low latency. For 6G systems, some Communications (UCDC) in [11]; ultra-High-Speed-with-
of the initial studies inherently analyze the relations Low-Latency Communications (uHSLLC) in [4]; Long-
between the future applications and prioritized require- Distance and High-Mobility Communications (LDHMC),
ments to propose candidate service types [1, 3–38]. Extremely Low-Power Communications (ELPC) in [5];
reliable eMBB; Mobile Broadband Reliable Low Latency
The following list exempli ies potential 6G applications: Communication (MBRLLC), massive URLLC (mURLLC),
drone and Unmanned Aerial Vehicle (UAV) networks,
drone taxi, fully automated Vehicle-to-Everything (V2X), 2 Comprehensive discussions on these potential service types can be
remote surgery, health monitoring, e-health, fully sen- found in the given references.

© International Telecommunication Union, 2020 123

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Potential
Potential Focusing on Focusing on An Inclusive The Future of
Service Types
Applications the Key the Key Perspective A Speci ic
Ref. and
and Key Re- Enablers in Enablers in for the Key Technology
Application
quirements General Detail Enablers or Concept
Groups
This work 3 3 3 3
[1] 3 3 3
[3] 3 3
[4–13] 3 3 3
[14] 3 3
[15–21] 3 3
[22, 23] 3
[24–28] 3 3 3
[29–39] 3 3
[40–83] 3

Table 2 – Scopes of 6G publications in the literature.

Human-Centric Services (HCS), Multi-Purpose Services Although technologies placed in different categories can
(MPS) in [7]. As it is seen from the names, some of the have overlapped regions, the given categorization dif-
service types (e.g., SuRLLC, uHSLLC, reliable eMMB, ferentiates these technologies regarding their lexibility
MBRLLC, mURLLC, MPS) try to be more inclusive than aspects.
the 5G service types to serve target applications. It is also
possible to see more speci ic service types (e.g., BigCom, Table 3 provides a summary of the example lexibility op-
3D-InteCom, UCDC, LDHMC, ELPC, HCS) in comparison tions achieved by the different technologies. It is worthy
with 5G. to emphasize that the different key enablers have their
own impact on the overall lexibility of the system. Ulti-
The aforementioned applications/services envisioned for mately all of them combine together to provide the com-
6G illustrates the expected richness of its requirements. plete infrastructure capable of realizing the lexible 6G vi-
These diverse requirements necessitate an ultra- lexible sion that we aspire to achieve.
perspective for the incorporation of key enabler technolo-
gies and concepts, described below, in future networks. 3.1 Flexible Multi-Band Utilization
3. ULTRA-FLEXIBLE PERSPECTIVE FOR 6G The inclination of communications technologies towards
high-frequency bands becomes more appealing due to
In this section, an inclusive categorization of promising the increased system capacity and throughput demands
key enablers is presented for 6G communications and of cellular users. Furthermore, lexible usage of available
their example lexibility aspects are discussed in detail. frequency bands, depending on the user and service
Then, several lexibility challenges are provided for 6G. requirements, is envisioned to be an inherent character-
Key enabler categories and their related subcategories istic of future wireless networks [74].
are shown in Fig. 4. Many of these technologies are
either super icially treated or not studied during 5G The millimeter Wave (mmWave) spectrum is starting
standardizations, such as Integrated Sensing and Com- to be exploited in 5G. It provides new bene its, such as
munications (ISAC) and intelligent communications. multi-gigabit data rates and reduced interference, how-
ever, the use of mmWave bands in 5G is limited by the
APPLICATIONS current International Mobile Telecommunications (IMT)
Each service type regulations. In World Radiocommunication Conference
has unique Applications are 2019 (WRC-19), additional 17.25 GHz of spectrum is
requirement sets mapped to
for application groups requirements identi ied for IMT, where only 1.9 GHz of bandwidth
was available before [85]. Therefore, it is expected that
spectrum availability in these bands and consequently
its lexible utilization will increase during the upcoming
SERVICE TYPES REQUIREMENTS years [42]. Moreover, beyond 52.6 GHz communications
is one of the agenda items for 3GPP Release 17 [86].
Requirements are grouped
under the service types
Frequency bands from 100 GHz to 3 THz are envisioned
Fig. 3 – A basic relationship between the applications, requirements, as a candidate spectrum for 6G communications [40]. If
and service types. THz communications is employed in 6G, it promises a way

124 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Space,
Waveform Aerial,
Coexistence, Terrestrial,
mmWave
Flexible and Undersea
and NOMA Network
Free Duplex Flying Integration Cell Free
microWave THz Multidimensional
Communications Access Points Networks
Modulation &
Waveform
ULTRA FLEXIBLE
FLEXIBLE MULTI-BAND ULTRA FLEXIBLE
HETEREGENOUS
UTILIZATION PHY & MAC
NETWORKS

Spectrum
Enhanced Advanced
Coexistence RAN Slicing
Scheduling Coordinated
and Visible Hybrid and
and and
Dynamic Light Equalization Network
User Centralized
Spectrum Virtualization
Association Networks
Access

Pervasive/ Ambient
Edge AI
Radio Collective Backscattering
and Wireless
Environment Big Data AI and ISAC Security PHY Security
Distributed Power Transfer
Map Context Energy
Intelligence
Awareness Harvesting

INTEGRATED
INTELLIGENT GREEN SECURE
SENSING and
COMMUNICATIONS COMMUNICATIONS COMMUNICATIONS
COMMUNICATIONS

Zero-energy
ISAC in ISAC in
Intelligent Holographic and Cross Layer
Radar Wi-Fi SWIPT MAC Security
Surfaces MIMO Narrow-band Security
Systems Networks
IoT

Fig. 4 – Categorization of the promising 6G key enablers under the ultra- lexible perspective.

of dealing with the spectrum scarcity issue by providing 3.2 Ultra-Flexible PHY and MAC
an additional degree of lexibility in assigning the most
suitable frequency resources for given scenarios [46]. One of the unique features of 5G, speci ically in the
context of PHY design, is the introduction of numerol-
ogy concept where different con igurations of the
time-frequency lattice are used to address the varying
Apart from mmWave and THz communications, Visible
requirements [88]. While the numerology concept paves
Light Communications (VLC) also provides spec-
the way for lexibility in beyond 5G networks, it is rather
trum lexibility as a candidate key enabler for 6G
limited considering the competing nature of require-
networks [27, 52, 61]. Moreover, a new degree of free-
ments expected for future 6G networks [25]. In addition
dom that is information source lexibility is exploited
to the standardized activities, the use of lexible Cyclic
using visible light sources.
Pre ix (CP) con igurations (e.g., individual CP, common
CP, etc.) is explored to enhance the multi-numerology
systems for 6G [89].
Spectrum coexistence is another important issue in need
of lexible spectrum utilization [50, 74]. Indeed, the
Taking one step beyond the use of different realiza-
coexistence of cellular communications, Wi-Fi, satellite
tions of the same parent waveform as in 5G, multiple
networks, and radar systems is inevitable in the future
waveforms can be accommodated in a single frame for
due to both scarce resources and increasing growth in
achieving 6G goals [49, 90]. In line with this, multi-
user demands. To exemplify, the coexistence of radar and
numerology structures can be designed for promising
cellular communications in mmWave frequency bands
alternative waveforms, that are more suitable for pro-
becomes more popular nowadays [87]. Moreover, the
viding additional parameterization options. Having
idea of Dynamic Spectrum Access (DSA) relies on the
these options enhances lexibility in the PHY layer via
spectrum coexistence [56].
increased adaptation capability for meeting a large
number of requirements. Moreover, waveform coexis-
tence in the same frame gives the opportunity to serve
As it is seen, there are several aspects of lexible multi- multiple networks such as radar sensing [91] and Wi-Fi
band utilization in 6G systems. Flexibility sources can communications together with 6G communications in
be summarized under three main perspectives: 1) multi- a lexible manner. There are also several waveform-
band lexibility, 2) information source lexibility, and 3) domain NOMA studies that exploit different resource
spectrum coexistence lexibility. utilization aspects in the literature [92–95]. Moreover,

© International Telecommunication Union, 2020 125

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

partial and full overlapping through available resources for rural/isolated areas and capacity enhancement for
can also be employed while designing new generation temporarily crowded places (such as stadiums/concert
NOMA techniques [30, 96]. The waveform-domain NOMA venues) [77]. FAP-based networks are expected to be an
concept provides an important lexibility by increasing important part of 6G not only for achieving deployment
the resource allocation possibilities in 6G networks [78]. lexibility but also for having better wireless propagation
Another lexibility aspect that can arise with 6G is the provided by a high probability of Line of Sight (LOS)
use of an alternate waveform domain rather than the communications [41, 63].
conventional time-frequency lattice employed by 5G and
older generations.
In addition to the aerial and terrestrial networks, the
integration of space (satellite) networks is another as-
In addition to the waveform itself, there is a large number
pect of the lexible heterogeneous networks [54]. Space
of new generation modulation options in the litera-
networks are also a promising solution for rural area
ture [97] and only a small set of them have appeared in
communications [31]. They are employed for wireless
the 5G standards. 6G can be enriched with the lexibility
backhaul communications in the previous cellular net-
provided by these options, particularly Index Modulation
works. However, space networks can also serve aerial
(IM) based solutions [11]. This concept can even be
user equipment such as drones and UAVs to increase
extended to multiple domains to provide an additional
coverage lexibility in 6G systems [72]. Moreover, under-
degree of freedom [98]. Moreover, modulation tech-
sea network integration with the other networks will be
niques are adaptively designed considering the other
useful while serving naval platforms.
key enablers such as Non-Orthogonal Multiple Access
(NOMA) [99] and Recon igurable Intelligent Surface
(RIS) [45] for 6G. Although, the integration of different networks is en-
sured, the cell structures of these networks are changing.
Since the con iguration of the PHY parameters is, to a Cell-less or cell-free networks are one of the poten-
large extent, controlled by the Medium Access Control tial 6G concepts considering the network architecture
(MAC) layer, it is imperative to develop the lexibility and richness [105, 106]. User equipment connects to the
adaptation capabilities of both layers simultaneously. network via multiple small cells in the cell-less networks.
Two important issues that require lexibility in PHY and Cell-centric design is transformed into the user-centric
MAC would be the “waveform parameter assignment” or system. Hence, it provides both handover-free com-
“numerology scheduling” paradigm under the context of munications and zero inter-cell interference. Cell-less
5G multi-numerology systems [25, 100], where the MAC networks may exploit a new dimension of network
layer is responsible for assignment of parameters of the Multi-Input Multi-Output (MIMO) lexibility in 6G. As
PHY signal. Similarly, adaptive guard utilization methods another network MIMO example, advanced coordinated
have been developed for the MAC layer [101–103] to and centralized networks [107] are addressed together
control the new type of interferences in 5G systems. with NOMA schemes for 6G communications [108, 109].
On this basis, it is expected that highly intelligent UE These networks are called multi-cell NOMA. Flexibility
capabilities, and con igurable network parameters, and comes with the number of the cells and architecture
lexible and ef icient MAC designs will play a key role in richness while exploiting other dimensions with NOMA.
6G networks due to the expected increased diversity in
service types and consequently requirements. From the network virtualization perspective, network
slices are used in 5G to customize and optimize the
Example lexibility perspectives for ultra- lexible PHY and network for service types or any other requirement
MAC technologies of potential 6G key enablers are given sets [110–112]. Hence, the overall performance is
in Table 3. increased by meeting different requirement sets with
virtually privatized networks. Network slicing brings
3.3 Ultra-Flexible Heterogeneous Networks an important lexibility in 5G since it enables different
network options under the same umbrella. The number
Flying Access Points (FAPs) provide enhanced lexibility
of network slices can increase for 6G and there may be
for network deployment by allowing dynamic (3-D)
network slices for each user equipment. This user-centric
positioning of the nodes or even optimized trajectory
network slicing architecture can provide full lexibility in
planning for different objective functions [47,55,59]. The
the network layer.
push in this direction occurred around the turn of the
century [104], and was further empowered by projects,
such as: 1) Google Loon project, 2) Facebook Aquila The number of examples for the lexibility aspects
project, 3) ABSOLUTE project, 4) Matternet project, and of promising 6G heterogeneous networks can be in-
5) Thales Stratobus project. The integration of FAPs creased with particular technologies and concepts such
with the terrestrial network can be leveraged to provide as blockchain systems [26, 33] and quantum communica-
coverage in disaster/emergency scenarios, connectivity tions [29] in the future.

126 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Key Enabler Categorization Flexibility Aspect Example Details

Multi-band flexibility microWave, mmWave, THz, visible light
Flexible Multi-Band
Information source flexibility Radio signals, visible light
Utilization
Spectrum coexistence flexibility DSA, CR, and coexistence of cellular networks, Wi-Fi networks and radar systems
Modulation-option flexibility BPSK, QPSK, 16QAM, 64QAM, 256QAM, 1024QAM, etc.
Multi-domain modulation flexibility IM (shape, interval, position, etc.), space, time, frequency, etc.
Multi-type coding New types of LDPC, block coding, polar coding, etc.
MCS option flexibility Ultra adaptive MCS
Multi-option waveform flexibility Multiple numerologies for any specific waveform
Waveform processing flexibility Adaptive windowing/filtering and the related configurable parameters
CP utilization flexibility Individual and common CP utilizations
Adaptive guard utilization flexibility Flexible guards for multi-waveform and multi-numerology designs
Ultra-Flexible
Multi-waveform flexibility Waveform coexistence in the same frame
PHY and MAC
Multi-network multi-waveform flexibility Waveform coexistence for cellular and Wi-Fi networks with radar sensing
Multi-domain NOMA flexibility Partial and fully overlapped resources with waveform-domain NOMA
Multi-domain waveform flexibility Alternative lattice flexibility together with the time-frequency lattice
Multiple access flexibility Fully flexible, both orthogonal and non-orthogonal
Receiver-type flexibility Fully flexible, hybrid equalization
Bandwidth option flexibility BWP, carrier aggregation, LAA, DSA, etc.
User association flexibility Multiple options under heteregonous networks, flexible user parameter assignment
Channel access flexibility GB transmission, GF transmission, and their coexistence over a resource pool
Positioning flexibility of the access points Flying access points can be positioned flexibly in the sky
Connection link and relaying flexibility User equipment can connect to different type of access points
Altitude-based multi-network flexibility Coexistence of space, HAP, terrestrial and undersea networks

Ultra-Flexible Coverage flexibility Rural area coverage with space and HAP networks
Heterogenous Networks Network architecture flexibility Ultra massive MIMO, small cell, D2D, relaying via different networks, etc.
Cell-free network flexibility User-centric network designs, handover-free communications
Multi-cell flexibility Network MIMO solutions, multi-cell NOMA, etc.
Network slice flexibility Network slices for each user equipment, user-specific virtual networks

Integrated Sensing and Multi-system flexibility All systems can collabarate with the wireless communications in different ways
Communications Awareness flexibility Awareness in spectrum, location, mobility, context, user, channel, interference, etc.
Alternative solution flexibility No need to get stuck on conventional algorithm designs
Edge computing flexibility Signal and data processing at the edge nodes
Intelligent Communications Channel control flexibility Different types of intelligent surfaces
Interference management flexibility Interference management with the channel control
Softwarization flexibility Programmable architecture options for holographic MIMO systems
Battery-free implementation flexibility Removing battery limitations and constraints
Green Communications
Interference exploitation flexibility Interference can be useful for the energy harvesting
Multi-domain security flexibility No need to get stuck on key sharing security mechanisms, complementary solutions
Secure Communications
Wireless channel exploitation flexibility PHY security methods exploit the characteristics of wireless channel

Table 3 – Example lexibility aspects for the key enabler categories.

3.4 Integrated Sensing and Communications leveraged in 6G [114]. The information pertaining to the
radio environment can be utilized in improving network
With the emphasis on use cases such as V2X communica- deployment, optimizing user association, providing se-
tions in recent years, sensing has attained increased im- cure communications and so on. Hence, one of the unique
portance leading to the integration of these two applica- novelties of 6G systems is the integration of many differ-
tions [113]. However, the use of sensing is not limited to ent sensor hardware with the heterogeneous communi-
V2X or autonomous driving. Rather, if there is any observ- cations networks as exempli ied in Fig. 5.
able data that can be utilized for the optimization or en-
hancement of the communications systems, it should be While it might sound like a novel idea to some, Integrated

© International Telecommunication Union, 2020 127

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

INTEGRATED SENSING AND

COMMUNICATIONS

Drone
cameras ULTRA-FLEXIBLE Camera
HETEROGENOUS NETWORKS systems

Space
Aerial
networks
Radar
ULTRA-FLEXIBLE networks
(zeppelins)
sensing PHY & MAC
Smart
Eavesdropper Flexible
wearables
waveforms
Aerial
networks
MIMO and
(balloons) PHY Aerial
Beam-space
security networks
Vehicle (UAVs)
sensors
Body
sensors
Aerial
CoMP Terrestrial
networks
Intelligent systems networks
(drones)
surfaces

Automated
vehicles Undersea Sensors
networks

Fig. 5 – The integration of many different sensor hardware with the heterogeneous communications networks under 6G systems.

Sensing and Communications (ISAC) has been studied granular REM information.
in different domains in the past. Cognitive Radio (CR)
applications triggered the ISAC research on the last two The complete information and awareness of the environ-
decades. Spectrum sensing and awareness is one of ment comes at the cost of a high volume of data, variety
the irst application areas in the ISAC research [115]. of sources and signi icant processing [80,82]. This neces-
Location awareness is exploited to improve the wire- sitates the use of big-data processing techniques [122]. A
less communications system design in [116]. Satellite signi icant challenge, however, in this regard is the over-
and drone images can be used to predict channel pa- head of data exchange between the sensing and process-
rameters [117]. Context-awareness is used to optimize ing nodes. A centralized solution might not be suitable
network architectures in wireless communications [118]. in such scenarios, rendering the use of edge-computing
ISAC systems are studied for radar sensing [91, 119] and imperative, particularly for low-latency use cases. More-
Wi-Fi network coexistence [120] in the literature. How- over, the usage of Arti icial Intelligence (AI) solutions can
ever, the complete list of sensing information that can be be helpful while processing big-data at the edge nodes.
useful for the next generation cellular communications
systems from the ISAC perspective has not yet been
3.5 Intelligent Communications
comprehensively studied [114].
The usage of AI in the communications society has
A Radio Environment Map (REM) is a realization of increased in recent years. Several survey and tutorial
the ISAC concept [121]. It is mainly used to obtain papers are published on the usage of Machine Learning
environmental information in the literature, however, (ML) for wireless communications [34, 123–129]. AI-
for the next generation systems the REM concept will be aided design and optimization has even been leveraged
generalized from environmental-awareness to complete- for the lexible implementation options provided in
awareness. REM may include all sensing information in 5G [25]. In many of the studies, AI is put at the cen-
a multi-dimensional manner for wireless communica- ter of 6G visions [6, 8, 24, 28, 32, 44, 67, 75, 76, 83] to
tions networks. To exemplify, REM can be a specialized complement the classical methods. Indeed, the use of
database for the ISAC. Therefore, the lexibility level of AI is inevitable to incorporate intelligence in the future
the ISAC systems can be determined by the dimensions in networks [130–132]. AI-aided methods can propose fast
REMs. Each dimension in a REM increase the awareness, and ef icient solutions in case enough data is available.
allowing better resource utilization. Moreover, control
of the con igurable options and parameters in different AI and ML also ind a range of applications in ISAC and
communications layers of 6G can be enhanced by more REM paradigms to extract information regarding the en-

128 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

vironment from sensed data. A lexible communications It is also possible to bene it from Wireless Power Trans-
system needs to bene it from the advantages of popular fer (WPT) while designing zero-energy IoT systems [139].
ML approaches such as reinforcement learning, deep Under the WPT concept, Simultaneous Wireless Informa-
learning, and edge computing [37, 48, 69, 73]. Especially tion and Power Transfer (SWIPT) is the most popular
distributed intelligence (edge AI) with edge computing technology that may be a candidate for 6G networks [60,
is a promising paradigm for 6G communications [36]. 140]. SWIPT designs are also used for interference ex-
The management of multi-band utilization, MAC layer ploitation purposes [141] since interference can be use-
control, heterogeneous and cell-less networks, and the ful for energy harvesting. Transformation of interference
ISAC systems cannot be done in an all centralized man- into an energy source introduces another lexibility per-
ner. Edge computing will play an important role at that spective.
point with the help of distributed intelligence so 6G big
data can be processed at the edge nodes without being
collected at a centralized network.
3.7 Secure Communications
Intelligent networks are not limited to AI-aided concepts.
For example, RIS technology is one of the most popular
research topics nowadays [133, 134]. Intelligent surfaces With applications such as eHealth, online banking, and
bring a new lexibility on the control of channel parame- autonomous driving etc., wireless communications
ters [57]. In the past, a wireless channel was just an ob- promises to be an enabler of innumerable sensitive ap-
servable medium. However, it can be controlled at some plications utilizing private data. However, the broadcast
level with new generation wireless systems. Interference nature of wireless communications makes it vulnerable
management lexibility is increased by controlling capa- to several security threats such as eavesdropping, imper-
bilities of the wireless channel. These lexibility aspects sonation, and jamming. In order to ensure security of
also affect the technology designs in different communi- such applications, PHY Layer Security (PLS) is an emerg-
cations layers [66, 71]. To exemplify, having a control ca- ing solution that has the capability to complement the
pability in multipath propagation, such as controlling de- conventional cryptography-based security techniques. In
lay spread, Doppler spread and the number of multipath fact, PLS is more suited for the increased heterogeneity
alleviates the constraints related to waveform design. RIS and power/processing restrictions of future wireless
technology can also be considered as passive holographic networks since it exploits the characteristics of the wire-
MIMO surfaces if it is located closer to the transmitter and less channel and PHY properties associated with the link
receiver antennas [53]. Additionally, it is possible to em- such as noise, fading, interference, and diversity [142].
ploy holographic MIMO surfaces as active elements. The It is also possible to increase this lexibility by designing
active holographic MIMO surfaces work similar to mas- cross-layer security algorithms with the PHY and MAC
sive MIMO but their softwarization lexibility is higher layer [143]. In several 6G papers, secure communications
than the conventional MIMO systems [53]. is discussed as one of the main topics [8, 35, 44, 81]. PHY
and cross-layer security concepts are expected to play a
3.6 Green Communications critical role in 6G networks because of their capability
to support joint design of security, reliability, and latency.
While candidate 6G key enablers are increasing the lex-
ibility in different domains, new architectural changes
of 6G should support energy ef iciency and green com-
munications [43, 64, 70]. Zero-energy Internet of Things As discussed in the previous subsections, ISAC and REM
(IoT) is one of the most important concepts since ultra concepts will be important enablers in 6G communica-
low-power wireless communications is necessary for 6G tions. However, a new security problem arises since there
connectivity [51]. In this context, Radio Frequency (RF) may be a large amount of con idential data for ISAC and
energy harvesting is studied with ambient backscatter REM concepts. In the literature, this problem is treated
technology for 6G communications [135, 136]. Thus, in [144] for ISAC security, and in [145] for REM security.
low-power wireless systems can obtain their energy from Thus, there is a need for more secure communications op-
the available high-power radio waves. Backscatter com- tions in 6G networks to meet new types of security re-
munications enables energy harvesting, simplifying the quirements, especially for ISAC and REM concepts. More-
implementation of zero-energy IoT designs. Provision over, in order to tackle spoo ing attacks, authentication at
of rich options for energy-ef iciency promises ful ilment the physical layer by using features of channel and hard-
of energy requirement variations belonging to different ware impairments can also provide a fast, lightweight,
applications. Within this direction, the Symbiotic Ra- and ef icient alternative for crypto-security for authen-
dio (SR) concept offers highly reliable backscattering tication in future wireless networks. Furthermore, the
communications together with mutualism spectrum physical layer solution will also provide ef icient robust-
sharing [137, 138]. ness against jamming attacks using terrestrial and lying
relay and other new multi-antenna-based solution.

© International Telecommunication Union, 2020 129

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Fig. 6 – The proposed framework includes lexible key enabler platform and lexible cognitive engine. The lexible cognitive engine can be de ined as a
bridge between the requirements and potential technology options with the related con igurations.

3.8 Flexibility Challenges and Opportunities can include multiple systems together with the commu-
in 6G nications networks. Generally, the amount of sensing
information increases in parallel to awareness capabili-
The exempli ied key enablers show that 6G will have ties. However, processing the sensing information causes
many different lexibility options while 5G systems have computational burdens. Additionally, investigating the
limited lexibilities. However, each lexibility is coming ways of exploiting this information to enrich the commu-
with unique challenges. In other words, lexibility oppor- nications systems is another important challenge.
tunities bring new challenges for the 6G networks. For the lexibility challenges of intelligent communica-
For lexible multi-band utilization, operating the cellular tions, irst of all, an ef icient work distribution between
system at multiple frequency bands needs advanced conventional and ML methods is required. A large data
front-end hardware. Additionally, spectrum coexistence sets and useful features need to be developed to make
of different networks causes new interference problems. ML mechanisms more functional. Additionally, edge
If the lexibility challenges on the PHY and MAC layer computing algorithm structures should be designed to
are investigated, one of the most important problems reduce the workload at transmission points.
is the necessity of a lexible waveform system. At that
point, either a single but an ultra- lexible waveform can If we summarize the challenges and opportunities, the fol-
be designed or multiple waveforms can be employed in lowing items can be listed:
the same frame. Designing a single waveform to meet all
types of requirements did not work for 5G networks. It • Need for a rich set of algorithms and techniques at
will be more dif icult for 6G with more types of require- different layers of the protocol stack that are opti-
ments. Moreover, waveform coexistence in the same mized for different applications with their own re-
frame causes new interferences (like inter-numerology quirements.
interference in 5G). Similarly, partial and fully overlapped
NOMA systems have the same interference problem. Con- • Integration of these rich sets of algorithms into the
trol and mitigation of these interferences is expected as lexibility framework with minimal overhead and
another challenge. complexity.

• Development of techniques that allows lexibility

Flexibility challenges for heterogeneous networks can be with a simple parameter change without signi i-
exempli ied with the developing optimal positioning and cantly impacting the rest of the system design.
relaying algorithms for lying access points. In addition
to these algorithms, interference management during • Integration of AI and ML techniques to solve complex
the coexistence of different networks is necessary. As system problems together with the classical model
another challenge, network MIMO structures provide based approaches. AI/ML can be applied in different
multi-cell lexibility, however, large amounts of data parts of our proposed framework, i.e. it can be ap-
need to be transferred at the backhaul systems and the plied for better sensing and learning, or for optimal
amount of burden increases. use of the given set of algorithms and approaches, or
developing better solutions in the transmission, re-
As discussed in the previous subsections, ISAC systems ception, and modeling of the system.

130 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Therefore, there is a need for general frameworks and the lexible key enabler platform. This engine can be de-
mechanisms to ease dealing with these challenges all to- signed in an AI-aided manner to optimize the key enabler
gether. Within this direction, an example framework is lexibilities jointly. An optimum work distribution should
proposed in the next section. be done for the lexible con igurations of key enablers to
meet all the system requirements in the most ef icient
4. ULTRA-FLEXIBLE 6G FRAMEWORK way. At the end, all system requirements should be met
optimally. The lexible cognitive engine will guaran-
Gathering together all potential 6G enablers in a lexible tee this optimization by the help of Key Performance
framework is an important challenge. Therefore, this sec- Indicators (KPIs) that show the success while meeting
tion brings the above-mentioned lexible perspectives for requirements. This lexibility optimizer considers also
the key enabler technologies and concepts under the um- complexity requirements while operating the system.
brella of a single ultra- lexible framework for 6G. Here, it
is important to realize that the presence of lexible op- ISAC technologies will be an important part of 6G tech-
tions in itself is not enough to render a network intel- nologies as discussed in the previous section. Any sensing
ligent. Rather, it needs the capability to make best use information can be exploited to make the wireless com-
of the available options. Therefore, some sort of intelli- munications more effective. The lexible cognitive engine
gence or cognition is imperative in future wireless net- can give decisions with more available information while
works. Keeping this in mind, the proposed framework meeting different requirements and handling with sev-
has the following primary components: 1) Flexible key eral impairments and constraints. Sensing information
enabler platform (like an advanced Mitola radio), 2) lex- increases the awareness and controlling capabilities of
ible cognitive engine, and 3) lexibility performance indi- the system. To provide these capabilities, AI tools in the
cators. Fig. 6 illustrates how these different components lexible cognitive engine provide useful and unnoticeable
are interconnected within the framework. The key points relationships without heuristic designs and theoretical
of this framework can be summarized as follows: analysis. Hence, the lexible cognitive engine needs three
important elements while optimizing the lexibility level
1. New technologies should be integrated into commu-
with key enablers: 1) Sensing information to increase
nications standards via a lexible key enabler plat-
awareness and controlling capabilities, 2) AI tools to
form without waiting for ten years.
increase the functionality and effectiveness of sensing
2. Key enabler technologies should work together in information, and 3) KPIs to monitor the overall system.
an optimal lexibility to meet different requirements.
Therefore, a lexible cognitive engine can make an KPIs are needed to measure several performances of the
optimization between different lexibility aspects. communications system. One of these KPIs can be the
lexibility performance indicator so that the achieved
3. The amount of lexibility needs to be measured while lexibility can be quanti ied. It is dif icult to decide on
making an optimization. Hence, developing new a speci ic lexibility performance indicator because there
lexibility performance indicators is necessary. are many different lexibility perspectives as shown in
Table 3. This indicator can be technology-speci ic and
The previous cellular communications generations were require separate metrics for different technology cate-
standardized approximately ten years apart. From a gories. 6G networks will need lexibility indicators sim-
different point of view, it took about a decade for the ilar to the other KPIs such as spectral ef iciency and re-
available technologies to be included in the cellular liability. Generally, the current key enabler technologies
standards. Waiting up to ten years to bene it from an are not designed to be called lexible technologies. Flexi-
available technology does not make sense if it is possible bility aspects of these key enablers are described mostly
to develop a platform that hosts different technologies based on the inferences. In ideal conditions, 6G technolo-
lexibly. For now, we need to tolerate the limited lexi- gies need to be designed considering the lexibility per-
bility of 5G technologies for the next decade. However, spective as one of the key criteria. At that point, lexibility
an advanced Mitola radio can work like a smart phone performance indicators should be employed to quantify
that has installable and updateable software. We call this the advantages and disadvantages of new designs in both
radio a lexible key enabler platform. In this concept, the PHY and MAC layer.
the platform has the ability to have new key enabler
technologies by a softwarization. Thus, the lexibility
5. CONCLUSION
level of the wireless communications system can be
enhanced with new technologies and the related updates. 5G systems were characterized by diverse applications
and requirements. 6G is expected to continue in the same
As it is shown in Fig. 6, each technology can bring dif- vein by enriching the application fabric even further.
ferent perspectives to the overall lexibility. There is a Ful illing such a wide variety of use cases is not possible
need for a multi-objective optimization unit to control unless lexibility is incorporated in the promising key
all con igurable and lexible aspects of the enablers in enabling technologies for the future networks. Driven by

© International Telecommunication Union, 2020 131

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

this, we have presented example lexibility aspects for the nologies,” in IEEE Vehicular Technology Maga-
potential 6G key enablers under a unique categorization. zine, vol. 14, no. 3, pp. 28-41, Sept. 2019, doi:
10.1109/MVT.2019.2921208.
We believe that 6G should be approached with lexibility
at its primary design criterion. Flexibility aspects of [6] L. Zhang, Y. Liang and D. Niyato, “6G Visions:
the potential key enablers need to play a leading role Mobile ultra-broadband, super internet-of-things,
in the design stages of 6G systems. To this end, we and arti icial intelligence,” in China Communica-
have presented a general framework comprising of the tions, vol. 16, no. 8, pp. 1-14, Aug. 2019, doi:
aforementioned lexible key enablers, empowered by 10.23919/JCC.2019.08.001.
a lexible cognitive engine and supported by different
aspects of sensing and AI. We believe that the presence of [7] W. Saad, M. Bennis and M. Chen, “A Vision of 6G
lexible options is imperative but only that is not enough Wireless Systems: Applications, Trends, Technolo-
to support the future applications. The ability to extract gies, and Open Research Problems,” in IEEE Network,
information regarding the operating environment and vol. 34, no. 3, pp. 134-142, May/June 2020, doi:
making related intelligent decisions are the way forward 10.1109/MNET.001.1900287.
in the wireless communications realm. The best possible
utilization of the lexibility offered by the key enablers is [8] G. Gui, M. Liu, F. Tang, N. Kato and F. Adachi, “6G:
determined with this vision. Opening New Horizons for Integration of Comfort, Se-
curity and Intelligence,” in IEEE Wireless Communi-
The realization of a lexible key enabler platform like the cations, doi: 10.1109/MWC.001.1900516.
one mentioned above is, however, not straightforward.
It requires the methods capable of performing ef icient [9] M. Giordani, M. Polese, M. Mezzavilla, S. Rangan
multi-objective optimization to address the various and M. Zorzi, “Toward 6G Networks: Use Cases
competing applications requirements. Furthermore, and Technologies,” in IEEE Communications Mag-
quantifying the lexibility by proposing novel perfor- azine, vol. 58, no. 3, pp. 55-61, Mar. 2020, doi:
mance indicators also remains a signi icant challenge on 10.1109/MCOM.001.1900411.
the way to ensure a fully-functional lexible, cognitive
[10] J. F. Monserrat, D. Martin-Sacristan, F. Bouch-
wireless communications network.
mal, O. Carrasco, J. Flores de Valgas and N. Car-
dona, “Key Technologies for the Advent of the
6G,” 2020 IEEE Wireless Communications and Net-
ACKNOWLEDGEMENT working Conference Workshops (WCNCW), Seoul,
The authors would like to thank Muhammad Sohaib J. So- Korea (South), 2020, pp. 1-6, doi: 10.1109/WC-
laija for his valuable comments and suggestions to im- NCW48565.2020.9124725.
prove the quality of the paper.
[11] S. Dang, O. Amin, B. Shihada and M. Alouini, “What
should 6G be?,” in Nature Electronics, vol. 3, pp. 20–
REFERENCES
29, 2020.
[1] I. F. Akyildiz, A. Kak and S. Nie, “6G and Beyond:
The Future of Wireless Communications Systems,” in [12] M. H. Alsharif et al., “Sixth Generation (6G) Wireless
IEEE Access, vol. 8, pp. 133995-134030, 2020, doi: Networks: Vision, Research Activities, Challenges
10.1109/ACCESS.2020.3010896. and Potential Solutions,” in Symmetry, vol. 12, no. 4,
pp. 1-21, 2020.
[2] H. Arslan and E. Basar, “Flexible and Cognitive Radio
Access Technologies for 5G and Beyond,” IET, 2020. [13] F. Tariq, M. R. A. Khandaker, K. -K. Wong, M.
A. Imran, M. Bennis and M. Debbah, “A Specula-
[3] L. Bariah et al., “A Prospective Look: Key Enabling
tive Study on 6G,” in IEEE Wireless Communica-
Technologies, Applications and Open Research Topics
tions, vol. 27, no. 4, pp. 118-125, Aug. 2020, doi:
in 6G Networks,” in IEEE Access, vol. 8, pp. 174792-
10.1109/MWC.001.1900488.
174820, 2020, doi: 10.1109/ACCESS.2020.3019590.

[4] B. Zong, C. Fan, X. Wang, X. Duan, B. Wang and [14] G. Liu et al., ”Vision, requirements and network ar-
J. Wang, “6G Technologies: Key Drivers, Core Re- chitecture of 6G mobile network beyond 2030,” in
quirements, System Architectures, and Enabling China Communications, vol. 17, no. 9, pp. 92-104,
Technologies,” in IEEE Vehicular Technology Mag- Sept. 2020, doi: 10.23919/JCC.2020.09.008.
azine, vol. 14, no. 3, pp. 18-27, Sept. 2019, doi:
10.1109/MVT.2019.2921398. [15] P. Yang, Y. Xiao, M. Xiao and S. Li, “6G Wireless
Communications: Vision and Potential Techniques,”
[5] Z. Zhang et al., “6G Wireless Networks: Vision, in IEEE Network, vol. 33, no. 4, pp. 70-75, July/Aug.
Requirements, Architecture, and Key Tech- 2019, doi: 10.1109/MNET.2019.1800418.

132 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[16] S. Elmeadawy and R. M. Shubair, “6G Wire- [26] S. Aggarwal, N. Kumar and S. Tanwar, “Blockchain
less Communications: Future Technologies Envisioned UAV Communication Using 6G Net-
and Research Challenges,” 2019 International works: Open issues, Use Cases, and Future Di-
Conference on Electrical and Computing Tech- rections,” in IEEE Internet of Things Journal, doi:
nologies and Applications (ICECTA), Ras Al 10.1109/JIOT.2020.3020819.
Khaimah, United Arab Emirates, 2019, pp. 1-5,
[27] S. Ariyanti and M. Suryanegara, “Visible Light
doi: 10.1109/ICECTA48151.2019.8959607.
Communication (VLC) for 6G Technology: The
[17] G. Wikströ m et al., “Challenges and Technologies for Potency and Research Challenges,” 2020 Fourth
6G,” 2020 2nd 6G Wireless Summit (6G SUMMIT), World Conference on Smart Trends in Systems,
Levi, Finland, 2020, pp. 1-5, doi: 10.1109/6GSUM- Security and Sustainability (WorldS4), Lon-
MIT49458.2020.9083880. don, United Kingdom, 2020, pp. 490-493, doi:
10.1109/WorldS450073.2020.9210383.
[18] Y. Yuan, Y. Zhao, B. Zong and S. Parolari, “Poten-
[28] K. B. Letaief, W. Chen, Y. Shi, J. Zhang and Y.
tial key technologies for 6G mobile communications,”
A. Zhang, “The Roadmap to 6G: AI Empowered
in Science China Information Sciences, vol. 63, no.
Wireless Networks,” in IEEE Communications Mag-
183301, pp. 1-19, 2020.
azine, vol. 57, no. 8, pp. 84-90, Aug. 2019, doi:
[19] L. U. Khan, I. Yaqoob, M. Imran, Z. Han and C. S. 10.1109/MCOM.2019.1900271.
Hong, “6G Wireless Systems: A Vision, Architectural [29] S. J. Nawaz, S. K. Sharma, S. Wyne, M. N. Patwary and
Elements, and Future Directions,” in IEEE Access, M. Asaduzzaman, “Quantum Machine Learning for 6G
vol. 8, pp. 147029-147044, 2020, doi: 10.1109/AC- Communication Networks: State-of-the-Art and Vi-
CESS.2020.3015289. sion for the Future,” in IEEE Access, vol. 7, pp. 46317-
46350, 2019, doi: 10.1109/ACCESS.2019.2909490.
[20] A. Dogra, R. K. Jha and S. Jain, “A Survey on be-
yond 5G network with the advent of 6G: Architec- [30] Y. Al-Eryani and E. Hossain, “The D-OMA Method for
ture and Emerging Technologies,” in IEEE Access, doi: Massive Multiple Access in 6G: Performance, Secu-
10.1109/ACCESS.2020.3031234. rity, and Challenges,” in IEEE Vehicular Technology
Magazine, vol. 14, no. 3, pp. 92-99, Sept. 2019, doi:
[21] S. P. Rout, “6G Wireless Communication: Its 10.1109/MVT.2019.2919279.
Vision, Viability, Application, Requirement, Tech-
nologies, Encounters and Research,” 2020 11th [31] E. Yaacoub and M. Alouini, “A Key 6G Challenge and
International Conference on Computing, Commu- Opportunity—Connecting the Base of the Pyramid: A
nication and Networking Technologies (ICCCNT), Survey on Rural Connectivity,” in Proceedings of the
Kharagpur, India, 2020, pp. 1-8, doi: 10.1109/ICC- IEEE, vol. 108, no. 4, pp. 533-582, Apr. 2020, doi:
CNT49239.2020.9225680. 10.1109/JPROC.2020.2976703.

[22] K. David and H. Berndt, “6G Vision and Require- [32] Y. Chen, P. Zhu, G. He, X. Yan, H. Baligh and J.
ments: Is There Any Need for Beyond 5G?,” in IEEE Wu, “From Connected People, Connected Things, to
Vehicular Technology Magazine, vol. 13, no. 3, pp. 72- Connected Intelligence,” 2020 2nd 6G Wireless Sum-
80, Sept. 2018, doi: 10.1109/MVT.2018.2848498. mit (6G SUMMIT), Levi, Finland, 2020, pp. 1-7, doi:
10.1109/6GSUMMIT49458.2020.9083770.
[23] A. Mourad, R. Yang, P. H. Lehne and A. de la Oliva,
[33] T. Hewa, G. Gü r, A. Kalla, M. Ylianttila, A. Bracken
“Towards 6G: Evolution of Key Performance Indica-
and M. Liyanage, “The Role of Blockchain in 6G:
tors and Technology Trends,” 2020 2nd 6G Wireless
Challenges, Opportunities and Research Direc-
Summit (6G SUMMIT), Levi, Finland, 2020, pp. 1-5,
tions,” 2020 2nd 6G Wireless Summit (6G SUMMIT),
doi: 10.1109/6GSUMMIT49458.2020.9083759.
Levi, Finland, 2020, pp. 1-5, doi: 10.1109/6GSUM-
[24] M. J. Piran and D. Y. Suh, “Learning-Driven Wire- MIT49458.2020.9083784.
less Communications, towards 6G,” 2019 Interna- [34] Y. Sun, J. Liu, J. Wang, Y. Cao and N. Kato, “When
tional Conference on Computing, Electronics & Com- Machine Learning Meets Privacy in 6G: A Survey,”
munications Engineering (iCCECE), London, United in IEEE Communications Surveys & Tutorials, doi:
Kingdom, 2019, pp. 219-224, doi: 10.1109/iC- 10.1109/COMST.2020.3011561.
CECE46942.2019.8941882.
[35] H. Chen, K. Tu, J. Li, S. Tang, T. Li and Z. Qing,
[25] A. Yazar and H. Arslan, “A Waveform Parameter “6G Wireless Communications: Security Technolo-
Assignment Framework for 6G With the Role of gies and Research Challenges,” 2020 International
Machine Learning,” in IEEE Open Journal of Vehic- Conference on Urban Engineering and Management
ular Technology, vol. 1, pp. 156-172, 2020, doi: Science (ICUEMS), Zhuhai, China, 2020, pp. 592-595,
10.1109/OJVT.2020.2992502. doi: 10.1109/ICUEMS50872.2020.00130.

© International Telecommunication Union, 2020 133

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[36] Y. Liu, X. Yuan, Z. Xiong, J. Kang, X. Wang and D. Computing, Networking and Communications
Niyato, “Federated learning for 6G communications: (ICNC), Big Island, HI, USA, 2020, pp. 631-635, doi:
Challenges, methods, and future directions,” in China 10.1109/ICNC47757.2020.9049790.
Communications, vol. 17, no. 9, pp. 105-118, Sept.
2020, doi: 10.23919/JCC.2020.09.009. [47] S. Zhang, H. Zhang and L. Song, “Beyond D2D:
Full Dimension UAV-to-Everything Communications
[37] J. Du, C. Jiang, J. WANG, Y. Ren and M. Debbah, “Ma- in 6G,” in IEEE Transactions on Vehicular Technol-
chine Learning for 6G Wireless Networks: Carry- ogy, vol. 69, no. 6, pp. 6592-6602, June 2020, doi:
Forward-Enhanced Bandwidth, Massive Access, and 10.1109/TVT.2020.2984624.
Ultrareliable/Low Latency,” in IEEE Vehicular Tech-
nology Magazine, doi: 10.1109/MVT.2020.3019650. [48] N. Kato, B. Mao, F. Tang, Y. Kawamoto and J. Liu, “Ten
Challenges in Advancing Machine Learning Tech-
[38] C. -X. Wang, J. Huang, H. Wang, X. Gao, X. You and Y. nologies toward 6G,” in IEEE Wireless Communica-
Hao, “6G Wireless Channel Measurements and Mod- tions, vol. 27, no. 3, pp. 96-103, June 2020, doi:
els: Trends and Challenges,” in IEEE Vehicular Tech- 10.1109/MWC.001.1900476.
nology Magazine, doi: 10.1109/MVT.2020.3018436.
[49] X. Liu, T. Xu and I. Darwazeh, “Coexistence of Or-
[39] Y. Zhou, L. Liu, L. Wang, N. Hui, X. Cui, J. Wu, Y. Peng, thogonal and Non-orthogonal Multicarrier Signals in
Y. Qi, C. Xing„ “Service-aware 6G: An intelligent and Beyond 5G Scenarios,” 2020 2nd 6G Wireless Sum-
open network based on the convergence of communi- mit (6G SUMMIT), Levi, Finland, 2020, pp. 1-5, doi:
cation, computing and caching,” in Digital Communi- 10.1109/6GSUMMIT49458.2020.9083780.
cations and Networks, vol. 6, no. 3, pp. 253-260, 2020.
[50] S. Lagen, N. Patriciello and L. Giupponi, “Cellular
[40] T. S. Rappaport et al., “Wireless Communications and Wi-Fi in Unlicensed Spectrum: Competition lead-
and Applications Above 100 GHz: Opportunities ing to Convergence,” 2020 2nd 6G Wireless Sum-
and Challenges for 6G and Beyond,” in IEEE Access, mit (6G SUMMIT), Levi, Finland, 2020, pp. 1-5, doi:
vol. 7, pp. 78729-78757, 2019, doi: 10.1109/AC- 10.1109/6GSUMMIT49458.2020.9083786.
CESS.2019.2921522.
[51] N. H. Mahmood, H. Alves, O. A. Ló pez, M. Shehab, D. P.
[41] X. Huang, J. A. Zhang, R. P. Liu, Y. J. Guo and L. Hanzo, M. Osorio and M. Latva-Aho, “Six Key Features of Ma-
“Airplane-Aided Integrated Networking for 6G Wire- chine Type Communication in 6G,” 2020 2nd 6G Wire-
less: Will It Work?,” in IEEE Vehicular Technology less Summit (6G SUMMIT), Levi, Finland, 2020, pp. 1-
Magazine, vol. 14, no. 3, pp. 84-91, Sept. 2019, doi: 5, doi: 10.1109/6GSUMMIT49458.2020.9083794.
10.1109/MVT.2019.2921244.
[52] M. Katz and I. Ahmed, “Opportunities and Chal-
[42] L. Zhu, Z. Xiao, X. Xia and D. Oliver Wu, “Millimeter- lenges for Visible Light Communications in 6G,”
Wave Communications With Non-Orthogonal 2020 2nd 6G Wireless Summit (6G SUMMIT),
Multiple Access for B5G/6G,” in IEEE Access, vol. Levi, Finland, 2020, pp. 1-5, doi: 10.1109/6GSUM-
7, pp. 116123-116132, 2019, doi: 10.1109/AC- MIT49458.2020.9083805.
CESS.2019.2935169.
[53] C. Huang et al., “Holographic MIMO Surfaces for
[43] T. Huang, W. Yang, J. Wu, J. Ma, X. Zhang and D. 6G Wireless Networks: Opportunities, Challenges,
Zhang, “A Survey on Green 6G Network: Architecture and Trends,” in IEEE Wireless Communications, doi:
and Technologies,” in IEEE Access, vol. 7, pp. 175758- 10.1109/MWC.001.1900534.
175768, 2019, doi: 10.1109/ACCESS.2019.2957648.
[54] C. Liu, W. Feng, Y. Chen, C. Wang and N. Ge, “Cell-Free
[44] F. Tang, Y. Kawamoto, N. Kato and J. Liu, “Future In- Satellite-UAV Networks for 6G Wide-Area Internet of
telligent and Secure Vehicular Network Toward 6G: Things,” in IEEE Journal on Selected Areas in Commu-
Machine-Learning Approaches,” in Proceedings of the nications, doi: 10.1109/JSAC.2020.3018837.
IEEE, vol. 108, no. 2, pp. 292-307, Feb. 2020, doi:
10.1109/JPROC.2019.2954595. [55] H. Hashida, Y. Kawamoto and N. Kato, “Intelli-
gent Re lecting Surface Placement Optimization
[45] E. Basar, “Recon igurable Intelligent Surface-Based in Air-Ground Communication Networks To-
Index Modulation: A New Beyond MIMO Paradigm ward 6G,” in IEEE Wireless Communications, doi:
for 6G,” in IEEE Transactions on Communications, 10.1109/MWC.001.2000142.
vol. 68, no. 5, pp. 3187-3196, May 2020, doi:
10.1109/TCOMM.2020.2971486. [56] R. K. Saha, “Licensed Countrywide Full-Spectrum
Allocation: A New Paradigm for Millimeter-Wave
[46] M. Yu, A. Tang, X. Wang and C. Han, “Joint Schedul- Mobile Systems in 5G/6G Era,” in IEEE Access, vol.
ing and Power Allocation for 6G Terahertz Mesh 8, pp. 166612-166629, 2020, doi: 10.1109/AC-
Networks,” 2020 International Conference on CESS.2020.3023342.

134 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[57] J. A. Hodge, K. V. Mishra and A. I. Zaghloul, “Intel- Intelligent 6G Networks,” in IEEE Network, doi:
ligent Time-Varying Metasurface Transceiver for In- 10.1109/MNET.011.2000195.
dex Modulation in 6G Wireless Networks,” in IEEE
[68] H. Huang, S. Hu, T. Yang and C. W. Yuan, “Full Du-
Antennas and Wireless Propagation Letters, doi:
plex Non-orthogonal Multiple Access with Layers-
10.1109/LAWP.2020.3025333.
based Optimized Mobile Relays Subsets Algorithm in
[58] A. Celik, A. Chaaban, B. Shihada and M. -S. B5G/6G Ubiquitous Networks,” in IEEE Internet of
Alouini, “Topology Optimization for 6G Net- Things Journal, doi: 10.1109/JIOT.2020.3033553.
works: A Network Information-Theoretic Ap- [69] Y. Liu, X. Wang, G. Boudreau, A. B. Sediq and H. Abou-
proach,” in IEEE Vehicular Technology Magazine, doi: zeid, “A Multi-Dimensional Intelligent Multiple Ac-
10.1109/MVT.2020.3017152. cess Technique for 5G Beyond and 6G Wireless Net-
[59] M. Kishk, A. Bader and M. -S. Alouini, “Aerial Base works,” in IEEE Transactions on Wireless Communi-
Station Deployment in 6G Cellular Networks Using cations, doi: 10.1109/TWC.2020.3032631.
Tethered Drones: The Mobility and Endurance Trade- [70] A. Mukherjee, P. Goswami, M. A. Khan, L. Manman,
off,” in IEEE Vehicular Technology Magazine, doi: L. Yang and P. Pillai, “Energy Ef icient Resource Allo-
10.1109/MVT.2020.3017885. cation strategy in Massive IoT for Industrial 6G Ap-
plications,” in IEEE Internet of Things Journal, doi:
[60] W. Lu et al., “SWIPT Cooperative Spectrum 10.1109/JIOT.2020.3035608.
Sharing for 6G-Enabled Cognitive IoT Net-
work,” in IEEE Internet of Things Journal, doi: [71] I. Yildirim, A. Uyrus and E. Basar, “Modeling and
10.1109/JIOT.2020.3026730. Analysis of Recon igurable Intelligent Surfaces for In-
door and Outdoor Applications in Future Wireless
[61] N. Chi, Y. Zhou, Y. Wei and F. Hu, “Visible light Networks,” in IEEE Transactions on Communications,
communication in 6G: Advances, challenges, and doi: 10.1109/TCOMM.2020.3035391.
prospects,” in IEEE Vehicular Technology Magazine,
doi: 10.1109/MVT.2020.3017153. [72] S. Wan, J. Hu, C. Chen, A. Jolfaei, S. Mumtaz and Q.
Pei, “Fair-Hierarchical Scheduling for Diversi ied Ser-
[62] J. Liu, W. Liu, X. Hou, Y. Kishiyama, L. Chen and T. vices in Space, Air and Ground for 6G-Dense Internet
Asai, “Non-Orthogonal Waveform (NOW) for 5G Evo- of Things,” in IEEE Transactions on Network Science
lution and 6G,” 2020 IEEE 31st Annual International and Engineering, doi: 10.1109/TNSE.2020.3035616.
Symposium on Personal, Indoor and Mobile Radio
[73] S. Han et al., “Arti icial-Intelligence-Enabled Air In-
Communications, London, United Kingdom, 2020, pp.
terface for 6G: Solutions, Challenges, and Standard-
1-6, doi: 10.1109/PIMRC48278.2020.9217361.
ization Impacts,” in IEEE Communications Maga-
[63] A. Vanelli-Coralli, A. Guidotti, T. Foggi, G. zine, vol. 58, no. 10, pp. 73-79, Oct. 2020, doi:
Colavolpe and G. Montorsi, “5G and Beyond 5G 10.1109/MCOM.001.2000218.
Non-Terrestrial Networks: trends and research [74] M. Matinmikko-Blue, S. Yrjö lä and P. Ahokan-
challenges,” 2020 IEEE 3rd 5G World Forum gas, “Spectrum Management in the 6G Era: The
(5GWF), Bangalore, India, 2020, pp. 163-169, Role of Regulation and Spectrum Sharing,” 2020
doi: 10.1109/5GWF49715.2020.9221119. 2nd 6G Wireless Summit (6G SUMMIT), Levi,
[64] L. Zhen, A. K. Bashir, K. Yu, Y. D. Al-Otaibi, C. Finland, 2020, pp. 1-5, doi: 10.1109/6GSUM-
H. Foh and P. Xiao, “Energy-Ef icient Random Ac- MIT49458.2020.9083851.
cess for LEO Satellite-Assisted 6G Internet of Re- [75] R. Sha in, L. Liu, V. Chandrasekhar, H. Chen,
mote Things,” in IEEE Internet of Things Journal, doi: J. Reed and J. C. Zhang, “Arti icial Intelligence-
10.1109/JIOT.2020.3030856. Enabled Cellular Networks: A Critical Path to
Beyond-5G and 6G,” in IEEE Wireless Communica-
[65] S. Liao, J. Wu, J. Li and K. Konstantin, “Information- tions, vol. 27, no. 2, pp. 212-217, Apr. 2020, doi:
Centric Massive IoT based Ubiquitous Connected 10.1109/MWC.001.1900323.
VR/AR in 6G: A Proposed Caching Consensus Ap-
proach,” in IEEE Internet of Things Journal, doi: [76] J. Zhu, M. Zhao, S. Zhang and W. Zhou, “Explor-
10.1109/JIOT.2020.3030718. ing the road to 6G: ABC — foundation for in-
telligent mobile networks,” in China Communica-
[66] R. Alghamdi et al., “Intelligent Surfaces for 6G Wire- tions, vol. 17, no. 6, pp. 51-67, June 2020, doi:
less Networks: A Survey of Optimization and Per- 10.23919/JCC.2020.06.005.
formance Analysis Techniques,” in IEEE Access, doi:
10.1109/ACCESS.2020.3031959. [77] V. Ziegler, H. Viswanathan, H. Flinck, M. Hoffmann,
V. Rä isä nen and K. Hä tö nen, “6G Architecture to Con-
[67] H. Yang, A. Alphones, Z. Xiong, D. Niyato, J. nect the Worlds,” in IEEE Access, vol. 8, pp. 173508-
Zhao and K. Wu, “Arti icial Intelligence-Enabled 173520, 2020, doi: 10.1109/ACCESS.2020.3025032.

© International Telecommunication Union, 2020 135

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[78] H. Li, F. Fang, Z. Ding, “Joint resource allocation for [90] Z. E. Ankarali, B. Pekö z and H. Arslan, “Flexible Ra-
hybrid NOMA-assisted MEC in 6G networks,” in Dig- dio Access Beyond 5G: A Future Projection on Wave-
ital Communications and Networks, vol. 6, no. 3, pp. form, Numerology, and Frame Design Principles,” in
241-252, 2020. IEEE Access, vol. 5, pp. 18295-18309, 2017, doi:
10.1109/ACCESS.2017.2684783.
[79] H. Xu, P. V. Klaine, O. Onireti, B. Cao, M. Imran,
L. Zhang, “Blockchain-enabled resource management [91] M. M. Sahin and H. Arslan, “Multi-functional Coex-
and sharing for 6G communications,” in Digital Com- istence of Radar-Sensing and Communication Wave-
munications and Networks, vol. 6, no. 3, pp. 261-269, forms,” in IEEE Vehicular Technology Conference
2020. (VTC-Fall), Victoria, Canada, 2020.
[80] Y. Wei, M. Peng, Y. Liu, “Intent-based networks for
[92] A. A. Sabah and H. Arslan, “NOMA for Multi-
6G: Insights and challenges,” in Digital Communica-
Numerology OFDM Systems,” in Hindawi Wireless
tions and Networks, vol. 6, no. 3, pp. 270-280, 2020.
Communications and Mobile Computing, vol. 2018,
[81] M. Wang, T. Zhu, T. Zhang, J. Zhang, S. Yu, W. Zhou, pp. 1-9, 2018, doi:10.1155/2018/8514314.
“Security and privacy in 6G networks: New areas and
new challenges,” in Digital Communications and Net- [93] A. Tusha, S. Doğ an and H. Arslan, “A Hybrid
works, vol. 6, no. 3, pp. 281-291, 2020. Downlink NOMA With OFDM and OFDM-IM for Be-
yond 5G Wireless Networks,” in IEEE Signal Pro-
[82] Y. Fu, K. N. Doan, T. Q. S. Quek, “On recommendation- cessing Letters, vol. 27, pp. 491-495, 2020, doi:
aware content caching for 6G: An arti icial intelli- 10.1109/LSP.2020.2979059.
gence and optimization empowered paradigm,” in
Digital Communications and Networks, vol. 6, no. 3, [94] M. M. Şahin and H. Arslan, “Waveform-Domain
pp. 304-311, 2020. NOMA: The Future of Multiple Access,” 2020
IEEE International Conference on Communi-
[83] Y. Chen, W. Liu, Z. Niu, Z. Feng, Q. Hu, T. Jiang, “Per- cations Workshops (ICC Workshops), Dublin,
vasive intelligent endogenous 6G wireless systems: Ireland, 2020, pp. 1-6, doi: 10.1109/ICCWork-
Prospects, theories and key technologies,” in Digital shops49005.2020.9145077.
Communications and Networks, vol. 6, no. 3, pp. 312-
320, 2020. [95] A. Maatouk, E. Çalışkan, M. Koca, M. Assaad, G. Gui
[84] International Telecommunication Union (ITU), “IMT and H. Sari, “Frequency-Domain NOMA With Two Sets
Vision – Framework and Overall Objectives of the Fu- of Orthogonal Signal Waveforms,” in IEEE Communi-
ture Development of IMT for 2020 and Beyond,” ITU cations Letters, vol. 22, no. 5, pp. 906-909, May 2018,
Publications, M.2083-0, 2015. doi: 10.1109/LCOMM.2018.2810118.

[85] International Telecommunication Union (ITU), “Fi- [96] M. B. Çelebi and H. Arslan, “Theoretical Analysis of
nal Acts World Radiocommunication Conference the Co-Existence of LTE-A Signals and Design of an
2019 (WRC-19),” ITU Publications, 2019. ML-SIC Receiver,” in IEEE Transactions on Wireless
Communications, vol. 14, no. 8, pp. 4626-4639, Aug.
[86] A. Ghosh, A. Maeder, M. Baker and D. Chandramouli, 2015, doi: 10.1109/TWC.2015.2424244.
“5G Evolution: A View on 5G Cellular Technol-
ogy Beyond 3GPP Release 15,” in IEEE Access, vol. [97] A. M. Jaradat, J. M. Hamamreh and H. Arslan, “Mod-
7, pp. 127639-127651, 2019, doi: 10.1109/AC- ulation Options for OFDM-Based Waveforms: Clas-
CESS.2019.2939938. si ication, Comparison, and Future Directions,” in
IEEE Access, vol. 7, pp. 17263-17278, 2019, doi:
[87] P. Kumari, S. A. Vorobyov and R. W. Heath, “Adap-
10.1109/ACCESS.2019.2895958.
tive Virtual Waveform Design for Millimeter-Wave
Joint Communication–Radar,” in IEEE Transactions [98] P. Yang, Y. Xiao, Y. L. Guan, M. Di Renzo, S. Li
on Signal Processing, vol. 68, pp. 715-730, 2020, doi: and L. Hanzo, “Multidomain Index Modulation for
10.1109/TSP.2019.2956689. Vehicular and Railway Communications: A Survey
[88] A. Yazar and H. Arslan, “Flexible Multi-Numerology of Novel Techniques,” in IEEE Vehicular Technology
Systems for 5G New Radio,” in River Publishers Jour- Magazine, vol. 13, no. 3, pp. 124-134, Sept. 2018, doi:
nal of Mobile Multimedia, vol. 14 no.4, pp. 367-394, 10.1109/MVT.2018.2814023.
2018.
[99] S. Doğ an, A. Tusha and H. Arslan, “NOMA With In-
[89] A. B. Kihero, M. S. J. Solaija and H. Arslan, “Inter- dex Modulation for Uplink URLLC Through Grant-
Numerology Interference for Beyond 5G,” in IEEE Free Access,” in IEEE Journal of Selected Topics in
Access, vol. 7, pp. 146512-146523, 2019, doi: Signal Processing, vol. 13, no. 6, pp. 1249-1257, Oct.
10.1109/ACCESS.2019.2946084. 2019, doi: 10.1109/JSTSP.2019.2913981.

136 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[100] A. Yazar and H. Arslan, “A Flexibility Metric and [111] D. Sattar and A. Matrawy, “Optimal Slice Allo-
Optimization Methods for Mixed Numerologies in 5G cation in 5G Core Networks,” in IEEE Networking
and Beyond,” in IEEE Access, vol. 6, pp. 3755-3764, Letters, vol. 1, no. 2, pp. 48-51, June 2019, doi:
2018, doi: 10.1109/ACCESS.2018.2795752. 10.1109/LNET.2019.2908351.
[101] A. F. Demir and H. Arslan, “Inter-Numerology Inter- [112] D. A. Chekired, M. A. Togou, L. Khoukhi and A. Ksen-
ference Management With Adaptive Guards: A Cross- tini, “5G-Slicing-Enabled Scalable SDN Core Network:
Layer Approach,” in IEEE Access, vol. 8, pp. 30378- Toward an Ultra-Low Latency of Autonomous Driving
30386, 2020, doi: 10.1109/ACCESS.2020.2972287. Service,” in IEEE Journal on Selected Areas in Com-
munications, vol. 37, no. 8, pp. 1769-1782, Aug. 2019,
[102] E. Memisoglu, A. B. Kihero, E. Basar and H. Ar- doi: 10.1109/JSAC.2019.2927065.
slan, “Guard Band Reduction for 5G and Beyond Mul-
tiple Numerologies,” in IEEE Communications Let- [113] Q. Zhang, H. Sun, Z. Wei and Z. Feng, “Sensing and
ters, vol. 24, no. 3, pp. 644-647, Mar. 2020, doi: Communication Integrated System for Autonomous
10.1109/LCOMM.2019.2963311. Driving Vehicles,” IEEE INFOCOM 2020 - IEEE
Conference on Computer Communications Work-
[103] A. Yazar and H. Arslan, “Reliability Enhancement shops (INFOCOM WKSHPS), Toronto, ON, Canada,
in Multi-Numerology Based 5G New Radio Using INI- 2020, pp. 1278-1279, doi: 10.1109/INFOCOMWK-
Aware Scheduling,” in EURASIP Journal on Wireless SHPS50562.2020.9162963.
Communications and Networking, vol. 2019 no. 110,
pp. 1-14, 2019. [114] H. Turkmen, M. S. J. Solaija, H. M. Furqan, H. Arslan,
“Generalized Radio Environment Monitoring for Next
[104] M. J. Colella, J. N. Martin and F. Akyildiz, “The Generation Wireless Networks,” arXiv:2008.06203
HALO networkTM ,” in IEEE Communications Maga- [eess.SP], Aug. 2020.
zine, vol. 38, no. 6, pp. 142-148, June 2000, doi:
10.1109/35.846086. [115] W. Lee and I. F. Akyildiz, “Optimal spectrum sens-
ing framework for cognitive radio networks,” in IEEE
[105] T. Han, X. Ge, L. Wang, K. S. Kwak, Y. Han and Transactions on Wireless Communications, vol. 7, no.
X. Liu, “5G Converged Cell-Less Communications 10, pp. 3845-3857, October 2008, doi: 10.1109/T-
in Smart Cities,” in IEEE Communications Maga- WC.2008.070391.
zine, vol. 55, no. 3, pp. 44-50, Mar. 2017, doi:
[116] S. Yarkan and H. Arslan, “Exploiting location aware-
10.1109/MCOM.2017.1600256CM.
ness toward improved wireless system design in
[106] L. Wang, T. Han, Q. Li, J. Yan, X. Liu and D. Deng, cognitive radio,” in IEEE Communications Maga-
“Cell-Less Communications in 5G Vehicular Networks zine, vol. 46, no. 1, pp. 128-136, Jan. 2008, doi:
Based on Vehicle-Installed Access Points,” in IEEE 10.1109/MCOM.2008.4427241.
Wireless Communications, vol. 24, no. 6, pp. 64-71, [117] H. F. Ates, S. M. Hashir, T. Baykas and B. K. Gun-
Dec. 2017, doi: 10.1109/MWC.2017.1600401. turk, “Path Loss Exponent and Shadowing Factor Pre-
[107] M. S. J. Solaija, H. Salman, A. B. Kihero, M. I. Saglam, diction From Satellite Images Using Deep Learning,”
H. Arslan, “Generalized Coordinated Multipoint in IEEE Access, vol. 7, pp. 101366-101375, 2019, doi:
Framework for 5G and Beyond,” arXiv:2008.06343 10.1109/ACCESS.2019.2931072.
[eess.SP], Aug. 2020. [118] D. Sabella et al., “A lexible and recon igurable 5G
[108] M. S. Ali, E. Hossain, A. Al-Dweik and D. I. Kim, networking architecture based on context and con-
“Downlink Power Allocation for CoMP-NOMA in tent information,” 2017 European Conference on Net-
Multi-Cell Networks,” in IEEE Transactions on Com- works and Communications (EuCNC), Oulu, 2017, pp.
munications, vol. 66, no. 9, pp. 3982-3998, Sept. 2018, 1-6, doi: 10.1109/EuCNC.2017.7980669.
doi: 10.1109/TCOMM.2018.2831206. [119] Z. Feng, Z. Fang, Z. Wei, X. Chen, Z. Quan and D. Ji,
[109] J. Ding and J. Cai, “Two-Side Coalitional Match- “Joint radar and communication: A survey,” in China
ing Approach for Joint MIMO-NOMA Clustering and Communications, vol. 17, no. 1, pp. 1-27, Jan. 2020,
BS Selection in Multi-Cell MIMO-NOMA Systems,” doi: 10.23919/JCC.2020.01.001.
in IEEE Transactions on Wireless Communications, [120] G. Naik, J. Park, J. Ashdown, W. Lehr, “Next Gener-
vol. 19, no. 3, pp. 2006-2021, Mar. 2020, doi: ation Wi-Fi and 5G NR-U in the 6 GHz Bands: Oppor-
10.1109/TWC.2019.2961654. tunities & Challenges,” arXiv:2006.16534, 2020.
[110] B. Han, J. Lianghai and H. D. Schotten, “Slice as [121] H. B. Yilmaz, T. Tugcu, F. Alagö z and S. Bayhan, “Ra-
an Evolutionary Service: Genetic Optimization for dio environment map as enabler for practical cogni-
Inter-Slice Resource Management in 5G Networks,” tive radio networks,” in IEEE Communications Mag-
in IEEE Access, vol. 6, pp. 33137-33147, 2018, doi: azine, vol. 51, no. 12, pp. 162-169, Dec. 2013, doi:
10.1109/ACCESS.2018.2846543. 10.1109/MCOM.2013.6685772.

© International Telecommunication Union, 2020 137

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[122] A. Imran, A. Zoha and A. Abu-Dayya, “Challenges in [132] Z. Chang, L. Lei, Z. Zhou, S. Mao and T. Ristaniemi,
5G: how to empower SON with big data for enabling “Learn to Cache: Machine Learning for Network Edge
5G,” in IEEE Network, vol. 28, no. 6, pp. 27-33, Nov.- Caching in the Big Data Era,” in IEEE Wireless Com-
Dec. 2014, doi: 10.1109/MNET.2014.6963801. munications, vol. 25, no. 3, pp. 28-35, June 2018, doi:
10.1109/MWC.2018.1700317.
[123] Q. Mao, F. Hu and Q. Hao, “Deep Learning for In-
telligent Wireless Networks: A Comprehensive Sur- [133] E. Basar, M. Di Renzo, J. De Rosny, M. Debbah,
vey,” in IEEE Communications Surveys & Tutorials, M. Alouini and R. Zhang, “Wireless Communica-
vol. 20, no. 4, pp. 2595-2621, Fourthquarter 2018, tions Through Recon igurable Intelligent Surfaces,” in
doi: 10.1109/COMST.2018.2846401. IEEE Access, vol. 7, pp. 116753-116773, 2019, doi:
10.1109/ACCESS.2019.2935192.
[124] C. Zhang, P. Patras and H. Haddadi, “Deep Learn-
ing in Mobile and Wireless Networking: A Survey,” [134] M. D. Renzo et al., “Smart Radio Environments Em-
in IEEE Communications Surveys & Tutorials, vol. powered by Recon igurable Intelligent Surfaces: How
21, no. 3, pp. 2224-2287, thirdquarter 2019, doi: it Works, State of Research, and Road Ahead,” in IEEE
10.1109/COMST.2019.2904897. Journal on Selected Areas in Communications, doi:
10.1109/JSAC.2020.3007211.
[125] N. C. Luong et al., “Applications of Deep Reinforce-
ment Learning in Communications and Networking: [135] C. Xu, L. Yang and P. Zhang, “Practical Backscatter
A Survey,” in IEEE Communications Surveys & Tutori- Communication Systems for Battery-Free Internet of
als, vol. 21, no. 4, pp. 3133-3174, Fourthquarter 2019, Things: A Tutorial and Survey of Recent Research,” in
doi: 10.1109/COMST.2019.2916583. IEEE Signal Processing Magazine, vol. 35, no. 5, pp.
16-27, Sept. 2018, doi: 10.1109/MSP.2018.2848361.
[126] Y. Sun, M. Peng, Y. Zhou, Y. Huang and S.
[136] N. Van Huynh, D. T. Hoang, X. Lu, D. Niy-
Mao, “Application of Machine Learning in Wire-
ato, P. Wang and D. I. Kim, “Ambient Backscat-
less Networks: Key Techniques and Open Issues,”
ter Communications: A Contemporary Survey,” in
in IEEE Communications Surveys & Tutorials, vol.
IEEE Communications Surveys & Tutorials, vol. 20,
21, no. 4, pp. 3072-3108, Fourthquarter 2019, doi:
no. 4, pp. 2889-2922, Fourthquarter 2018, doi:
10.1109/COMST.2019.2924243.
10.1109/COMST.2018.2841964.
[127] M. Chen, U. Challita, W. Saad, C. Yin and M.
[137] R. Long, Y. Liang, H. Guo, G. Yang and R. Zhang,
Debbah, “Arti icial Neural Networks-Based Machine
”Symbiotic Radio: A New Communication Paradigm
Learning for Wireless Networks: A Tutorial,” in
for Passive Internet of Things,” in IEEE Internet of
IEEE Communications Surveys & Tutorials, vol. 21,
Things Journal, vol. 7, no. 2, pp. 1350-1363, Feb. 2020,
no. 4, pp. 3039-3071, Fourthquarter 2019, doi:
doi: 10.1109/JIOT.2019.2954678.
10.1109/COMST.2019.2926625.
[138] Q. Zhang, Y. Liang and H. V. Poor, ”Symbi-
[128] J. Wang, C. Jiang, H. Zhang, Y. Ren, K. Chen otic Radio: A New Application of Large Intelli-
and L. Hanzo, “Thirty Years of Machine Learning: gent Surface/Antennas (LISA),” 2020 IEEE Wire-
The Road to Pareto-Optimal Wireless Networks,” less Communications and Networking Conference
in IEEE Communications Surveys & Tutorials, doi: (WCNC), Seoul, Korea (South), 2020, pp. 1-6, doi:
10.1109/COMST.2020.2965856. 10.1109/WCNC45663.2020.9120455.
[129] X. Wang, Y. Han, V. C. M. Leung, D. Niyato, X. [139] S. Buzzi, C. I, T. E. Klein, H. V. Poor, C. Yang
Yan and X. Chen, “Convergence of Edge Comput- and A. Zappone, “A Survey of Energy-Ef icient Tech-
ing and Deep Learning: A Comprehensive Survey,” niques for 5G Networks and Challenges Ahead,”
in IEEE Communications Surveys & Tutorials, vol. in IEEE Journal on Selected Areas in Communica-
22, no. 2, pp. 869-904, Secondquarter 2020, doi: tions, vol. 34, no. 4, pp. 697-709, Apr. 2016, doi:
10.1109/COMST.2020.2970550. 10.1109/JSAC.2016.2550338.
[130] C. Jiang, H. Zhang, Y. Ren, Z. Han, K. Chen and [140] J. Huang, C. Xing and C. Wang, “Simultaneous Wire-
L. Hanzo, “Machine Learning Paradigms for Next- less Information and Power Transfer: Technologies,
Generation Wireless Networks,” in IEEE Wireless Applications, and Research Challenges,” in IEEE Com-
Communications, vol. 24, no. 2, pp. 98-105, Apr. 2017, munications Magazine, vol. 55, no. 11, pp. 26-32, Nov.
doi: 10.1109/MWC.2016.1500356WC. 2017, doi: 10.1109/MCOM.2017.1600806.

[131] R. Li et al., “Intelligent 5G: When Cellular Networks [141] T. D. Ponnimbaduge Perera, D. N. K. Jayakody,
Meet Arti icial Intelligence,” in IEEE Wireless Com- S. K. Sharma, S. Chatzinotas and J. Li, “Simulta-
munications, vol. 24, no. 5, pp. 175-183, Oct. 2017, neous Wireless Information and Power Transfer
doi: 10.1109/MWC.2017.1600304WC. (SWIPT): Recent Advances and Future Challenges,”

138 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

in IEEE Communications Surveys & Tutorials, vol. AUTHORS

20, no. 1, pp. 264-302, Firstquarter 2018, doi:
A. Yazar received his B.Sc. degree in
10.1109/COMST.2017.2783901.
electrical engineering from Eskise-
[142] J. M. Hamamreh, H. M. Furqan and H. Arslan, “Clas- hir Osmangazi University, Eskisehir,
si ications and Applications of Physical Layer Secu- Turkey in 2011, M.Sc. degree in elec-
rity Techniques for Con identiality: A Comprehen- trical engineering from Bilkent Uni-
sive Survey,” in IEEE Communications Surveys & Tu- versity, Ankara, Turkey in 2013, and
torials, vol. 21, no. 2, pp. 1773-1828, Secondquarter Ph.D. degree in electrical engineer-
2019, doi: 10.1109/COMST.2018.2878035. ing from Istanbul Medipol Univer-
sity, Istanbul, Turkey in 2020. He
[143] J. M. Hamamreh, M. Yusuf, T. Baykas and H. is currently general coordinator as
Arslan, “Cross MAC/PHY layer security design a member of the Communications,
using ARQ with MRC and adaptive modulation,” Signal Processing, and Networking
2016 IEEE Wireless Communications and Net- Center (CoSiNC) at Istanbul Medipol University. His cur-
working Conference, Doha, 2016, pp. 1-7, doi: rent research interests are lexible waveform design, ra-
10.1109/WCNC.2016.7564987. dio resource management techniques, and the role of ma-
chine learning in wireless communications systems.
[144] S. Dwivedi, M. Zoli, A. N. Barreto, P. Sen and G. Fet-
tweis, “Secure Joint Communications and Sensing us- S. Doğan Tusha received the B.Sc.
ing Chirp Modulation,” 2020 2nd 6G Wireless Sum- degree in electronics and telecom-
mit (6G SUMMIT), Levi, Finland, 2020, pp. 1-5, doi: munication engineering from Ko-
10.1109/6GSUMMIT49458.2020.9083884. caeli University, Kocaeli, Turkey, in
[145] Y. Hu and R. Zhang, “A Spatiotemporal Approach for 2015, and the Ph.D. degree in electri-
Secure Crowdsourced Radio Environment Map Con- cal and electronics engineering from
struction,” in IEEE/ACM Transactions on Networking, Istanbul Medipol University, Istan-
doi: 10.1109/TNET.2020.2992939. bul, Turkey, in 2020. She is currently
a post-doctoral researcher in the
Communications, Signal Processing,
and Networking Center (CoSiNC) at
Istanbul Medipol University, Istan-
bul, Turkey. Her research interests include index mod-
ulation, millimeter-wave frequency bands, nonorthogo-
nal multiple accessing (NOMA), and random access tech-
niques for next-generation wireless networks.

H. Arslan (IEEE Fellow, IEEE Dis-

tinguished Lecturer) received his BS
degree from the Middle East Tech-
nical University (METU), Ankara,
Turkey in 1992; his MS and Ph.D.
degrees were received respectively
in 1994 and 1998 from Southern
Methodist University (SMU), Dallas,
TX. From January 1998 to August
2002, he was with the research
group of Ericsson, where he was
involved with several projects re-
lated to 2G and 3G wireless communication systems.
Since August 2002, he has been with the Electrical Engi-
neering Department, at the University of South Florida,
where he is a Professor. In December 2013, he joined
Istanbul Medipol University to found the Engineering
College, where he has worked as the Dean of the School
of Engineering and Natural Sciences. In addition, he has
worked as a part-time consultant for various companies
and institutions including Anritsu Company and The
Scienti ic and Technological Research Council of Turkey.

© International Telecommunication Union, 2020 139

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Dr. Arslan conducts research in wireless systems, with

emphasis on the physical and medium access layers of
communications. His current research interests are on 5G
and beyond radio access technologies, physical layer se-
curity, interference management (avoidance, awareness,
and cancellation), cognitive radio, multi-carrier wireless
technologies (beyond OFDM), dynamic spectrum access,
coexistence issues, non-terrestial communications (High
Altitude Platforms), joint radar (sensing) and commu-
nication designs. Dr. Arslan has been collaborating
extensively with key national and international industrial
partners and his research has generated signi icant
interest in companies such as InterDigital, Anritsu, NTT
DoCoMo, Raytheon, Honeywell, Keysight technologies.
Collaborations and feedback from industry partners has
signi icantly in luenced his research. In addition to his
research activities, Dr. Arslan has also contributed to
wireless communication education. He has integrated
the outcomes of his research into education which lead
him to develop a number of courses at the University of
South Florida. He has developed a unique “Wireless Sys-
tems Laboratory” course (funded by the National Science
Foundation and Keysight technologies) where he was
able to teach not only the theory but also the practical
aspects of wireless communication system with the most
contemporary test and measurement equipment.

Dr. Arslan has served as general chair, technical pro-

gram committee chair, session and symposium organizer,
workshop chair, and technical program committee mem-
ber in several IEEE conferences. He is currently a member
of the editorial board for the IEEE Surveys and Tutorials
and the Sensors Journal. He has also served as a member
of the editorial board for the IEEE Transactions on Com-
munications, the IEEE Transactions on Cognitive Com-
munications and Networking (TCCN), and several other
scholarly journals by Elsevier, Hindawi, and Wiley Pub-
lishing.

140 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

ON THE EVOLUTION OF INFRASTRUCTURE SHARING IN MOBILE

NETWORKS: A SURVEY

Lorela Cano1 , Antonio Capone2 , Brunilde Sansò3

1,2
Politecnico di Milano, Milan, Piazza Leonardo da Vinci, 32, 20133, Milano MI, Italy, 3 Polytechnique Montréal,
2500 Chemin de Polytechnique, Montréal, QC H3T 1J4, Canada,

NOTE: Corresponding author: Lorela Cano ([email protected])

Abstract – Infrastructure sharing for mobile networks has been a prolific research topic for more than three decades
now. The key driver for Mobile Network Operators to share their network infrastructure is cost reduction. Spectrum
sharing is often studied alongside infrastructure sharing although on its own it is a vast research topic outside the scope
of this survey. Instead, in this survey we aim to provide a complete picture of infrastructure sharing both over time
and in terms of research branches that have stemmed from it such as performance evaluation, resource management
etc. We also put an emphasis on the relation between infrastructure sharing and the decoupling of infrastructure from
services, wireless network virtualization and multi-tenancy in 5G networks. Such a relation reflects the evolution of
infrastructure sharing over time and how it has become a commercial reality in the context of 5G.
Keywords – 5G, infrastructure sharing, mobile networks, multi-tenancy, spectrum sharing, wireless network virtu-
alization

1. INTRODUCTION among them and hence make their business more prof-
itable. In these lines, infrastructure sharing has accom-
Infrastructure sharing in mobile networks is a multi-
panied the technology migrations from 2G to 3G and
faceted problem involving not only academic and indus-
from 3G to 4G due to the high upfront cost met by
trial research entities but also national and international
MNOs during these migrations. In turn, in 5G networks,
regulatory entities [51, 52, 65, 66], standardization bod-
infrastructure sharing, besides from being a means for
ies [1–4] and vendors [45,107]. In essence, infrastructure
cost-reduction, it is also an important pillar of the 5G
sharing in mobile networks is the shared use of existing
architecture. Another paradigm strongly linked to in-
or jointly deployed network infrastructure among mul-
frastructure sharing is spectrum sharing. The need for
tiple Mobile Network Operators (MNOs).
spectrum sharing comes from spectrum being an intrin-
Based on which network elements (nodes) MNOs agree sically scarce resource, even more so in the context of
to/can share, there are two main types of sharing: pas- 5G, given its target throughputs. However, spectrum
sive and active, the latter comprising the former. Pas- sharing alone is a really vast research topic and will be
sive sharing (also referred to as site sharing or co- outside the scope of this survey unless combined with
location [49]) implies the sharing of the site physical infrastructure sharing.
space and of the non-active elements on the site (such
as shelter, cabinet, mast, etc. [49, 104]). Instead, active What’s more, in this paper we will also address some
sharing extends to active elements of the Radio Access literature on Wireless Network Virtualization (WNV)
Network (RAN) (such as antennas, Base Transceiver [91] and network slicing (enabling multi-tenancy) in the
Stations/Base Station Controller for 2G, Node B/Radio context of 5G [6], since both are based on infrastruc-
Network Controller for 3G, eNode B for 4G, and gN- ture and spectrum sharing. Conversely, WNV and net-
odeB for 5G) and part of the core nodes (in fact, core work slicing can be seen as enablers for infrastructure
node elements related to user billing and accounting are and spectrum sharing. Besides, another concept closely
not shared). related to infrastructure and spectrum sharing is that
of the decoupling of infrastructure from services, which
The phenomenon of infrastructure sharing has disrupted was envisioned by some of the early literature on in-
the business model of a conventional MNO, that is, an frastructure sharing (see Section 2). The concept has
MNO which is by itself responsible for (i) purchasing a been further carried out in the context of WNV and
spectrum license, (ii) deploying and managing the net- then in the context of network slicing. In fact, the dif-
work infrastructure, (iii) tailoring services for their sub- ferent research efforts on introducing Software-Defined
scribers (e.g., voice, data, etc.) and (iv) handling their Networking (SDN), virtualization in general and Net-
billing and accounting. The main reason for MNOs to work Functions Virtualization (NFV) in particular into
share infrastructure is to divide the infrastructure cost mobile networks seem to have converged into the 5G

© International Telecommunication Union, 2020 141

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

architecture as enablers for network slicing. 3GPP 3rd Generation Partnership Project
BS Base Station
Infrastructure sharing in this broader sense has been a C-RAN Cloud Radio Access Network
very prolific research topic over the last three decades. DCS Digital Cellular System
Samdanis et al. in [125] provide a compelling analysis of EDGE Enhanced Data rates for GSM Evolution
the path from infrastructure sharing to multi-tenancy. GERAN GSM EDGE RAN
However, to the best of our knowledge, our survey is GSM Global System for Mobile communications
the first1 comprehensive study on how the infrastructure IaaS Infrastructure as a Service
sharing topic in mobile networks has evolved over time, InP Infrastructure Provider
i.e., with the advent of the different mobile network gen- IoT Internet of Things
IP Integer Programming
erations, and which research branches have spurred from
JV Joint Venture
this topic. Reviewing this evolution is particularly im- MLFG Multi-Leader-Follower Game
portant now that networking slicing is being introduced mmWave millimiter Wave
in 5G (from release 16 onwards) and operators are look- MNO Mobile Network Operator
ing for models for sharing infrastructure costs and to MVNO Mobile Virtual Network Operator
invest more in new services and applications, collaborat- NaaS Network as a Service
ing with different players of vertical industrial sectors. NFV Network Functions Virtualization
Moreover, it is becoming clear to the telecommunica- NSP Network Service Provider
tions industry sector that some form of infrastructure OTT Over The Top
sharing will be the common basis on which networks PRB Physical Resource Block
will be deployed in different countries and services will QoS Quality of Service
RAN Radio Access Network
evolve and diversify, going beyond 5G and preparing the
RRH Remote Radio Head
ground for the next generation. SaaS Software as a Service
SDN Software-Defined Networking
This survey is organized in the following fashion. Due
SINR Signal-to-Interference-plus-Noise Ratio
to the change in the nature of problems studied over
SLA Service Level Agreement
time, we first make a broad chronological classification SP Service Provider
of the literature into early works and recent and up-to- UMTS Universal Mobile Telecommunications System
date works. For the latter, we further identify several VNO Virtual Network Operator
research branches/categories. The overall picture of our VO Virtual Operator
classification is depicted in Fig. 1. An overview of the W-CDMA Wideband Code Division Multiple Access
early works on the topic is provided in Section 2. Fur- WNV Wireless Network Virtualization
ther, in Section 3, we focus on the more recent and up- xG xth mobile network Generation
to-date works. Then in Section 4 we make a critical
discussion of the research area related to infrastructure Table 1 – Definitions of acronyms and abbreviations
sharing and provide an outlook of future research direc- • state regulatory standpoints and provide guidelines
tions. Finally, conclusions are drawn in Section 5. for the latter and
For readers’ ease, in Table 1 we provide the definitions
of the acronyms and abbreviations used in the paper. • conceive new paradigms for the mobile market.

2. EARLY WORKS In [118], which dates back to 1994, Ramsdale states that
national roaming3 is part of the specifications of the Dig-
[16,49,55,70,115,118,142] are among the earliest articles ital Cellular System at 1800 MHz (DCS 1800), unlike
on infrastructure sharing (combined at times also with the Global System for Mobile Communications at 900
spectrum sharing). With the exception of [70]2 , these MHz (GSM 900), which supported international roam-
articles have tended to: ing only. National roaming was introduced in the DCS
1800 to improve coverage due to smaller cell sizes at
• address technical issues of different sharing alterna-
1800 MHz (as opposed to 900 MHz).
tives,
Instead, the work in [55] shows the positive impact of
• assess the financial profitability through techno- infrastructure sharing in financial terms for the Univer-
economic approaches, sal Mobile Telecommunications Systems (UMTS), espe-
1 This survey is based on the PhD thesis of Lorela Cano [24].
cially for lowly populated areas in which network de-
2 The study in [70] is an early work on the problem of schedul- ployment is dictated by coverage instead of capacity.
ing users of multiple operators arising from the case when a 3G,
facility-based MNO hosts several Mobile Virtual Network Oper- 3 National roaming is an infrastructure sharing alternative that
ators (MVNOs): the authors propose a non-pre-emptive priority allows users of an operator which does not provide coverage in
queuing model for circuit-switched traffic applied through an ad- certain areas of a country to be served by the network of another
mission control scheme. operator of that country covering such areas.

142 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

Fig. 1 – Literature classification map

In turn, MVNOs are suggested as a means to monetize mobile market, the former being responsible for network
spare resources of an MNO. planning, deployment and management while the latter
for dealing only with the development of novel services
Park et al. in [115] discuss issues faced by MNOs world- (possibly specialized and targeting specific market seg-
wide when deploying Wideband Code Division Multiple ments [16]).
Access (W-CDMA) and propose spectrum trading and
infrastructure sharing as means to accelerate the deploy- When analyzing these early works on infrastructure
ment of W-CDMA. However, they emphasize that such sharing in mobile networks, we have to consider the spe-
means should be cautiously treated by regulators. cific technical limitations that have constrained the ap-
proaches for 2G, 3G and partially 4G network, to some
The study in [49] proposes a spreadsheet-based financial aspects of the problem only. In particular, being the
model to estimate the economic profitability of multiple spectrum one of the most important assets of a mobile
sharing alternatives and shows that cost can be further network and being it easily shared among physically sep-
reduced if the network operations are outsourced or a arated networks, it has been widely studied considering
joint venture is created. the locality of interference generated and the re-usability
in different geographical areas.
The authors in [142] discuss technical aspects concern- As far as the physical infrastructure is concerned, the
ing the infrastructure sharing alternatives at the time; first works on sharing have focused on economic aspects
they also anticipate two crucial paradigms: (i) dynamic and market regulation policies associated to the intro-
spectrum trading and (ii) the decoupling of the network duction of MVNOs. However, the main limitation of
infrastructure from services, enabled by infrastructure these approaches was due to the mobile technology that
sharing. It is worth noticing that both these paradigms prevented a significant service and performance differen-
are ongoing research topics even nowadays. Similarly, tiation among users of MVNOs and MNOs. Therefore,
according to [16], the advantages of network sharing go sharing policies had to be based on other objectives such
beyond cost reduction: based on the product life cy- as cooperative coverage of low population areas and cost
cle model, the authors suggest that, under an appropri- sharing of radio towers.
ate regulatory framework, network sharing can steer the As mentioned above, the key aspect that we take from
monolithic mobile networks industry toward the decou- these works is the decoupling of network services from
pling of the network infrastructure from services for end the infrastructure that provides them. Only recently,
users. In other words, based on [142] and [16] infras- however, this concept has become fully exploitable
tructure sharing would lead to new stakeholders such thanks to the network virtualization technologies that
as network/infrastructure providers (InPs) and service allow a fine grain differentiation of the network behav-
providers (SPs) which were expected to emerge in the ior with respect to different applications and groups of

© International Telecommunication Union, 2020 143

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

users. This radical change of the technology scenario, decide to pool together their respective networks. Ki-
mainly due to the new architectural solutions and ser- bilda et al. [81] resort to stochastic geometry to cal-
vice definition of 5G, did not cancel the main issues culate the gains of sharing for the cases of infrastruc-
analyzed by the early works on infrastructure sharing, ture and/or spectrum pooling. Their key finding is that
such as the economic aspects of cost sharing and their the infrastructure and spectrum sharing gains do not
relation with resource allocation or partitioning. sum up when combined since full sharing (infrastruc-
ture+spectrum) introduces a trade-off between the data
3. MORE RECENT AND UP-TO- rate and coverage.
DATE WORKS As 5G is expected to make use of the millimeter wave
In the more recent and up-to-date literature, there is a (mmWave) frequencies [11], the gains of infrastructure
tendency to address specific problems, e.g., the problem and/or spectrum in these frequencies have become the
of resource management, for specific sharing scenarios, object of several recent articles. For instance, Gupta
e.g., infrastructure and spectrum sharing at the RAN. et al. in [54] provide a stochastic geometry-based the-
There are at least two ways to go about the classifi- oretical analysis on the gains of spectrum sharing us-
cation of this literature, one being problem-centric and ing a simplified antenna and channel model for the
the other being methodology-centric. We have opted mmWave frequency range. In particular, in [54] it is
for the first one in order to highlight the fact that there shown how narrow beams are key for spectrum sharing
are many aspects to infrastructure sharing and hence in the mmWaves. A very similar investigation to [81]
provide the reader with the bigger picture on the topic. is carried by Rebato et al. in [120] for mmWaves; the
Methodology details are discussed only when deemed authors highlight the impact of the channel model ac-
necessary. curacy when carrying out a quantitative analysis of the
Under the problem-centric classification, we have iden- sharing gains. The recent work in [71] also addresses
tified the following research branches/categories for the infrastructure and spectrum sharing at mmWaves and
revised articles: (i) performance evaluation, (ii) resource it resorts to stochastic geometry to derive the proba-
management, (iii) enablers and architectures, (iv) energy bility of Signal-to-Interference-plus-Noise Ratio (SINR)
efficiency, (v) strategic modeling and (vi) miscellaneous. coverage as a performance metric.

In Table 2 we provide a visual overview of the classi-

It is worth pointing out that some of the articles may fication of the different articles that were included in
fit in more than one category, but for each such article, the performance evaluation category. As can be seen
we have opted for a single category, the one we believe from the table, methodologically-wise, the authors use
is the most salient. mainly stochastic geometry, simulation and optimiza-
tion. The other method found was empirical analysis.
With respect to the type of measures used to evalu-
3.1 Performance evaluation ate performance, we can see that network measures are
fairly diverse: even though most work in this category
Several authors have addressed the gains of particu- deals with physical layer measures such as SINR, net-
lar infrastructure and/or spectrum sharing scenarios in working measures such as traffic load, sector overload
terms of network performance metrics, such as through- or packet drop probability are also considered. Not sur-
put, coverage probability etc. (see e.g., [71, 114, 139, prisingly, less diversity can be found in the economic
145]) and/or economic ones such as CAPEX/OPEX re- measures’ category.
duction (see e.g., [67, 75, 80, 106]). The common ap-
proach is to benchmark such scenarios against the base-
3.2 Resource management
line case when no sharing takes place and the involved
MNOs build individual networks instead. Methodology- Problems of resource management arise whenever infras-
wise, both theoretical, mainly stochastic geometry anal- tructure sharing is combined with spectrum sharing, as
ysis (see e.g., [54,71,81,145]), and simulation approaches users of multiple MNOs/MVNOs have to be assigned
(see e.g., [40,114,120]) have been adopted. For instance, resources from a shared pool.
the work in [114] proposes a virtualized architecture to
enable two types of spectrum sharing other than the Several studies ([34, 53, 99, 137]) have proposed algo-
classical one and capacity sharing (national roaming) rithms for a multi-operator scheduler, namely when
and compares the different sharing alternatives with no users of multiple MNOs have to be scheduled in the
sharing case. The performance metrics considered in finite resources available in a shared Base Station (BS).
[114] are the sector load and packet drop probability. Assuming MNOs agree a priori on the resource shares,
i.e., how to split the available BS resources among them,
The authors in [40] analyse how the time and space cor- the work in [137] adopts the concept of Generalized Pro-
relation of the MNO individual traffic loads impacts the cessor Sharing for a multi-operator scheduler. For the
gains of infrastructure sharing in the case when MNOs same setting, Malanchini et al. [99] explore the trade-

144 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

hhhh Methodology
hhh
Stochastic Geometry Simulation Optimization Other
Classification
hhh
hhhh
h
traffic load [40] [67]
SINR [54], [71] [120] [80]
SINR coverage [71], [81]
Network
probability
Measures
Throughput [67]
User rate [81], [145]
Sector overload [114]
Packet drop proba- [114]
bility
Revenue [75]
Economic CAPEX [75] [106]
Measures OPEX [75]
Miscellaneous [75] [67], [80]

Table 2 – Performance evaluation classification

off between satisfying the resource shares and improving the fact that a conventional MVNO competes with its
the overall (system) spectral efficiency when the agreed MNO, while there is no such competition between an
resource shares are violated in a controlled fashion. The InP and its SPs/VOs/VNOs/MVNOs. In these lines,
work in [53] considers a global scheduler taking deci- some articles tend to “misuse” the term InP when they
sions for clusters of BSs and therefore scheduling users of consider the InP to provide services also to end users.
multiple MNOs over a 3D time-frequency-space resource Additionally an InP is also referred to as a Network
grid. In [53] scheduling is performed with the objective Service Provider (NSP). Moreover, the work in [91] en-
of maximizing the overall system utility. The authors visions three different types of stakeholders in line with
in [34] propose a BS virtualization scheme which per- the ones in the cloud computing domain, i.e., the InP
forms scheduling in two levels, namely, among MNOs, providing Infrastructure as a Service (IaaS), the MVNO
and for each MNO, among its user flows. Hew et al. in providing Network as a Service (NaaS) and the SP pro-
[57] consider a network shared by multiple MNOs, each viding Software as a Service (SaaS). For instance, in [89]
of them serving both a set of end users and a set of the authors address a scenario in which there are mul-
MVNOs. In this context, the problem of resource allo- tiple InPs, a single MVNO and multiple SPs where the
cation is tackled in two steps: first, the resource sharing MVNO acts as a reseller of resources from InPs to SPs.
among MNOs, and then the resource sharing among the It should also be noted that the terms slicing and slice
users and the MVNOs of each MNO, where the resource are also misused in some articles in non-5G contexts,
sharing at each step is modeled as a bargaining problem. in the sense that, such articles do not consider problem
The study in [69] suggests an algorithm that fairly allo- instances that account for 5G service requirements.
cates the shared radio resources among MNOs. In [105]
the authors propose Remote Radio Head (RRH) assign- There is a large body of literature on resource man-
ment algorithms for an SDN-based Cloud Radio Access agements concerning InPs and SPs in the context of
Network (C-RAN) shared by multiple MNOs. WNV. The vast majority of articles in this litera-
ture considers a single InP and multiple SPs (see e.g.,
Concerning WNV, the problem of resource management [7, 8, 17, 31, 42, 50, 58, 59, 62, 63, 68, 73, 76, 79, 97, 123, 130,
is crucial in the interaction between an InP and its SPs. 132, 133, 136, 138, 146, 151, 154, 155]). However, there are
In the context of this paper, an InP is an entity which exceptions: e.g., the work in [32] considers a single InP
is responsible for the infrastructure deployment, man- and a single VNO which serves multiple users through
agement and operation and does not serve end users an SDN-based virtualized network provided by the InP.
directly whereas an SP is an entity which does not have The VNO faces the problem of scheduling its users, each
any resources of its own but purchases or rents resources characterized by a maximum delay over a finite time
from an InP to provision services for its end users. It is period, through resources rented by the InP with the
worth noticing that the terminology concerning the SP objective of minimizing the payments made to the InP
varies across different articles: such an entity is also re- for the rented resources. There are also articles which
ferred to as a Virtual Operator (VO), a Virtual Network consider both multiple InPs and multiple SPs (and few
Operator (VNO) or a Mobile Virtual Network Operator other variations with multiple InPs) which however are
(MVNO). Also notice that the conventional MVNO ob- more pertinent to Section 3.5 hence discussed therein.
tains resources from an MNO which serves end users
of its own, unlike the InP. The key difference lies in As for the literature on a single InP and multiple SPs,

© International Telecommunication Union, 2020 145

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

it can be broadly classified into two groups based on categories:

whether the resource management is driven by pric-
ing ([7, 8, 17, 31, 42, 58, 59, 76, 136, 146, 154, 155]) or not 1. partitioning and allocation of resources shared by
([50, 62, 63, 68, 73, 79, 97, 123, 130, 132, 133, 138, 151]). For multiple operators,
instance, Ho et al. in [58] consider the case when there
2. the literature on Wireless Network Virtualization
is a single InP serving multiple MVNOs, each charac-
that mostly deals with the sharing of resources
terized by a fixed number of users and a Service Level
between a single InP and multiple SPs; this sub-
Agreement (SLA) given in terms of a minimum resource
category, can be further subdivided into:
requirement and a maximum aggregate rate (over all its
users). The InP has to decide how to price and allo- (a) articles that base their modelling on pricing
cate its available BS resources among all users of all issues and
MVNOs so as to maximize its profit while guaranteeing
(b) articles that base their modelling on other is-
the SLA of each MVNO. In this work MVNOs are also
sues, such as performance metrics, and
self-interested as the goal of each MVNO is to maximize
its own profit given by the difference between the total 3. a large body of 5G literature that deals with re-
rate obtained from resources allocated by the InP and source management and network slicing.
their cost. The problem is then modeled as a one-leader
multi-follower variant of the Stackelberg game with the 3.3 Enablers and architectures
InP being the leader and each MVNO being a follower.
Instead, Kamel et al. in [73] address a scheduling prob- Although the different alternatives for infrastructure
lem over one time frame which is modeled through math- and spectrum sharing can be financially attractive for
ematical programming. In details, there is a single InP MNOs, they where not always supported by the 3GPP
and a set of VOs, each having a fixed number of users specifications; in fact, while a basic type of network shar-
and a minimum resource requirement (total Physical ing was supported as of Release 5, there was no sup-
Resource Blocks (PRBs) over the time frame). The InP port for more involved network sharing scenarios for the
has to decide to which user to assign each PRB and the 3GPP GSM EDGE4 RAN (GERAN) prior to Release
amount of power to allocate to each PRB so as to max- 10 ([3]).
imize the total rate over the time frame while satisfying
the maximum power constraint, the minimum resource Standardization apart, the research community has
requirement of each VO and a VO-specific proportional largely contributed on the topics of enabling network
fairness constraint for cell-center and cell-edge users. sharing, e.g., through novel architectures. While passive
sharing (i.e., site/tower sharing) is the simplest network
In 5G, the problem of resource management reemerges sharing alternative to implement, the different types of
in the context of multi-tenancy and its enabler, network active sharing demand architectural changes in mobile
slicing ([5, 6]). Tenants (such as MVNOs, Over The networks e.g., to guarantee the isolation of the involved
Top (OTT) providers and vertical industries) have dis- MNOs in terms of their private information in order to
tinct requirements to support their services which have avoid harming competition, or they demand changes at
to be translated into appropriate network resources. It the protocol stack level to implement the novel resource
is worth noting that network slicing does not involve management algorithms etc. According to [64], radio re-
only the RAN segment but it can be end-to-end. How- source management should be delegated to a third party
ever, the problem of resource management at the RAN provider to ensure isolation and therefore not to inter-
segment has brought about a significant amount of at- fere with competition. In [56] the authors introduce
tention from the research community due to the intrin- AppRAN which relies on a centralized scheduler to per-
sically complex nature of the radio (wireless) access. form application-level resource allocation for a shared
For instance, the authors in [125] propose the “5G Net- RAN.
work Slice Broker”, a centralized scheduler based on the In particular, different flavors of virtualization have been
3rd Generation Partnership Project (3GPP) specifica- widely considered by the research community as candi-
tions for network sharing. The proposed scheduler has date enablers for network sharing. For instance, the
a global view of the shared network and applies admis- virtualized network architecture proposed in [60] can
sion control and resource allocation, translating the ten- support network sharing. Other papers that resort to
ants’ request, with given SLAs, into available network virtualization are e.g., [10, 34, 72, 117, 152]. In particu-
resources. Other examples on resource management at lar, the authors of [38] and of [43] propose the “Network
the RAN in the context of multi-tenancy/network slic- without Borders”, namely the virtualized pool of (het-
ing are given in [9, 44, 47, 122, 144, 148, 153]. erogeneous) wireless resources for which infrastructure
and spectrum pooling are essential. Costanzo et al. in
Summarizing, the Resource Management category is [37] suggest an architecture for 4G RAN sharing based
a very rich part of the infrastructure sharing litera- on SDN and NFV.
ture. Within this category, we have identified three sub- 4 Enhanced Data rates for GSM Evolution

146 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

In the context of enabling network slicing in 5G net- Blogowski et al. in [18] deal with the particular scenario
works, there is a myriad of papers that propose archi- when two MNOs have to deploy BSs over a given set of
tectures or test prototypes based on (i) NFV and/or candidate sites. For each site, each MNO has to decide
SDN (see e.g, [35, 36, 85, 86, 113, 119]), (ii) changes to whether to install a BS or not; in the former case, if
the RAN protocol stack (see e.g., [48, 116, 124]), or (iii) both MNOs decide to install a BS, it is assumed that it
using features of the new 5G radio ([44]) etc. In partic- is profitable for both to install a single shared BS. The
ular, the work in [29] proposes an architecture to sup- problem is formulated as a non-cooperative game where
port network slicing in ultra-dense networks, the one in the payoff of each player (MNO) is given by its total
[74] presents an architecture that supports Internet Of profit (revenues - cost), calculated over all BSs. It is as-
Things (IoT) slices whereas the one in [129] dwells on sumed that each site can serve a given (arbitrary) num-
combining 3GPP specifications for 5G with NFV. ber of users, e.g., those under its coverage area, which
means there are no capacity constraints associated with
3.4 Energy efficiency the sites. Instead, coverage constraints are present and
they are expressed as a minimum percentage of users to
Infrastructure and spectrum sharing allow to reduce the be served by each MNO (a common constraint associ-
energy-consumption OPEX cost particularly in cases ated for spectrum licensees). When the coverage con-
when the aggregated network resources (infrastructure straint is absent, MNOs can decide independently for
and/or spectrum) are redundant. For instance, in ru- each site. Otherwise, the game is no longer separable.
ral areas where capacity is not an issue, MNOs can de- The authors describe the propriety of the Nash equilib-
commission a subset of the aggregated BSs and/or op- ria of the game for different relationships of the payoff
erate at a subset of the aggregated frequency carriers matrix (i.e., by establishing relations between the pay-
[49], which reduces the energy consumption and (indi- offs obtained under different strategy profiles) and also
rectly) the environmental impact. In these lines, since suggest a centralized solution which Pareto dominates
MNOs dimension their networks based on the peak- all Nash Equilibria.
load traffic predictions, there is intrinsically resource
redundancy during the off-peak periods in their indi- [108,109,127] address the problem of infrastructure and
vidual networks. Consequently, MNOs can agree to spectrum sharing arising when a set of MNOs, each
roam users of each other during the off-peak periods, with a given number of users (market share) and own
e.g., overnight, and switch off a subset of their BSs (see spectrum license, plan a greenfield Long-Term Evolu-
e.g., [13, 21]). While the vast majority of infrastruc- tion (LTE) deployment. The strategic problem of coali-
ture (and spectrum) sharing problems revolve around tion formation, namely, which subsets of MNOs volun-
economic and technical aspects, some papers (see e.g., tarily sign long-term infrastructure and spectrum shar-
[12–15,19–22,46,61,84,103,110–112,143]) have taken an ing agreements, is modeled by means of non-cooperative
energy-efficiency/green networking perspective. game theory. We address a very similar problem to
[108, 109, 127] in [25, 26] resorting to cooperative game
theory in [26] and non-cooperative game theory in [25].
3.5 Strategic modeling Unlike in [108,109,127], in [25,26] we (i) account for both
This branch consists of articles that deal with decision- the technical and economic aspects of sharing reflected
making problems such as MNOs deciding whether to en- in the payoff function definition and (ii) do not split the
ter a sharing agreement or not, SPs selecting InPs from shared infrastructure cost among MNOs a priori; how
which to obtain resources etc. In these lines we can these cost are split is an outcome of the model (game).
further split this category into two subcategories: (i) in- In turn in [27], we address a similar scenario to [25, 26]
frastructure sharing among conventional MNOs and (ii) but without spectrum pooling. Moreover, in [27] we
infrastructure sharing for decoupled infrastructure from consider two different cases deriving from two different
services (involving InPs and SPs etc.). Such articles perspectives, the one of a regulatory entity favoring the
naturally resort to mathematical programming and to users and the MNOs’ perspective as profit-maximizers.
game theory in particular when the involved actors are We model the former case through Mixed Integer Linear
assumed rational, self-interested and payoff-maximizing Programming and the latter through cooperative game
entities. theory.

The authors in [41] consider the case when a set of

3.5.1 Infrastructure sharing among conven- MNOs agrees to pool together their current individual
tional MNOs RAN networks but make joint decisions for future de-
commissions, network expansion and upgrades of their
The following articles concern either greenfield deploy- shared network; a greedy procedure is proposed to solve
ment of shared networks [18, 25–27, 108, 109, 127] or the the multi-period network planning.
case when shared networks are created by pooling to-
gether the existing network infrastructure of at least two Similarly to the “sale-leaseback” approach of Tower
MNOs [39, 41, 94, 126]. Companies (see e.g., [90]), the work in [39] assumes a set

© International Telecommunication Union, 2020 147

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

of self-interested MNOs decide to pool together their re- acts as the employer whereas the InPs as employees.
spective network infrastructures and create a Joint Ven-
ture (JV), responsible for managing their shared net- Instead, Wei et al. in [147] take a centralized approach.
work. In turn, MNOs will leaseback network capacity Specifically, the work in [147] considers multiple InPs
from the JV. The authors propose a Stackelberg game and multiple VNOs (analogous to SPs) in the context
to determine the shares MNOs obtain from the JV and of WNV. Here, each InP has a given set of users of its
the prices set by the JV to the MNOs and by the MNOs own; resources allocated to its own users are referred to
to their respective users. as local slices and the total rate across the local slices
should be above a given minimum for each InP. Instead,
Notably, the user perspective is considered in [94], which resources allocated to users of an MVNO are referred to
investigates the problem of user-to-BS association when as foreign slices. Each InP is characterized by a given
multiple MNOs decide to pool together their respective bandwidth (number of subchannels) and power budget
network infrastructures. The authors propose a non- for the downlink of a BS. The problem consists in deter-
cooperative game to model the problem of each user mining the number of subchannels and amount of power
selecting its serving BS from the shared pool, indepen- to allocate to each slice by each InP. The objective is to
dently, so that its individual data rate is maximized. maximize the total rate across all slices while satisfying
the bandwidth and power constraints and the minimum
rate requirement for the local slices of each InP. Conse-
The work in [126] represents a fresh take on infrastruc- quently, the problem is formulated by means of an Inte-
ture sharing. Its authors consider a set of MNOs with ger Programming (IP) model. In this model an MVNO
individual but overlapping infrastructures (BSs) and in- can be simultaneously served by multiple InPs, likewise
dividual spectrum licenses; in this setting one of the an InP can simultaneously serve multiple MVNOs.
MNOs (the buyer) can purchase the use of BSs of the
other MNOs (the sellers) for serving its own users at The authors in [156] propose a hierarchical (two layer)
its own licensed spectrum. The buyer MNO evaluates combinatorial auction to model the interactions among
whether it can provide a given (Quality of Service) QoS multiple InPs, multiple MVNOs (analogous to SPs), and
to its own users through its own infrastructure by in- multiple end users concerning the resource allocation
creasing the transmission power of its BSs or by pur- at the BS level (the resources here being transmission
chasing BSs from the seller MNOs. In the latter case, power, number of channels and number of antennas).
the buyer MNO has to decide from which seller MNOs
to buy from and what fraction of their BSs to purchase In [28] we propose a novel framework based on a Multi-
so as to minimize its expenditures while satisfying the Leader-Follower Game (MLFG) to study the techno-
QoS of its users. In turn, the seller MNOs have to decide economic interactions among multiple InPs and multiple
the fraction of their own BSs to sell so as to maximize SPs in a 5G context.
their profit (payment from the buyer MNO minus cost
Table 3 summarizes the main issues that are considered
of sold BSs) where the competition in quantity among
in this subcategory: what are the actors that intervene
the seller MNOs is modeled as a Cournot market.
in the infrastructure sharing scheme and what is the
modelling and mathematical approach that is taken in
3.5.2 Infrastructure sharing for decoupled in- each case.
frastructure from services
Article Actors Approach
We remind the reader that we have discussed the vary- [30] many InPs - one SP contract theory
ing terminology used across different articles related to [147] many InPs - many SPs IP model
the infrastructure sharing for decoupled infrastructure [156] many InPs - many SPs auction theory
from services in Section 3.2 and that we have maintained [28] many InPs - many SPs MLFG
the authors’ terminology for the considered stakehold-
ers when describing their articles and, when necessary, Table 3 – Infrastructure sharing for decoupled infrastructure from
we provide clarifications on how they compare to our services
definitions of InPs and SPs.

It is worth pointing out that, across the different articles 3.6 Miscellaneous
very distinct mathematical approaches have been used
to study the interaction among InPs and SPs. 3.6.1 Infrastructure sharing for mobile net-
work segments other than the access
Rather exceptionally, the study in [30] tackles the in-
teraction among InPs and MVNOs (analogous to SPs) Infrastructure sharing and multi-tenancy can also be ap-
from the MVNO perspective. In fact, the authors in [30] plied to specific segments of a mobile network other than
consider multiple InPs but a single MVNO and propose the access. For instance, the studies in [23, 98, 128, 140,
a model based on contract theory in which the MVNO 141] address sharing of the backhaul network whereas

148 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

the one in [83] deals with the sharing of the core net- different countries, that over the years have favoured the
work. introduction of MVNOs and roaming mechanisms, we
can observe that the wide service availability at reason-
3.6.2 Infrastructure sharing among different able prices has been considered particularly important
types of networks also with respect to natural market competition.

In the following paragraph we provide some examples With the arrival of 5G, we are observing a renewed inter-
of heterogeneous infrastructure sharing. The work in est in sharing strategies due to the specific virtualization
[78] studies sharing among different Radio Access Tech- technology available and the standardization of network
nologies (RATs), the one in [102] addresses sharing be- slicing. For the first time, the dynamic allocation of net-
tween LTE femtocells and Wi–Fi hotspots whereas the work resources and the service specialization on different
one in [100] investigates 3G offloading over Wi–Fi. Ki- slices, allow serving different groups of users according
bilda et al. [82] deal with sharing among MNOs and to different quality levels. This will likely generate the
OTTs. In [101] the authors propose a RAN architecture interest of new players specialized in vertical applica-
for both infrastructure and spectrum sharing between tion domains, in order to take the role of slice tenants
the MNOs and safety services. Instead the study in and sharing the resources of the network infrastructures
[95] concerns infrastructure sharing between mobile ser- with others. Even extreme scenarios where communi-
vices and smart grid utilities or intelligent transporta- cations resources are traded in real time on automated
tion services. Lin et al. in [93] address backhaul shar- markets are now possible [92], like it already happens in
ing among mobile networks and fixed networks whereas other sectors like energy networks. With this regard, the
Simo-Reigadas et al. in [131] suggest exploiting the com- use of the large toolbox created by research over several
munity infrastructure as backhaul for 3G. years will certainly be an important asset to be used to
shape sharing and trading instruments.
3.6.3 Infrastructure sharing for networks other
There are however, other important evolution trends of
than mobile the technology that will probably influence the sharing
The concept of infrastructure sharing is not exclusive to methods beyond 5G. We want to mention here two of
mobile networks. In fact, it has been applied to fixed them that we believe are particularly relevant:
access networks and problems related to the latter have
been recently addressed in the literature (see e.g., [33] • the extreme distribution of access infrastructure
and [77]). Apart from fixed access networks, infrastruc- with the so-called smart radio environments, and
ture sharing has also been proposed for Wi–Fi networks,
e.g., in [121].
• the full virtualization of connectivity in open and
cloud-based architectures.
3.6.4 Spectrum sharing
As previously stated, the overall literature on the dif-
ferent types of spectrum sharing alone (i.e., not com- As for the smart radio environments, they consider the
bined with infrastructure sharing) is per se very vast. introduction of new equipment at the radio interface of
Unsurprisingly, as spectrum is a scarce resource for the mobile networks, which includes smart repeaters with
MNOs, many papers within this literature resort to dif- large antenna arrays and controllable reflective surfaces.
ferent game theory models (see e.g., [94, 135, 149, 150]). This kind of evolution trend is making clear that in the
future the deployment of multiple physical infrastruc-
tures by different operators will become more and more
3.6.5 MVNO business model difficult and the focus of MNO attention will shift from
The relation among the MNO, its MVNO(s) and the end optimizing their own network to that of efficiently con-
users has been largely addressed through game theory trolling the resources leased from the common infras-
as well (see e.g., [57, 87, 88, 96, 134]). tructure.

While the full virtualization of connectivity will be

4. DISCUSSION AND OUTLOOK
complete relatively soon, the evolution started years
The large body of literature analyzed in this survey and ago with the separation of the network logic from the
the impact it has had over the years on the standard- pure transmission technology. There are here how-
ization of mobile technologies and the practices adopted ever, important novelties associated with open source
around the world, testify that mobile networks are im- approaches like Open RAN that are becoming popular
portant infrastructures with high costs which can be and that are fostering the transition to cloud-based so-
shared in some scenarios in order to offer better and lutions where the value for service providers will be in
more convenient services to end users. Also, in terms of designing and implementing advanced services based on
regulatory strategies adopted by national authorities in the effective use of transmission resources.

© International Telecommunication Union, 2020 149

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

5. CONCLUSION [9] O. Al-Khatib, W. Hardjawana, and B. Vucetic.

Spectrum sharing in multi-tenant 5G cellular net-
Infrastructure sharing in mobile networks has been a
works: Modeling and planning. IEEE Access,
pervasive research topic over the last three decades and
7:1602–1616, 2019.
has produced a significant body of work.
[10] L. Anchora, M. Mezzavilla, L. Badia, and
One interesting takeaway from this survey is that while
M. Zorzi. A performance evaluation tool for spec-
researchers sought enabling technologies to materialize
trum sharing in multi-operator LTE networks.
infrastructure sharing in 3G and 4G networks, in 5G
Computer Communications, 35(18):2218–2226,
networks instead, infrastructure sharing became an im-
November 2012.
portant pillar of the 5G architecture which means that
in turn 5G enables infrastructure sharing from a busi- [11] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly,
ness point of view. In these lines one can easily argue A. Lozano, A. C. Soong, and J. C. Zhang. What
the presence of infrastructure sharing also in future net- will 5g be? IEEE Journal on selected areas in
works. communications, 32(6):1065–1082, 2014.
As for the mobile ecosystem, it is worth noting that the [12] C. Anglano, M. Guazzone, and M. Sereno.
concepts of decoupling infrastructure from services and Maximizing profit in green cellular networks
dynamic spectrum trading have been anticipated in the through collaborative games. Computer Networks,
literature over two decades ago but they have come into 75:260–275, 2014.
being only recently (mainly in 5G networks).
[13] A. Antonopoulos, E. Kartsakli, A. Bousia,
Of the several research branches within the infrastruc- L. Alonso, and C. Verikoukis. Energy-efficient
ture and spectrum sharing topics, resource management infrastructure sharing in multi-operator mobile
in the context of resource sharing has been and will be networks. IEEE Communications Magazine,
one of the most active research branches given the cur- 53(5):242–249, 2015.
rent and future need for dynamic resource sharing.
[14] O. Aydin, E. A. Jorswieck, D. Aziz, and A. Zap-
pone. Energy-spectral efficiency tradeoffs in 5G
REFERENCES multi-operator networks with heterogeneous con-
[1] 3GPP. TR 22.852, v13.1.0, Study on Radio Ac- straints. IEEE Transactions on Wireless Commu-
cess Network (RAN) sharing enhancements (Re- nications, 16(9):5869–5881, 2017.
lease 13), September 2014.
[15] Y. Bao, J. Wu, S. Zhou, and Z. Niu. Bayesian
[2] 3GPP. TR 22.951, v15.0.0, Service aspects and re- mechanism based inter-operator base station shar-
quirements for network sharing (Release 15), July ing for energy saving. In Communications (ICC),
2018. 2015 IEEE International Conference on, pages
49–54. IEEE, 2015.
[3] 3GPP. TS 23.251, v15.1.0, Network sharing;
Architecture and functional descriptions (Release [16] C. Beckman and G. Smith. Shared networks:
15), September 2018. Making wireless communication affordable. IEEE
Wireless Communications, 12(2):78–85, April
[4] 3GPP. TS 32.130, v15.0.0, Telecommunication 2005.
management; Network sharing; Concepts and re-
quirements (Release 15), June 2018. [17] D. Bega, M. Gramaglia, A. Banchs, V. Sciancale-
pore, K. Samdanis, and X. Costa-Perez. Optimis-
[5] 5G NORMA. Deliverable D3.3, 5G NORMA net-
ing 5G infrastructure markets: The business of
work architecture - final report, October 2017.
network slicing. In IEEE INFOCOM 2017-IEEE
[6] 5GPPP Architecture Working Group. View on 5G Conference on Computer Communications, pages
architecture, December 2017. 1–9. IEEE, 2017.
[7] H. Ahmadi, I. Macaluso, I. Gomez, L. DaSilva, [18] A. Blogowski, P. Chrétienne, and F. Pascual. Net-
and L. Doyle. Virtualization of spatial streams work sharing by two mobile operators: beyond
for enhanced spectrum sharing. In 2016 IEEE competition, cooperation. RAIRO-Operations Re-
Global Communications Conference (GLOBE- search, 49(3):635–650, 2015.
COM), pages 1–6. IEEE, 2016.
[19] A. Bousia, E. Kartsakli, A. Antonopoulos,
[8] Ö. U. Akgül, I. Malanchini, V. Suryaprakash, and L. Alonso, and C. Verikoukis. Game theoretic
A. Capone. Dynamic resource allocation and pric- approach for switching off base stations in multi-
ing for shared radio access infrastructure. In 2017 operator environments. In 2013 IEEE Inter-
IEEE International Conference on Communica- national Conference on Communications (ICC),
tions (ICC), pages 1–7. IEEE, 2017. pages 4420–4424. IEEE, 2013.

150 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[20] A. Bousia, E. Kartsakli, A. Antonopoulos, [30] Z. Chang, D. Zhang, T. Hämäläinen, Z. Han, and
L. Alonso, and C. Verikoukis. Auction-based of- T. Ristaniemi. Incentive mechanism for resource
floading for base station switching off in heteroge- allocation in wireless virtualized networks with
neous networks. In 2016 European Conference on multiple infrastructure providers. IEEE Transac-
Networks and Communications (EuCNC), pages tions on Mobile Computing, 19(1):103–115, 2018.
335–339. IEEE, 2016.
[31] Z. Chang, K. Zhu, Z. Zhou, and T. Ris-
[21] A. Bousia, E. Kartsakli, A. Antonopoulos, taniemi. Service provisioning with multiple service
L. Alonso, and C. Verikoukis. Game-theoretic in- providers in 5g ultra-dense small cell networks. In
frastructure sharing in multioperator cellular net- IEEE PIMRC 2015, pages 1895–1900, 2015.
works. IEEE Transactions on Vehicular Technol-
ogy, 65(5):3326–3341, 2016. [32] X. Chen, H. Zhang, and Z. Han. Delay-tolerant
resource scheduling in large-scale virtualized ra-
[22] A. Bousia, E. Kartsakli, A. Antonopoulos, dio access networks. In 2017 IEEE International
L. Alonso, and C. Verikoukis. Multiobjec- Conference on Communications (ICC), pages 1–6.
tive auction-based switching-off scheme in het- IEEE, 2017.
erogeneous networks: To bid or not to bid?
IEEE Transactions on Vehicular Technology, [33] B. Cornaglia, G. Young, and A. Marchetta. Fixed
65(11):9168–9180, 2016. access network sharing. Optical Fiber Technology,
26:2–11, 2015.
[23] C. Caillouet, D. Coudert, and A. Kodjo. Ro-
bust optimization in multi-operators microwave [34] X. Costa-Pérez, J. Swetina, T. Guo, R. Mahin-
backhaul networks. In Global Information Infras- dra, and S. Rangarajan. Radio access network
tructure Symposium-GIIS 2013, pages 1–6. IEEE, virtualization for future mobile carrier networks.
2013. IEEE Communications Magazine, 51(7):27–35,
July 2013.
[24] L. Cano. Game-theoretic frameworks for the
techno-economic aspects of infrastructure sharing [35] S. Costanzo, I. Fajjari, N. Aitsaadi, and R. Lan-
in current and future mobile networks. PhD the- gar. DEMO: SDN-based network slicing in C-
sis, Polytechnique Montréal and Politecnico di Mi- RAN. In 2018 15th IEEE Annual Consumer Com-
lano, 2020. munications & Networking Conference (CCNC),
pages 1–2. IEEE, 2018.
[25] L. Cano, A. Capone, G. Carello, M. Cesana,
and M. Passacantando. Cooperative infrastruc- [36] S. Costanzo, I. Fajjari, N. Aitsaadi, and R. Lan-
ture and spectrum sharing in heterogeneous mo- gar. A network slicing prototype for a flexible
bile networks. IEEE Journal on Selected Areas in cloud radio access network. In 2018 15th IEEE
Communications, 34(10):2617–2629, oct 2016. Annual Consumer Communications & Networking
Conference (CCNC), pages 1–4. IEEE, 2018.
[26] L. Cano, A. Capone, G. Carello, M. Cesana, and
M. Passacantando. A non-cooperative game ap- [37] S. Costanzo, D. Xenakis, N. Passas, and L. Mer-
proach for RAN and spectrum sharing in mobile akos. Augmented RAN with SDN Orchestration
radio networks. In 22th European Wireless Confer- of Multi-tenant Base Stations. Wireless Personal
ence, pages 1–6, Oulu, Finland, May 18-20, 2016. Communications, 96(2):2009–2037, 2017.

[27] L. Cano, A. Capone, G. Carello, M. Cesana, [38] L. A. DaSilva, J. Kibilda, P. DiFrancesco, T. K.

and M. Passacantando. On optimal infrastruc- Forde, and L. E. Doyle. Customized services over
ture sharing strategies in mobile radio networks. virtual wireless networks: The path towards net-
IEEE Transactions on Wireless Communications, works without borders. In Future Network and
16(5):3003–3016, may 2017. Mobile Summit (FutureNetworkSummit), 2013,
pages 1–10. IEEE, 2013.
[28] L. Cano, G. Carello, M. Cesana, M. Passacan-
tando, and B. Sansò. Modeling the techno- [39] X. Deng, J. Wang, and J. Wang. How to Design
economic interactions of infrastructure and ser- a Common Telecom Infrastructure for Competi-
vice providers in 5G networks with a multi-leader- tors to be Individually Rational and Collectively
follower game. IEEE Access, 7:162913–162940, dec Optimal. IEEE Journal on Selected Areas in Com-
2019. munications, 35(3):736–750, 2017.
[29] C.-Y. Chang, N. Nikaein, O. Arouk, K. Kat- [40] P. Di Francesco, F. Malandrino, and L. A.
salis, A. Ksentini, T. Turletti, and K. Samdanis. DaSilva. Mobile network sharing between opera-
Slice orchestration for multi-service disaggregated tors: a demand trace-driven study. In Proceedings
ultra-dense RANs. IEEE Communications Maga- of the 2014 ACM SIGCOMM workshop on Capac-
zine, 56(8):70–77, 2018. ity sharing workshop, pages 39–44. ACM, 2014.

© International Telecommunication Union, 2020 151

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[41] P. Di Francesco, F. Malandrino, T. K. Forde, and infrastructure. https://tmt.mckinsey.com/con-

L. A. DaSilva. A sharing-and competition-aware tent/industry/Telecommunications/page/19,
framework for cellular network evolution planning. 2012. [Online; Accessed: 2016-03-30].
IEEE Transactions on Cognitive Communications
and Networking, 1(2):230–243, June 2015. [52] GSMA. Mobile infrastructure sharing.
https://www.gsma.com/mobilefordevelop-
[42] S. D’Oro, F. Restuccia, T. Melodia, and ment/programme/connected-society/mobile-
S. Palazzo. Low-complexity distributed radio ac- infrastructure-sharing-report/, November 2008.
cess network slicing: Algorithms and experimental Accessed: 11-01-2019.
results. IEEE/ACM Transactions on Networking, [53] A. Gudipati, L. E. Li, and S. Katti. RadioVisor:
26(6):2815 – 2828, 2018. A slicing plane for Radio Access Networks. In Pro-
[43] L. Doyle, J. Kibilda, T. K. Forde, and L. DaSilva. ceedings of the Third Workshop on Hot Topics in
Spectrum without bounds, networks without bor- Software Defined Networking, HotSDN ’14, pages
ders. Proceedings of the IEEE, 102(3):351–365, 237–238, 2014.
2014. [54] A. K. Gupta, J. G. Andrews, and R. W. Heath.
On the feasibility of sharing spectrum licenses in
[44] S. E. Elayoubi, S. B. Jemaa, Z. Altman, and mmwave cellular systems. IEEE Transactions on
A. Galindo-Serrano. 5G RAN slicing for verticals: Communications, 64(9):3981–3995, 2016.
Enablers and challenges. IEEE Communications
Magazine, 57(1):28–34, 2019. [55] J. Harno. 3G business case successful-
ness within the constraints set by competi-
[45] Ericsson. Network sharing. https://www.er- tion, regulation and alternative technologies.
icsson.com/us/ourportfolio/networks- JOURNAL-COMMUNICATIONS NETWORK,
services/network-sharing?nav=marketcate- 1(2):159–165, 2002.
gory004 [Online; Accessed: 2017-05-15].
[56] J. He and W. Song. Appran: Application-oriented
[46] M. J. Farooq, H. Ghazzai, E. Yaacoub, A. Kadri, radio access network sharing in mobile networks.
and M.-S. Alouini. Green virtualization for multi- In Communications (ICC), 2015 IEEE Interna-
ple collaborative cellular operators. IEEE Trans- tional Conference on, pages 3788–3794. IEEE,
actions on Cognitive Communications and Net- 2015.
working, 3(3):420–434, 2017.
[57] S. L. Hew and L. B. White. Cooperative re-
[47] A. Fendt, S. Lohmuller, L. C. Schmelz, and source allocation games in shared networks: Sym-
B. Bauer. A network slice resource allocation metric and asymmetric fair bargaining models.
and optimization model for end-to-end mobile net- IEEE Transactions on Wireless Communications,
works. In 2018 IEEE 5G World Forum (5GWF), 7(11):4166–4175, November 2008.
pages 262–267. IEEE, 2018. [58] T. M. Ho, N. H. Tran, S. A. Kazmi, and C. S.
[48] R. Ferrus, O. Sallent, J. Pérez-Romero, and Hong. Dynamic pricing for resource allocation
R. Agusti. On 5G radio access network slic- in wireless network virtualization: A stackelberg
ing: Radio interface protocol features and con- game approach. In IEEE ICOIN 2017, pages
figuration. IEEE Communications Magazine, 429–434, 2017.
56(5):184–192, 2018. [59] T. M. Ho, N. H. Tran, L. B. Le, Z. Han, S. A.
Kazmi, and C. S. Hong. Network virtualization
[49] T. Frisanco, P. Tafertshofer, P. Lurin, and R. Ang. with energy efficiency optimization for wireless
Infrastructure sharing for mobile network oper- heterogeneous networks. IEEE Transactions on
ators; From a deployment and operations view. Mobile Computing, 18(10):2386–2400, 2018.
In IEEE International Conference on Information
Networking (ICOIN), pages 1–5, January 2008. [60] M. Hoffmann and M. Staufer. Network virtualiza-
tion for future mobile networks: General architec-
[50] L. Gao, P. Li, Z. Pan, N. Liu, and X. You. Vir- ture and applications. In 2011 IEEE international
tualization framework and VCG based resource conference on communications workshops (ICC),
block allocation scheme for LTE virtualization. In pages 1–5. IEEE, 2011.
2016 IEEE 83rd Vehicular Technology Conference
(VTC Spring), pages 1–6. IEEE, 2016. [61] M. F. Hossain, K. S. Munasinghe, and A. Ja-
malipour. Energy-efficient inter-RAN cooperation
[51] F. Grijpink, S. Newman, S. Sandoval, for non-collocated cell sites with base station se-
M. Strandell-Jansson, and W. Torfs. A “New lection and user association policies. Wireless Net-
Deal”: Driving investment in Europe’s telecoms works, 25(1):269–285, 2019.

152 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[62] F.-T. Hsu and C.-H. Gan. Resource allocation [72] M. Kalil, M. Youssef, A. Shami, A. Al-Dweik,
with spectrum aggregation for wireless virtual net- and S. Ali. Wireless resource virtualization: op-
work embedding. In 2015 IEEE 82nd Vehicu- portunities, challenges, and solutions. Wire-
lar Technology Conference (VTC2015-Fall), pages less Communications and Mobile Computing,
1–5. IEEE, 2015. 16(16):2690–2699, 2016.

[63] M. Hu, Y. Chang, Y. Sun, and H. Li. Dy- [73] M. I. Kamel, L. B. Le, and A. Girard. LTE wire-
namic slicing and scheduling for wireless net- less network virtualization: Dynamic slicing via
work virtualization in downlink LTE system. In flexible scheduling. In 2014 IEEE 80th Vehicu-
2016 19th International Symposium on Wireless lar Technology Conference (VTC2014-Fall), pages
Personal Multimedia Communications (WPMC), 1–5. IEEE, 2014.
pages 153–158. IEEE, 2016. [74] E. Kapassa, M. Touloupou, P. Stavrianos, and
D. Kyriazis. Dynamic 5G Slices for IoT ap-
[64] J. Hultel, K. Johansson, and J. Markendahl. Busi- plications with diverse requirements. In 2018
ness models and resource management for shared Fifth International Conference on Internet of
wireless networks. In IEEE 60th Vehicular Tech- Things: Systems, Management and Security,
nology Conference (VTC2004-Fall), volume 5, pages 195–199. IEEE, 2018.
pages 3393–3397, September 2004.
[75] M. Katsigiannis, T. Smura, T. Casey, and A. Sorri.
[65] Industry Canada. Framework for manda- Techno-economic modeling of value network con-
tory roaming and antenna tower and site figurations for public wireless local area access.
sharing. http://www.ic.gc.ca/eic/site/smt- NETNOMICS: Economic Research and Electronic
gst.nsf/eng/h𝑠 𝑓10290.ℎ𝑡𝑚𝑙, 2013.[𝑂𝑛𝑙𝑖𝑛𝑒; 𝐴𝑐𝑐𝑒𝑠𝑠𝑒𝑑 ∶ Networking, 14(1):27–46, November 2013.
2016 − 03 − 30].
[76] S. M. A. Kazmi and C. S. Hong. A matching
[66] ITU. Mobile and wireless network regulation. game approach for resource allocation in wireless
http://www.ictregulationtoolkit.org/2.6. [Online; network virtualization. In Proceedings of the 11th
Accessed: 2016-03-28]. International Conference on ubiquitous informa-
tion management and communication, pages 1–6,
[67] T. Janssen, R. Litjens, and K. W. Sowerby. On 2017.
the expiration date of spectrum sharing in mobile [77] K. J. Kerpez, J. M. Cioffi, P. J. Silverman, B. Cor-
cellular networks. In 2014 12th International Sym- naglia, and G. Young. Fixed access network shar-
posium on Modeling and Optimization in Mobile, ing. IEEE Communications Standards Magazine,
Ad Hoc, and Wireless Networks (WiOpt), pages 1(1):82–89, 2017.
490–496. IEEE, 2014.
[78] M. A. Khan, A. C. Toker, C. Troung, F. Sivrikaya,
[68] Y. Jia, H. Tian, S. Fan, P. Zhao, and K. Zhao. and S. Albayrak. Cooperative game theoretic ap-
Bankruptcy game based resource allocation algo- proach to integrated bandwidth sharing and al-
rithm for 5G Cloud-RAN slicing. In IEEE WCNC location. In IEEE International Conference on
2018, pages 1–6, 2018. Game Theory for Networks (GameNets ’09), pages
1–9, May 2009.
[69] K. Johansson. Cost effective deployment strategies
for heterogeneous wireless networks. PhD thesis, [79] S. Khatibi and L. M. Correia. Modelling of
KTH Information and Communication Technol- virtual radio resource management for cellular
ogy, 2007. heterogeneous access networks. In 2014 IEEE
25th Annual International Symposium on Per-
[70] K. Johansson, M. Kristensson, and U. Schwarz. sonal, Indoor, and Mobile Radio Communication
Radio resource management in roaming based (PIMRC), pages 1152–1156. IEEE, 2014.
multi-operator wcdma networks. In Vehicular [80] J. Kibilda and L. A.DaSilva. Efficient coverage
Technology Conference, 2004. VTC 2004-Spring. through inter-operator infrastructure sharing in
2004 IEEE 59th, volume 4, pages 2062–2066. mobile networks. In IEEE IFIP Wireless Days
IEEE, 2004. (WD), pages 1–6, November 2013.

[71] R. Jurdi, A. K. Gupta, J. G. Andrews, and [81] J. Kibiłda, N. J. Kaminski, and L. A. DaSilva. Ra-
R. W. Heath. Modeling infrastructure sharing in dio access network and spectrum sharing in mo-
mmWave networks with shared spectrum licenses. bile networks: A stochastic geometry perspective.
IEEE Transactions on Cognitive Communications IEEE Transactions on Wireless Communications,
and Networking, 4(2):328–343, 2018. 16(4):2562–2575, 2017.

© International Telecommunication Union, 2020 153

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[82] J. Kibilda, F. Malandrino, and L. A. DaSilva. in radio access networks. IEEE Transactions on
Incentives for infrastructure deployment by over- Mobile Computing, 2020.
the-top service providers in a mobile network:
A cooperative game theory model. In 2016 [93] P. Lin, J. Zhang, Q. Zhang, and M. Hamdi.
IEEE International Conference on Communica- Enabling the femtocells: A cooperation frame-
tions (ICC), pages 1–6. IEEE, 2016. work for mobile and fixed-line operators. IEEE
Transactions on Wireless Communications,
[83] W. Kiess, M. R. Sama, J. Varga, J. Prade, H.-J. 12(1):158–167, 2013.
Morper, and K. Hoffmann. 5G via evolved packet
core slices: Costs and technology of early deploy- [94] Y.-T. Lin, H. Tembine, and K.-C. Chen. Inter-
ments. In 2017 IEEE 28th Annual International operator spectrum sharing in future cellular sys-
Symposium on Personal, Indoor, and Mobile Ra- tems. In Global Communications Conference
dio Communications (PIMRC), pages 1–7. IEEE, (GLOBECOM), 2012 IEEE, pages 2597–2602.
2017. IEEE, 2012.

[84] G. Koutitas, G. Iosifidis, B. Lannoo, M. Tahon, [95] Y. Lostanlen. From heterogeneous wireless net-
S. Verbrugge, P. Ziridis, Ł. Budzisz, M. Meo, works to sustainable efficient ICT infrastructures.
M. A. Marsan, and L. Tassiulas. Greening the In 2013 7th European Conference on Antennas and
airwaves with collaborating mobile network oper- Propagation (EuCAP), pages 1360–1363. IEEE,
ators. IEEE Transactions on Wireless Communi- 2013.
cations, 15(1):794–806, 2016.
[96] M. H. Lotfi and S. Sarkar. The economics of
[85] F. Kurtz, C. Bektas, N. Dorsch, and C. Wiet- competition and cooperation between mnos and
feld. Network slicing for critical communications mvnos. In 2017 51st Annual Conference on Infor-
in shared 5G infrastructures-an empirical evalua- mation Sciences and Systems (CISS), pages 1–6.
tion. In 2018 4th IEEE Conference on Network IEEE, 2017.
Softwarization and Workshops (NetSoft), pages
[97] X. Lu, K. Yang, Y. Liu, D. Zhou, and S. Liu.
393–399. IEEE, 2018.
An elastic resource allocation algorithm enabling
[86] L.-V. Le, B.-S. P. Lin, L.-P. Tung, and D. Sinh. wireless network virtualization. Wireless Commu-
SDN/NFV, Machine Learning, and Big Data nications and Mobile Computing, 15(2):295–308,
Driven Network Slicing for 5G. In 2018 IEEE 5G 2015.
World Forum (5GWF), pages 20–25. IEEE, 2018.
[98] J. Lun and D. Grace. Software defined net-
[87] H. Le Cadre and M. Bouhtou. An interconnection work for multi-tenancy resource sharing in back-
game between mobile network operators: Hidden haul networks. In 2015 IEEE Wireless Commu-
information forecasting using expert advice fusion. nications and Networking Conference Workshops
Computer networks, 54(17):2913–2942, 2010. (WCNCW), pages 1–5. IEEE, 2015.

[88] H. Le Cadre and M. Bouhtou. Modelling MNO [99] I. Malanchini, S. Valentin, and O. Aydin. Wire-
and MVNO’s dynamic interconnection relations: less resource sharing for multiple operators: Gen-
is cooperative content investment profitable for eralization, fairness, and the value of prediction.
both providers? Telecommunication Systems, Computer Networks, 100:110–123, 2016.
51(2-3):193–217, 2012.
[100] L. Mamatas, I. Psaras, and G. Pavlou. Incentives
[89] T. LeAnh, N. H. Tran, D. T. Ngo, and C. S. Hong. and algorithms for broadband access sharing. In
Resource allocation for virtualized wireless net- Proceedings of the 2010 ACM SIGCOMM work-
works with backhaul constraints. IEEE Commu- shop on Home networks, pages 19–24, 2010.
nications Letters, 21(1):148–151, 2017.
[101] D. Marabissi and R. Fantacci. Heterogeneous pub-
[90] T. Levine, P. Eijsvoogel, and M. Reede. Pas- lic safety network architecture based on RAN slic-
sive infrastructure sharing. http://www.al- ing. IEEE Access, 5:24668–24677, 2017.
lenovery.com/SiteCollectionDocuments/Pas-
sive2012. [Online; Accessed: 2016-03-28]. [102] J. Markendahl and M. Nilson. Business models for
deployment and operation of femtocell networks;
[91] C. Liang and F. R. Yu. Wireless virtualization for – Are new operation strategies needed for mobile
next generation mobile cellular networks. IEEE operators? In 21st European Regional ITS Con-
wireless communications, 22(1):61–69, 2015. ference, Copenhagen, September 2010.

[92] A. Lieto, I. Malanchini, S. Mandelli, E. Moro, and [103] M. A. Marsan and M. Meo. Network sharing and
A. Capone. Strategic network slicing management its energy benefits: A study of European mobile

154 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

network operators. In 2013 IEEE Global Com- architectures, and challenges. IEEE Communi-
munications Conference (GLOBECOM), pages cations Magazine, 55(5):80–87, 2017.
2561–2567. IEEE, 2013.
[114] J. S. Panchal, R. D. Yates, and M. M. Buddhikot.
[104] D.-E. Meddour, T. Rasheed, and Y. Gourhant. Mobile network resource sharing options: Perfor-
On the role of infrastructure sharing for mobile mance comparisons. IEEE Transactions on Wire-
network operators in emerging markets. Computer less Communications, 12(9):4470–4482, 2013.
Networks, 55(7):1576–1591, May 2011.
[115] J. S. Park, M. Kim, and H. J. Lee. Analysis of Eu-
[105] O. Narmanlioglu and E. Zeydan. Efficient RRH ropean 3G markets and advanced strategies for 3G
assignments for mobile network operators in development. In The 7th International Conference
shared cellular network architecture. In 2017 on Advanced Communication Technology, 2005,
IFIP/IEEE Symposium on Integrated Network ICACT 2005., volume 1, pages 428–431. IEEE,
and Service Management (IM), pages 1103–1108. 2005.
IEEE, 2017.
[116] J. Pérez-Romero, O. Sallent, R. Ferrús, and
[106] B. Naudts, M. Kind, F.-J. Westphal, S. Ver- R. Agustí. On the configuration of radio resource
brugge, D. Colle, and M. Pickavet. Techno- management in a sliced RAN. In NOMS 2018-
economic analysis of software defined networking 2018 IEEE/IFIP Network Operations and Man-
as architecture for the virtualization of a mobile agement Symposium, pages 1–6. IEEE, 2018.
network. In 2012 European workshop on software
defined networking, pages 67–72. IEEE, 2012. [117] M. Rahman, C. Despins, and S. Affes. Analy-
sis of CAPEX and OPEX benefits of wireless ac-
[107] Nokia. White paper - network sharing: Delivering cess virtualization. In IEEE International Confer-
mobile broadband more efficiently and at lower ence on Communications (ICC) Workshops, pages
cost. nokia.com. [Online; Accessed: 2017-05-15]. 436–440, June 2013.
[108] F. Offergelt, F. Berkers, and G. Hendrix. If you
[118] P. Ramsdale. Personal communications in the
can’t beat ’em, join ’em; Cooperative and non-
UK—Implementation of PCN using DCS 1800.
cooperative games in network sharing. In IEEE
International Journal of Wireless Information
15th International Conference on Intelligence in
Networks, 1(1):29–36, 1994.
Next Generation Networks (ICIN), pages 196–201,
October 2011. [119] R. Ravindran, A. Chakraborti, S. O. Amin, A. Az-
[109] F. H. Offergelt. Saphyre: Cooperation among gin, and G. Wang. 5G-ICN: Delivering ICN ser-
competitors – analysing sharing scenarios for mo- vices over 5G using network slicing. IEEE Com-
bile network operators using game theory. Mas- munications Magazine, 55(5):101–107, 2017.
ter’s thesis, Leiden University, The Netherlands, [120] M. Rebato, M. Mezzavilla, S. Rangan, and
2011. M. Zorzi. Resource sharing in 5g mmwave cellular
[110] M. Oikonomakou, A. Antonopoulos, L. Alonso, networks. In Computer Communications Work-
and C. Verikoukis. Cooperative base station shops (INFOCOM WKSHPS), 2016 IEEE Con-
switching off in multi-operator shared heteroge- ference on, pages 271–276. IEEE, 2016.
neous network. In 2015 IEEE Global Commu- [121] M. Richart, J. Baliosian, J. Serrati, J.-L. Gorri-
nications Conference (GLOBECOM), pages 1–6. cho, R. Agüero, and N. Agoulmine. Resource al-
IEEE, 2015. location for network slicing in WiFi access points.
[111] M. Oikonomakou, A. Antonopoulos, L. Alonso, In 2017 13th International conference on net-
and C. Verikoukis. Evaluating cost allocation im- work and service management (CNSM), pages
posed by cooperative switching off in multiopera- 1–4. IEEE, 2017.
tor shared hetnets. IEEE Transactions on Vehic-
[122] R. Riggio, A. Bradai, D. Harutyunyan,
ular Technology, 66(12):11352–11365, 2017.
T. Rasheed, and T. Ahmed. Scheduling wireless
[112] M. Oikonomakou, A. Antonopoulos, L. Alonso, virtual networks functions. IEEE Transactions on
and C. Verikoukis. Fairness in multi-operator en- network and service management, 13(2):240–252,
ergy sharing. In 2017 IEEE International Confer- 2016.
ence on Communications (ICC), pages 1–6. IEEE,
2017. [123] B. Rouzbehani, L. M. Correia, and L. Caeiro.
Radio resource and service orchestration for vir-
[113] J. Ordonez-Lucena, P. Ameigeiras, D. Lopez, J. J. tualised multi-tenant mobile Het-Nets. In 2018
Ramos-Munoz, J. Lorca, and J. Folgueira. Net- IEEE Wireless Communications and Networking
work slicing for 5G with SDN/NFV: Concepts, Conference (WCNC), pages 1–5. IEEE, 2018.

© International Telecommunication Union, 2020 155

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[124] O. Sallent, J. Perez-Romero, R. Ferrus, and ing market. In 2016 IEEE International Confer-
R. Agusti. On radio access network slicing from ence on Communications (ICC), pages 1–6. IEEE,
a radio resource management perspective. IEEE 2016.
Wireless Communications, 24(5):166–174, 2017.
[135] F. Teng, D. Guo, and M.-L. Honig. Sharing of un-
[125] K. Samdanis, X. Costa-Perez, and V. Sciancale- licensed spectrum by strategic operators. In IEEE
pore. From network sharing to multi-tenancy: The Global Conference for Signal Processing and Com-
5G network slice broker. IEEE Communications munications (GlobalSIP), pages 288–292, Decem-
Magazine, 54(7):32–39, 2016. ber 2014.

[126] T. Sanguanpuak, S. Guruacharya, E. Hossain, [136] T. D. Tran and L. B. Le. Resource alloca-
N. Rajatheva, and M. Latva-aho. Infrastructure tion for efficient bandwidth provisioning in vir-
sharing for mobile network operators: analysis of tualized wireless networks. In 2017 IEEE Wire-
trade-offs and market. IEEE Transactions on Mo- less Communications and Networking Conference
bile Computing, 17(12):2804–2817, 2018. (WCNC), pages 1–6. IEEE, 2017.

[127] SAPHYRE. Deliverable D5.5, Business models, [137] S. Valentin, W. Jamil, and O. Aydin. Extending
cost analysis and advices for spectrum policy and generalized processor sharing for multi-operator
regulation for scenario III (full sharing), 2013. scheduling in cellular networks. In Wireless Com-
munications and Mobile Computing Conference
[128] O. Semiari, W. Saad, M. Bennis, and Z. Dawy. (IWCMC), 2013 9th International, pages 485–490.
Inter-operator resource management for mil- IEEE, 2013.
limeter wave multi-hop backhaul networks.
IEEE Transactions on Wireless Communications, [138] J. van de Belt, H. Ahmadi, L. E. Doyle, and
16(8):5258–5272, 2017. O. Sallent. A prioritised traffic embedding mecha-
nism enabling a public safety virtual operator. In
[129] M.-K. Shin, S. Lee, S. Lee, and D. Kim. A way 2015 IEEE 82nd Vehicular Technology Conference
forward for accommodating NFV in 3GPP 5G sys- (VTC2015-Fall), pages 1–5. IEEE, 2015.
tems. In 2017 International Conference on Infor-
mation and Communication Technology Conver- [139] F. Vaz, P. Sebastiao, L. Goncalves, and A. Cor-
gence (ICTC), pages 114–116. IEEE, 2017. reia. Femtocell deployment in LTE-A networks: A
sustainability, economical and capacity analysis.
[130] F. Shirzad and M. Ghaderi. Cloud-based spec- In IEEE 24th International Symposium on Per-
trum sharing in virtual wireless networks. In 2016 sonal Indoor and Mobile Radio Communications
IEEE 24th International Symposium on Model- (PIMRC), pages 3423–3427, September 2013.
ing, Analysis and Simulation of Computer and
Telecommunication Systems (MASCOTS), pages [140] D. P. Venmani, Y. Gourhant, and D. Zeghlache.
196–204. IEEE, 2016. Divide and share: A new approach for optimiz-
ing backup resource allocation in LTE mobile
[131] J. Simo-Reigadas, E. Municio, E. Morgado, networks backhaul. In 2012 8th International
E. M. Castro, A. Martinez, L. F. Solorzano, and Conference on Network and Service management
I. Prieto-Egido. Sharing low-cost wireless infras- (CNSM) and 2012 Workshop on Systems Vir-
tructures with telecommunications operators to tualization Management (SVM), pages 189–193.
bring 3g services to rural communities. Computer IEEE, 2012.
Networks, 93:245–259, 2015.
[141] D. P. Venmani, Y. Gourhant, and D. Zegh-
[132] H. M. Soliman and A. Leon-Garcia. A novel neuro- lache. ROFL: Restoration of failures through
optimization method for multi-operator schedul- link-bandwidth sharing. In 2012 IEEE Globecom
ing in cloud-RANs. In 2016 IEEE International Workshops, pages 30–35. IEEE, 2012.
Conference on Communications (ICC), pages 1–6.
IEEE, 2016. [142] J. Village, K. Worrall, and D. Crawford. 3g shared
infrastructure. In Third International Conference
[133] H. M. Soliman and A. Leon-Garcia. QoS- on 3G Mobile Communication Technologies (Conf.
aware frequency-space network slicing and ad- Publ. No. 489), pages 10–16. IET, 2002.
mission control for virtual wireless networks. In
2016 IEEE Global Communications Conference [143] M. Vincenzi, A. Antonopoulos, E. Kartsakli,
(GLOBECOM), pages 1–6. IEEE, 2016. J. Vardakas, L. Alonso, and C. Verikoukis. Coop-
eration incentives for multi-operator C-RAN en-
[134] F. Sun, B. Liu, F. Hou, H. Zhou, L. Gui, and ergy efficient sharing. In 2017 IEEE International
J. Chen. Cournot equilibrium in the mobile vir- Conference on Communications (ICC), pages 1–6.
tual network operator oriented oligopoly offload- IEEE, 2017.

156 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

[144] P. L. Vo, M. N. Nguyen, T. A. Le, and N. H. An auction approach. Computer Communications,

Tran. Slicing the edge: Resource allocation for 114:22–35, 2017.
RAN network slicing. IEEE Wireless Communi-
cations Letters, 7(6):970–973, 2018. [155] K. Zhu, Z. Cheng, B. Chen, and R. Wang. Wire-
less virtualization as a hierarchical combinatorial
[145] S. Wang, K. Samdanis, X. C. Perez, and auction: An illustrative example. In 2017 IEEE
M. Di Renzo. On spectrum and infrastructure Wireless Communications and Networking Con-
sharing in multi-operator cellular networks. In ference (WCNC), pages 1–6. IEEE, 2017.
2016 23rd International Conference on Telecom-
munications (ICT), pages 1–4. IEEE, 2016. [156] K. Zhu and E. Hossain. Virtualization of 5g cellu-
lar networks as a hierarchical combinatorial auc-
[146] J. Wei, K. Yang, G. Zhang, and Z. Hu. Pricing- tion. IEEE Transactions on Mobile Computing,
based power allocation in wireless network virtu- 15(10):2640–2654, 2016.
alization: A game approach. In 2015 International
Wireless Communications and Mobile Comput-
ing Conference (IWCMC), pages 188–193. IEEE,
2015.

[147] J. Wei, K. Yang, G. Zhang, and X. Lu. A QoS-

Aware Joint Power and Subchannel Allocation Al-
gorithm for Mobile Network Virtualization. Wire-
less Personal Communications, 104(2):507–526,
2019.

[148] D. Wu, Z. Zhang, S. Wu, J. Yang, and R. Wang.

Biologically inspired resource allocation for net-
work slices in 5G-enabled Internet of Things.
IEEE Internet of Things Journal, 6(6):9266–9279,
2018.

[149] Y. Xiao, Z. Han, C. Yuen, and L. A. DaSilva. Car-

rier aggregation between operators in next genera-
tion cellular networks: A stable roommate market.
IEEE Transactions on Wireless Communications,
15(1):633–650, 2016.

[150] Y. Xiao, C. Yuen, P. Di Francesco, and L. A.

DaSilva. Dynamic spectrum scheduling for car-
rier aggregation: A game theoretic approach. In
Communications (ICC), 2013 IEEE International
Conference on, pages 2672–2676. IEEE, 2013.

[151] D. Xu and Q. Li. Resource allocation in wireless

virtualized networks with energy harvesting. In
2016 IEEE International Conference on Commu-
nication Systems (ICCS), pages 1–6. IEEE, 2016.

[152] Y. Zaki, L. Zhao, C. Goerg, and A. Timm-Giel.

LTE mobile network virtualization: Exploiting
multiplexing and multi-user diversity gain. Mobile
Networks & Applications, 16(4):424–432, August
2011.

[153] J. Zheng, P. Caballero, G. De Veciana, S. J.

Baek, and A. Banchs. Statistical multiplex-
ing and traffic shaping games for network slic-
ing. IEEE/ACM Transactions on Networking,
26(6):2528–2541, 2018.

[154] R. Zhou, X. Yin, Z. Li, and C. Wu. Virtualized

resource sharing in cloud radio access networks:

© International Telecommunication Union, 2020 157

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

AUTHORS extensively for industry and the mainstream media and

has been part of major international committees. She
leads the LORLAB, a research group dedicated to de-
veloping effective applied mathematics methods to the
design and performance of wireless and wireline telecom-
munication networks. She has a special interest in net-
work robustness and sustainability.

L. Cano (BSc. 12, MSc. 14, PhD 20) is currently a re-

searcher at Politecnico di Milano. She recently obtained
a double degree PhD at Politecnico di Milano and Poly-
technique Montréal. Her main research interests are in
the techno-economic characterization of infrastructure
sharing in networks based on game theoretical models.

A. Capone (PhD 1998) is full professor at Politec-

nico di Milano where is he also the Dean of the School
of Industrial and Information Engineering, member of
the university strategy team POLIMI2040, and direc-
tor of the Advanced Network Technologies Laboratory
(ANTLab). He is associate editor with IEEE Trans. on
Mobile Computing and Elsevier Computer Communi-
cations, and member of the TPC of major conferences
in networking. His research interests include radio re-
source management and planning of wireless networks,
software defined networks and switching architectures.
On these topics he has published more than 300 papers.
He is a fellow of the IEEE.

B. Sansò (Ph.D.89) is a full professor of telecommu-

nication networks in the department of Electrical En-
gineering of Polytechnique Montréal and a member of
GERAD, a world-renowned applied mathematics re-
search center. Over her long career, she has received
many awards and honors, has published and consulted

158 © International Telecommunication Union, 2020

 ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

INDEX OF AUTHORS
A L
Agustí, Ramon ........................................... 103 Liaskos, Christos .......................................... 55
Akyildiz, Ian F. ............................................ 55 Lin, Zhiqiang ................................................ 89
Armada, Ana Garcia .................................... 13 Liu, Fang ....................................................... 89
Arslan, Huseyin .......................................... 121 Liu, Yong ..................................................... 13
Athalye, Akshay .............................................. 1 Loscrí, Valeria ............................................. 79
Atzori, Luigi ................................................. 37
M
C Milotta, Giuseppe Massimiliano ................... 37
Campolo, Claudia ........................................ 37 Morabito, Giacomo ....................................... 37
Cano, Lorela ............................................... 141
O
Capone, Antonio ........................................ 141
Ojaroudi, Mohammad .................................. 79
Chen-Hu, Kun .............................................. 13
Christodoulou, Michail ................................ 55
P
Pérez-Romero, Jordi ................................... 103
D
Pitilakis, Alexandros .................................... 55
Das, Samir R. ................................................. 1
Pitsillides, Andreas ...................................... 55
Djurić, Petar M................................................ 1
Pompili, Dario............................................... 25
F Pyrialakos, Georgios G. ............................... 55
Ferrús, Ramon ............................................ 103
Q
H Quattropani, Salvatore .................................. 37
Haas, Zygmunt J. ........................................... 1
R
Hajisami, Abolfazl ....................................... 25
Ren, Wenbo .................................................. 89
I S
Iera, Antonio ................................................ 37
Sallent, Oriol .............................................. 103
Ioannidis, Sotiris .......................................... 55
Sansò, Brunilde .......................................... 141
K Shroff, Ness B. .............................................. 89
Kantartzis, Nikolaos. ..................................... 55 Singh, Rahul ................................................ 89
Stanaćević, Milutin ......................................... 1

© International Telecommunication Union, 2020 159

ITU Journal on Future and Evolving Technologies, Volume 1 (2020), Issue 1

T
Tsioliaridou, Ageliki .................................... 55
Tusha, Seda Doğan .................................... 121

V
Vegni, Anna Maria ....................................... 79
Vilà, Irene ................................................... 103

X
Xuan, Dong .................................................. 89

Y
Yazar, Ahmet ............................................. 121

160 © International Telecommunication Union, 2020

 International
Telecommunication
Union
Telecommunication
Standardization Bureau (TSB)
Place des Nations
CH-1211 Geneva 20
Switzerland

ISSN: 2616-8375
Published in Switzerland
Geneva, 2020