Proyecto Fin de Carrera

Trabajo Fin de Máster

Ingeniería
Máster de
en Telecomunicación
Ingeniería de Telecomunicación

Formato de Publicación
An Open de la for
Source Project Escuela Técnica
Learning 5G
Superior de Ingeniería
Waveforms at an Electrical Engineering
Graduate Level

Autor:Autor: Guillermo
F. Javier Palomino Lozano
Payán Somet
Tutor:Tutor: Juan Murillo
Juan José AntonioFuentes
Becerra González

Dep.
Dep. de Teoría
Teoría de la de la Señal
Señal y Comunicaciones
y Comunicaciones
Escuela
Escuela Técnica
Técnica Superior
Superior de Ingeniería
de Ingeniería
Universidad
Universidad de Sevilla
de Sevilla
Sevilla, 2020
Sevilla, 2013
 Trabajo Fin de Máster
Máster en Ingeniería de Telecomunicación

An Open Source Project for Learning 5G

Waveforms at an Electrical Engineering
Graduate Level

Autor:

Guillermo Palomino Lozano

Tutor:

Juan Antonio Becerra González

Profesor Sustituto Interino

Dep. de Teoría de la Señal y Comunicaciones

Escuela Técnica Superior de Ingeniería
Universidad de Sevilla
Sevilla, 2020
 Trabajo Fin de Máster: An Open Source Project for Learning 5G Waveforms at an Electrical
Engineering Graduate Level

Autor: Guillermo Palomino Lozano

Tutor: Juan Antonio Becerra González

El tribunal nombrado para juzgar el trabajo arriba indicado, compuesto por los siguientes profesores:

Presidente:

Vocal/es:

Secretario:

acuerdan otorgarle la calificación de:

El Secretario del Tribunal

Fecha:
 Acknowledgements

C on este documento termina mi formación educativa en un campo tan apasionante como es la Ingeniería
de Telecomunicaciones. Atrás quedan 6 hermosos años, en los que he ido recorriendo un camino
fascinante descubriendo y aprendiendo de este mundo. Y te vienen a la mente muchísimas personas que, en
mayor o menor medida, te han ayudado o han formado parte de esta aventura.
En primer lugar, los profesores. Gracias a todos y cada uno de ellos, que con su esfuerzo consiguen que
año tras año muchas personas salgan formadas como ingenieros, la profesión más bonita del mundo. Mención
especial merece el tutor de este trabajo, Juan Antonio, que siempre ha estado ahí al pie del cañón. Siempre te
estaré agradecido por todo tu empeño y dedicación a lo largo de toda mi formación académica. También me
gustaría agradecer a la profesora María José Madero, que colaboró en la redacción del artículo asociado a
este trabajo, y también por su extraordinaria docencia durante los últimos años.
En segundo lugar, los compañeros. Este proceso no solo te forma para poder ejercer tu profesión, también te
forma para ser mejor persona. Y gran culpa de ello depende de la gente de la que te rodeas. Y, afortunadamente,
me he sentido muy bien rodeado durante estos 6 años. Compañeros de clase convertidos en grandes amigos
que se convertirán en compañeros de profesión. La unión hace la fuerza, creo que ha quedado más que
demostrado. El Team Polinino nunca morirá.
En tercer lugar, merece mención especial este último año de mi vida, donde gracias a la beca Erasmus
he podido vivir en el extranjero el mejor año de mi vida. Sin duda, la experiencia vivida allí, así como la
formación recibida en una universidad extranjera, una vez más han servido para crecer aún más como futuro
ingeniero y como persona. Gracias a todos y cada uno de vosotros, chicxs.
Por último, la familia. Esos que siempre están ahí y nunca se irán, que te agarran cuando te caes, que
te echan una mano cuando todo parece desmoronarse. Gracias papás por educarnos tanto a mí como a mi
hermana en unos valores que quizá no entendimos de pequeño, pero que sin embargo ahora sí que entendemos
e intentaremos transmitir a las futuras generaciones: educación y respeto. Y gracias a ti, Sara, por entrar en
mi mundo hace ya más de 7 años y hacerme la persona más feliz del mundo.
Mi etapa educativa finaliza en un momento de muchísima incertidumbre, en la que nuestras vidas se han
visto alteradas en los últimos meses de forma drástica, marcadas por la pandemia COVID-19. Habrá cosas
que nunca volverán a ser como antes, pero la ilusión y motivación permanecerán igual que el primer día.
THE END.

Guillermo Palomino Lozano

Alumno de 2º de MUIT
Entre Cáceres y Lund, 2020

I
 Resumen

E l presente documento describe la implementación de un software libre y de código abierto para la

generación de formas de onda 5G-NR. El aprendizaje de las tecnologías de comunicaciones digitales
requiere experiencia, y los estudiantes a menudo exigen contenidos más prácticos para entender los con-
ceptos básicos de dichas asignaturas. Paralelamente, el desarrollo comercial de la quinta generación de
comunicaciones móviles (5G) comenzó en 2020. 5G pretende ser una tecnología inalámbrica revolucionaria
en términos de latencia, velocidad de datos, eficiencia energética y comunicación masiva. Los estudiantes
deberán saber cómo funciona la capa física de 5G, conocida como 5G New Radio (5G-NR) y, por lo tanto,
deben hacer frente a los desafíos de diseño del sistema que existen en esta tecnología. La herramienta
presentada está contextualizada en el marco de las Ingenierías de Telecomunicación. Además, el software
presentado aquí también será útil para los investigadores que podrán utilizar las formas de onda de 5G-NR
para sus propios proyectos de investigación.

III
 Abstract

T his paper describes the implementation of a free and open-source software for 5G-NR waveform gen-
eration. Learning digital communication technologies skills requires experience and the students often
demand more practical contents in order to understand the basic concepts of their courses. Linked to this,
the commercial development of the fifth generation wireless system (5G) started in 2020. 5G aims to be a
revolutionary wireless technology in terms of latency, data rates, energy efficiency and massive communica-
tion. Digital Communication students must know how 5G works and, therefore, they must study its physical
layer and the system design challenges that exist in this technology. The presented tool is contextualized in
Engineering curricula, more precisely in a program named Electrical and Computer engineering. This tool
developed will also be helpful to researchers that will be able to use 5G-NR waveforms for their projects.

V
 Summarized contents

Resumen III

Abstract V

Summarized contents VII

Abbreviations and Acronyms XI

1 Introduction 1
1.1 Scope and challenges 2
1.2 Planning 3
1.3 Structure of this work 4
1.4 Software employed 5

2 5G New Radio Overview 7

2.1 Motivation 7
2.2 OFDM as 5G-NR waveform 7
2.3 Numerology 8

2.4 Duplex schemes 9

2.5 Time domain 10

2.6 Frequency domain 11

3 Design of the proposed tool 13

3.1 Software designed 13

3.2 Code availability 19

4 Results and discussions 23

4.1 Simulations and results 23

4.2 Conclusions and future lines 31

Appendix A Implementation of the project 33

VII
VIII Summarized contents

A.1 Simulation script that uses the toolbox presented in this work 33
A.2 Main 5G-NR toolbox script for waveform generation 37
A.3 Some other util functions 44
A.4 Extra code for some figures generation 57

List of Figures 61
List of Tables 63
List of Codes 65
Bibliography 67
 Contents

Resumen III
Abstract V
Summarized contents VII
Abbreviations and Acronyms XI

1 Introduction 1
1.1 Scope and challenges 2
1.2 Planning 3
1.3 Structure of this work 4
1.4 Software employed 5

2 5G New Radio Overview 7

2.1 Motivation 7
2.2 OFDM as 5G-NR waveform 7
2.3 Numerology 8
2.3.1 Discrete Fourier Transform 8
2.3.2 Cyclic Prefix 9
2.4 Duplex schemes 9
2.4.1 Time-Division Duplex (TDD) 9
2.4.2 Frequency-Division Duplex (FDD) 10
2.4.3 Half-Duplex FDD 10
2.5 Time domain 10
2.6 Frequency domain 11

3 Design of the proposed tool 13

3.1 Software designed 13
3.1.1 Flowchart description 13
3.1.2 Block diagram 15
Random data generator 15
Modulator 16
Oversampling 16
IFFT 17
Cyclic prefix 18
Peak clipping 18
Spectrum shaping filter 18
3.2 Code availability 19
3.2.1 Codeocean 20
3.2.2 PyPI 21
3.2.3 Github 22

IX
X Contents

4 Results and discussions 23

4.1 Simulations and results 23
4.1.1 Simulation number 1: Multicarrier generation stage 23
Analysing the signal in the time domain 24
Peak clipping stage motivation 25
4.1.2 Simulation number 2: AWGN channel and most typical figures of merit 25
Bit Error Rate (BER) 26
Error Vector Magnitude (EVM) 27
4.1.3 Simulation number 3: Sending the generated signal to an external RF
WebLab 27
RF Weblab at Chalmers University 27
Programming the script and simulation 28
4.2 Conclusions and future lines 31

Appendix A Implementation of the project 33

A.1 Simulation script that uses the toolbox presented in this work 33
A.2 Main 5G-NR toolbox script for waveform generation 37
A.3 Some other util functions 44
A.4 Extra code for some figures generation 57

List of Figures 61
List of Tables 63
List of Codes 65
Bibliography 67
 Abbreviations and Acronyms

3GPP Third-Generation Partnership Project

5G Fifth Generation wireless system
AW GN Aditive White Gaussian Noise
BER Bit Error Rate
CDMA Code-division Multiple Access
DPD Digital Pre-Distorsion
eMBB enhanced Mobile Broadband
FDD Frequency Division Duplex
FOSS Free and Open-Source Software
FR Frequency Range
GSM Global System for Mobile Communication
ICI Intercarrier Interference
ISI Intersymbol Interference
IoT Internet of Things
IMT − 2020 International Mobile Telecommunications 2020
ITU − R International Telecommunications Union - Radio communica-
tions sector
LT E Long Term Evolution
mMTC massive Machine Type Communications
M2M Machine-To-Machine
NMSE Normalised Mean Squared Error
NR New Radio
OFDM Orthogonal Frequency Division Multiplexing
PAPR Peak to Average Power Ratio
PyPI Python Package Index
RB Resource Block
RE Resource Element
T DD Time-Division Duplex
UMT S Universal Mobile Telecommunications System
URLLC Ultra-Reliable and Low-Latency Communications

XI
 1 Introduction

Why should I try to make you believe the things I believe in?

Dennis Rodman, a controversial NBA basketball player

M obile communications have redefined the relationship between human beings in the last 40 years
in terms of society and technology. Everything started in 1973 when Martin Cooper (a Motorola
executive) made the first phone call ever. At that moment a race for bringing this new technology to the user
market started, and it meant that around 1980 the first generation of mobile communication emerged. This
very first system only allowed voice serviced but, for the first time in history, allowed ordinary people to
make phone calls to every part of the world.
The second generation of mobile communication (2G) came out in the first years of the 1990s. But there
was a lack of a unified system for serving calls all over the world. One of the most famous systems was the
Global System for Mobile Communication (GSM) that was developed by some European countries. GSM
dominated the implementation of 2G because it spread to other non-European countries and it is considered
the first global system for mobile communication. 2G was mainly designed for serving voice calls but it
included digital transmission to provided limited data services.
People started to use a cell phone in their daily lives so it was necessary to introduce a new generation.
In early 2000 the third generation of mobile communication (3G) was born, with the main difference that
now it uses packet switching instead of circuit switching for data transmission. This technology was known
as High-Speed Packet Access (HSPA) [GMB, 2006]. Additionally, a new core network architecture was
completely designed, called Universal Mobile Telecommunications System (UMTS), and the channel access
technique used was Code-Division Multiple Access (CDMA). Therefore, a big step of quality was taken
and users observed how data rates increased. A new era had just begun, and users started to connect to the
Internet with their cellphones.
The fourth generation (4G) was deployed in 2013 and went one step further providing not only higher data
rates but also more devices connected. 4G is a technology that understands that the number of devices in
the society is growing exponentially and therefore the services that those devices can offer are more than
just calls and messages. Cellphones are now smartphones, and they demand a stable Internet connection for
video streaming, social media, augmented reality, etc. This generation involves the Long Term Evolution
(LTE) technology and the channel access is Orthogonal Frequency Division Multiplexing (OFDM).
Almost half a century after Martin Cooper made the first call, the fifth generation (5G) wireless sys-
tem deployment is starting. A summarized evolution of the different generations of mobile communica-
tions is shown in Fig. 1.1. But 5G is not just an upgrade of last generations of mobile communication
[Gupta and Jha, 2015, Ghosh et al., 2019]. Far from being only an evolution of the latest generation wireless
technology, 5G is a system with new services capabilities that expect to meet the demands imposed by the
users in the coming years. According to [Cis, 2020], by 2023 a 66% of the population will have Internet
access —up from 51% in 2018— and the 29.3 billion devices connected to this global network will represent
more than thrice the global population. Additionally, there will be an increase in the number of devices in
machine-to-machine (M2M) communication, growing from 33% of the total devices in 2018 to 50% by

1
2 Chapter 1. Introduction

2023. The increment in data rates and the number of connected devices to legacy mobile generations pose a
challenge to standards that are fulfilled with 5G [Ijaz et al., 2016, Henry et al., 2020].

Figure 1.1 Road to 5G [Remmert, 2019].

After initial 5G launches in 2019, it is expected that commercial telecommunication companies start to offer
5G connection in most of their territory in the following years. The key for a successful technology is in its
process of standardization that must allow interoperability between infrastructure and devices from different
companies. The specifications of the new standard have been created by the Third-Generation Partnership
Project (3GPP), as they did with previous cellular technologies such as Universal Mobile Telecommunica-
tions System (UMTS) and Long-Term Evolution (LTE) [ROA, 2020]. 3GPP is following the International
Telecommunications Union - Radio communications sector (ITU-R) guidelines, submitting a candidate
technology that will be evaluated following the International Mobile Telecommunications 2020 (IMT-2020)
requirements [Marcus, 2015]. Taking into account the necessities of the society today, IMT-2020 specified
three use case scenarios [ITU-R, 2015]:
• Enhanced mobile broadband (eMBB): Requirement of high data rates across a wide coverage area.
Since the end-user data-rate demand is continuously increasing, higher frequency deployment is
particularly suitable for dedicating more bandwidth for the transmission.
• Ultra-reliable and low-latency communications (URLLC): This case is considered for scenarios where
latency requirement is crucial. Additionally, very high availability and reliability must be guaranteed.
Examples of scenarios that cover this situation are those related to autonomous driving and automatic
control.
• Massive machine-type communications (mMTC): services that require a large number of entities
connected simultaneously. These low-cost devices such as sensors or actuators consume ultra-low
power. Their requirements also include ultra-low latency, high reliability, and availability.

1.1 Scope and challenges

This work presents a free and open-source software (FOSS) platform for learning 5G New Radio (5G
NR), the radio access technology of 5G, that is intended for its use at undergraduate and graduate-level
courses. Previous studies have presented FOSS frameworks for 5G deployment, but focusing on other
aspects, such as a techno-economic assessment [Gomes et al., 2018] or communication between networks
 1.2 Planning 3

[Lin et al., 2017, Kaltenberger et al., 2019, Salama and Elmesalawy, 2019]. This software is developed in
the Python programming language and it is publicly available within SciPy, the Python-based ecosystem of
open-source software that is a common tool for laboratory activities at engineering programs, and Codeocean,
a research collaboration platform by IEEE.
The objectives of this work can be enumerated as follow:
1. Study the origin of the mobile communication systems and how they have progressed during the last
decades, highlighting in the most recent technologies (LTE and 5G NR).
2. Try to know better about how 5G-NR waveforms are generated and how to choose the parameters
needed for the transmission efficiency.
3. Develop a FOSS platform for learning 5G-NR waveforms using Python programming language.
4. Learn which are the most important figures of merit in 5G NR and try to understand which values are
considered as admissible parameters in this technology.
5. Provide an on-line tool for future students that helps them to understand how 5G NR works.
6. Bring a closed-box package that generates 5G-NR signals that researchers and users, in general, can
use to develop their simulations and experiments.

1.2 Planning

Since this works consists of a real project that is going to be implemented, it was considered to create a Gantt
chart to schedule and define all the tasks that had to be performed. The fact of creating this chart was helpful
because it allowed me to separate tasks, and it was also helpful to the supervisor, that knew in every moment
how was the project status. The Gantt chart is represented in Fig. 1.2. It is possible to identify two different
groups of tasks. One called software development and another one called documentation and publication
then all the documentation and publication process. All tasks are explained in detail next:
• Software development.
– Problem definition. This part consisted of some meetings where the project supervisor explained
to the student the problem they had and the reasons why they wanted to develop a platform like
this.
– 5G-NR readings. It is true that the student had some previous knowledge in 5G NR but it was
necessary to read some books to get background knowledge of how exactly 5G-NR standardization
process is doing. Once all this knowledge was gotten, I was ready to start coding.
– Toolbox implementation. This task is the main one of this part, where all implementation is
done taking more than two months. This task was coordinated with the next one, which is the
verification and simulation process. As the reader can think, the verification process will return
modifications in the toolbox: that is the reason why both share space in time.
– Verification and simulations. Once most of the tool is developed, it is convenient to start with
the simulations that will be explained in the following chapters. Additionally, some tests were
performed to verify the system developed.
• Documentation and publication.
– IEEE Access paper preparation. It was convenient to write a paper explaining the tool per-
formed and the impact it will have on Electrical Engineering students. Therefore, a task was
defined, whose objectives were to start explaining and defining the software.
– Pypi package preparation. Additionally, the tool was uploaded to the Pypi package. It needs
some modifications regarding the structure and some files that must be defined.
– Codeocean capsule preparation. Doing the same way, the purpose of this task is to adapt the
tool to the Codeocean format.
4 Chapter 1. Introduction

– Thesis elaboration. Finally, this project finishes with the thesis elaboration, that is this document
that you are reading. The thesis elaboration includes a defense with a presentation that must be
also prepared in these tasks.

WEEKS: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

1 Software development

1.1 Problem definition

1.2 5G-NR readings

1.3 Toolbox implementation

1.4 Verification and simulations

2 Documentation and publication

2.1 IEEE Access paper preparation

2.2 Pypi package preparation

2.3 Codeocean capsule preparation

2.4 Thesis elaboration

Figure 1.2 Gantt diagram associated to the project planning.

1.3 Structure of this work

This thesis is organized into four chapters, where the first one is the one you are reading now, and it is an
introduction for the rest of the work. As you can understand, this first chapter is useful for the reader in terms
of motivation and understanding. The rest of this is organized in three chapters, very different from each
other:

• Chapter 2: 5G New Radio overview. This second chapter presents an overview of 5G-NR technology
on which this project is based. We will motivate the use of numerology, key for considerate several
usage scenarios. A brief recap of how OFDM works is also included here, and how 5G-NR adapt
OFDM for their transmissions. Signal generation, both frequency and time domain is also considered
here. This chapter finishes with some necessary definitions that everyone who works with 5G NR
should be familiar with.

• Chapter 3: design of the proposed tool. The complete description of the designed tool is introduced
in this chapter. The design is organized in seven different blocks: random data generator, modulator,
oversampling, Inverse Fast-Fourier Transform (IFFT), Cyclic Prefix, Peak clipping, and Spectrum
shaping filter. Every block is explained in detail in a mathematical way, related at the same time with
the code that has been developed. This chapter is the key to the whole thesis and it is a basic one that
the reader must understand before reading the next one, that is where all simulations and results are
presented.

• Chapter 4: results and discussions. Once the tool has been completely explained, the next step is to
present all the simulations that have been carried out and the results that are extracted after that. The
tool presented here allows a lot of functionalities, therefore a variety of different simulations may be
performed. The motivation for creating new simulations is also explained here. This chapter ends with
some future lines that could continue the work done here and also some conclusions.
 1.4 Software employed 5

1.4 Software employed

For the platform implementation, the chosen programming language was Python. The reasons why Python
was selected are explained next, but it is also important to know that the author has already some experience
with this programming language. Hence, most of the work is to focus on developing the tool itself and not
learning the programming language. However, there is always something new that must be learned, like
new libraries or some code techniques that the developer must understand and therefore spend some time on
learning those tasks. In any way, some of the Python features that are helpful for this particular project are
[Oliphant, 2007, Fangohr, 2004, Ozgur et al., 2017]:
• It is a FOSS, so there is an active community of developers continuously creating projects that users can
employ for their activities. Hence, it helps to reduce development costs significantly. It is commonly
known that Python has strengthened the past years. If you go search in Google which are the most
famous programming languages Python will be for sure the top 3 in each of them.
• There are a large available number of libraries that help to optimize the code. Some of the most
important libraries and packages for signal processing used in this project are: Numpy, that pro-
vides support for matrices and vectors enabling efficient implementation in a high-level language
[van der Walt et al., 2011]; Matplotlib, a data visualization package with a graphical interface similar
to the Matlab programming language that provides high-quality output in many formats, including
portable document format (PDF), encapsulated postscript (EPS), and portable network graphics (PNG)
[Ari and Ustazhanov, 2014]; and SciPy, more for a general purpose on scientific and technical comput-
ing [Oliphant, 2004]. The latter is part of the Numpy project.
• The Python Package Index (PyPi) is the official software repository for Python, where hundreds of
thousands of developers upload their packages. Users can easily download, install a package, and
add them to their projects. Precisely, this project is uploaded to the PyPi, so it is easy to install and
download for anyone.
• Users can run external applications with Python, as well as embed Python into another one. Conse-
quently, the abstraction level is increased. As an example, in this project, an application developed in
Matlab located in an external server is run.
To put it briefly, Python is one of the most important programming languages in the world of today, and it
also applies in mobile communication where every day more and more projects are developed in Python
because of its robustness.
 2 5G New Radio Overview

Good ideas are always crazy until they are not.

Elon Musk, Billionaire CEO of SpaceX and Tesla.

2.1 Motivation

T he purpose of this second chapter is to give a 5G background before start commenting on how devel-
opment tool has been designed. Our starting point is the fourth generation of mobile communication,
LTE. One reason for doing that is that both are developed by the same organization, 3GPP. Hence, 5G NR
can be considered as an upgrade of LTE in terms of radio access, providing the following benefits:
• Lower latency, that allows new scenarios.
• It is possible to exploit higher frequencies, which is directly related to the use of more spectrum for
getting more data rates.
• Researchers have focused on what is known as ultra-lean design. It improves network energy efficiency,
managing the transmissions in a way that the "always-on" transmissions are minimized, according to
an Ericsson report [Zaidi, 2017]. This technique is without any doubt an advantage because of two
reasons: first, it is better for a sustainable society and second, it reduces operational expenses.
• 5G NR is prepared for forwarding compatibility, for future use cases and technologies.
• If we have to describe 5G-NR technology with one word, it is clear: flexibility. It will be mentioned
during all the chapters but 5G NR must address all of the different use cases already mentioned in the
last chapter (eMBB, mMTC, URLLC), so a range of carrier frequencies, order modulation and, in
general, different transmission schemes must be defined dynamically. The software that is presented in
this project covers most of these aspects, see next chapter for further details.

2.2 OFDM as 5G-NR waveform

Following the same procedure which was done with the development of LTE, orthogonal frequency division
multiplexing (OFDM) was chosen as the waveform for 5G-NR transmissions [Guan et al., 2017]. OFDM is
primarily chosen due to its spectrum efficiency, relative immunity to selective fading, and simple channel
equalization. Additionally, it is also easy to exploit both the time and frequency domains. OFDM allows
dividing the available bandwidth into parallel subchannels that are called subcarriers. Some of the advantages
of using OFDM include:
• No equalization of delay spread is needed. As a consequence, the OFDM received symbols does not
experience any ISI.

7
8 Chapter 2. 5G New Radio Overview

Table 2.1 Numerology in 5G NR.

TCP [µs] for

µ ∆ f [kHz] Tu [µs]
symbols 1 to 6
0 15 66.7 4.7
1 30 33.3 2.3
2 60 16.7 1.2
3 120 8.33 0.59
4 240 4.17 0.29

• It is a modulation that is robust to selective fading. The channel is divided into orthogonal signals: a
selective fading is therefore converted into a flat fading for one subcarrier.

• High spectral efficiency. The subcarriers are close to each other which is good for spectral efficiency.

But OFDM has also some disadvantages, such as high Peak to Average Power Ratio (PAPR), which is
defined as the ratio between the peak power and the average power of a signal, commonly expressed in
decibels. As a consequence of a high PAPR, there will be a high dynamic range, therefore a highly linear
power amplification will be needed. Amplifiers with a high dynamic range usually are power inefficient.
Another technique commonly used is to reduce PAPR using peak clipping techniques. This one is the one
that is implemented in this project. Another disadvantage of OFDM is that it is sensitive to carrier frequency
offset, that arises a phenomenon known as Intercarrier Interference (ICI), that deteriorates the quality of the
transmission.

The physical layer designed in 5G NR allows OFDM for both the downlink and uplink. Additionally,
discrete Fourier transform (DFT) precoded OFDM can be used as a waveform in the uplink due to its
efficiency in energy-limited scenarios. According to [Truong et al., 2014], the energy consumption of the
power amplifier may be reduced by a factor of two. But the use of DFT-precoded OFDM also has some
drawbacks. To give an example, spatial multiplexing becomes more complex, and also using this scheme the
possibility of allocating resources in non-contiguous frequencies is lost, so it implies scheduling restrictions
as well as losing frequency diversity.

2.3 Numerology

In a 5G-NR context, the term numerology is associated to the design parameters that can be defined for a
specific waveform. In LTE, there is only one type of subcarrier spacing of 15 kHz. In 5G this is completely
different since one of the main challenges is to support different use case scenarios [Osseiran et al., 2014]
therefore it is not possible to have just one numerology. A scalable numerology is needed. As a result, a
completely new one is defined and it is shown in Table 2.1. To give an explanation of how numerology works
in 5G NR, it is convenient to discuss regarding the definition of Discrete Fourier Transform operator (DFT).

2.3.1 Discrete Fourier Transform

Waveform generation in 5G NR is based on OFDM, which is straightforward related to the DFT, defined as

1 N−1
X(k) = ∑ x(n)e− j2πkn/N ,
N k=0
(2.1)

where x(n) is the signal represented in a sequence of N complex number in discrete-time domain, and X(k)
the output of the DFT operation, also a sequence of complex numbers. It is also possible to define the inverse
operation, the Inverse Discrete Fourier Transform (IDFT),

1 N−1
x(n) = ∑ X(k)e j2πkn/N . (2.2)
N n=0
 2.4 Duplex schemes 9

The most common algorithm that computes the DFT and the IDFT is called the fast Fourier Transform
(FFT) and the inverse Fourier Transform (IFFT) respectively, and they are widely used in 5G-NR waveform
generation. Each subcarrier is modulated depending on one transmitted symbol, i.e., the modulation is
performed in parallel, as it is shown on Fig. 3.3. It means that every Tu seconds an IFFT with FFTsize points
is performed. The sampling frequency Fs is defined as follows,

Fs = FFTsize · ∆ f , (2.3)
where FFTsize is the FFT size, that must be power of two in order to be computationally efficient, and ∆ f is
the subcarrier spacing, whose expression is

∆ f = 2µ · 15, (2.4)

where µ ∈ [0,4] is the parameter that defines the numerology. Available configurations are shown in Table
2.1.

2.3.2 Cyclic Prefix

After the IFFT operation, the cyclic prefix (CP) is added. In LTE, two types of CP are specified: normal
CP (NCP) and extended CP (ECP). In practice, ECP was not used, so 5G only defines a NCP except for
∆ f = 60 kHz, situation in which both types of CP may be used, so that a CP similar to the ∆ f = 15 KHz case
can be maintained. Following the 3GPP standard [ETS, 2018], the length of the CP, in samples, is defined as
follows,

−µ
512κ · 2
 ECP
NCP = 144κ · 2−µ + 16κ NCP for symbol 0 (2.5)
µ

144κ · 2−µ NCP for symbols 1 to 6




where κ is a constant that relates LTE and 5G basic unit times and takes a value of κ = 64. Note that the first
symbol of every 7 symbols exhibits an extended duration in time. The CP duration follows

TCP = NCP · Tc , (2.6)

where Tc is the basic 5G time unit that is defined as Tc = 1/(480000 · 4096).

Since the program allows to generate a signal with a custom sampling frequency of Fs , it is necessary to
adjust the number of samples accordingly,

0 Tu
NCP = NCP · , (2.7)
Ts
0
where NCP is the new length of the CP after the adjustment. Once the numerology has been defined, the next
step is to create the time structure of the signal. In 5G the transmission schemes are defined in two domains:
time domain and frequency domain. Both are reviewed next.

2.4 Duplex schemes

One of the keys of the wireless systems is how to maintain the communication link in both directions.
Spectrum flexibility is one of the features of NR. It ensures separation of uplink and downlink in frequency
and/or time domains. Duplex schemes defined in NR are Time-Division Duplex (TDD), Frequency-Division
Duplex (FDD), and Half-Duplex FDD. All of them are briefly explained next.

2.4.1 Time-Division Duplex (TDD)

It uses only a single frequency for transmission and reception. Hence, there is not exist and overlap-
ping between uplink and downlink transmissions. What is more, NR uses what is called dynamic TDD
10 Chapter 2. 5G New Radio Overview

[Dahlman et al., 2018]. Dynamic TDD ensures the system to dynamically allocate resources for uplink or
downlink depending on the demand in a certain moment. It means that it is easy to adjust the capacity in
either direction. It is useful in small-cell and isolated cell scenarios. However, for large distances it is not a
good technique: a guard period is added to solves the propagation time issues.

2.4.2 Frequency-Division Duplex (FDD)

It requires one channel for uplink transmissions and another for downlink transmissions. Nevertheless, both
links can be transmitted at the same time. If the capacity must be adjusted, it is possible to re-allocate
channels. This technique solves the problem of the period guard but the complexity is higher (duplex-filters
must be added). Therefore, the costs increase. That is the reason why another duplex scheme is introduced.

2.4.3 Half-Duplex FDD

In this case, the device uses different frequencies for uplink and downlink and they do not transmit at the
same time. This is a simplified technique that reduces device costs, for cases where there are no duplex-filters.
Since the devices cannot transmit in all uplink/downlink subframes, the data rate is reduced in this case.

2.5 Time domain

5G transmissions in time domain are structured into frames of 10 ms. One frame is, in turn, divided into
10 subframes of 1 ms each. A subframe is again divided into several slots that contain 14 OFDM symbols
each. The number of slots allocated in a subframe depends on the numerology. An example of how slots are
organized in 5G is shown in Fig. 2.1, in which shaded symbols denote a symbol whose CP is longer. It is
possible to see in the figure that the structure is scaled dividing by a power of two, with the origin in 15 KHz
for coexistence with LTE. Another benefit of this division is that mixing different numerologies in the same
subcarrier is avoided.

One subframe (1 ms)

14 symbols / 1 slot
Δf = 15 kHz

28 symbols / 1 slot
Δf = 30 kHz

56 symbols / 1 slot
Δf = 60 kHz

112 symbols / 1 slot

Δf = 120 kHz

224 symbols / 1 slot

Δf = 240 kHz

Figure 2.1 Slots in 5G according to subcarrier spacings of 15, 30, 600, 120, and 240 kHz.

It is also important to comment that there is a technique implemented in 5G NR to support low latency
transmissions. It consists on separate the concept of transmission duration from slot duration. In other words,
it is possible to use only a portion of the slot for transmission. It is known as mini-slot transmission.
 2.6 Frequency domain 11

Table 2.2 Frequency range related to numerology in 5G.

Frequency range ∆ f (kHz) Number of RB BWch (MHz)

15 25-270 5, 10, 15, 20, 25,
FR1 30 11-273 30, 40, 50, 60, 70,
60 11-135 80, 90, 100
60 66-264
FR2 50, 100, 200, 400
120 32-264

2.6 Frequency domain

The basic element in terms of frequency is a Resource Element (RE). It consists of one subcarrier during one
OFDM symbol. Another division commonly used is referred to as a Resource Block (RB), which consists
of twelve consecutive subcarriers. It is important to remark here the difference of a RB in LTE, which was
twelve subcarriers during one OFDM symbol and one slot. Here an RB is a one-dimensional measure, and
it can be explained if we understand that 5G NR is more flexible. As was commented in the last section, now
the transmission does not have to occupy the entire slot, and that is the reason why the definition has changed.
3GPP introduced in Release 15 two ranges of frequency where a 5G transmission can be allocated, which
are called frequency ranges (FRs). Sub 6 GHz range is called FR1 and the millimeter-wave range is called
FR2. FR1 covers bands from 450 MHz to 7.125 GHz and FR2 from 24.25 GHz to 52.60 GHz. 1 The selection
of an FR involves a different utilization of the spectrum. Note that not every numerology is allowed in each
FR. Table 2.2 shows the number of RBs allowed for each situation as well as the channel bandwidth specified
for each case.

1
For the full list of the bands involved in the different frequency ranges, see https:// [Link]/ wirelesstechnology/ 4glte/ 5g-
frequency-bands-lte/
 3 Design of the proposed tool

Your love makes me strong. Your hate makes me unstoppable.

Cristiano Ronaldo, football player.

A fter introducing all 5G-NR concepts that are necessary for the complete understanding of this work,
next step is to introduce how the development tool looks like and also how it has been designed and
implemented. This chapter is organized into two parts. The first one, where the tool itself is presented and
we will discuss everything based on the flowchart and the block diagram of the tool. The second part of this
chapter consists of a description of the repositories used to upload this project.

3.1 Software designed

3.1.1 Flowchart description

To start with, the flowchart of the full project is shown and can be found in Fig. 3.1, and the full code of the
developed tool is included in the appendix (Code A.2 and Code A.1). It consists of a full simulation of 5G
generation and their following reception. As it has been already commented during the first chapter, the tool
has been developed in Python, which is an open-source programming language.

The software starts with some configuration parameters, that can be set by the user but they must be
according to the standard defined, that has been commented in the last chapter (table 2.1 and 2.2). More
precisely, the parameters that the user can set are:

• Number of Resource Blocks (NRB). Here the number of RB is set and must be defined according to
Table 2.2.

• Subcarrier spacing (Df). Subcarrier spacing is another parameter that must be set in order to prepare
everything for the transmission.

• Number of slots (Nslots). It indicates the number of slots that are going to be generated. The more
slots that are generated, the more accurate our calculates will be, but it also will take more resources
on our computer, and more time will be spent to calculate the results.

• Frequency range (FR). As has been commented, 5G-NR transmission can be allocated in two frequency
ranges. By setting FR1 or FR2 we fix the numerology we are using, that is completely different.

• Modulation order (M). It represents the size of the constellation used for the transmission. One of
the following schemes supported must be chosen [ETS, 2018]: binary phase-shift keying (BPSK),
quadrature phase-shift keying (QPSK), and quadrature-amplitude modulation (QAM) of 16QAM,
64QAM, and 256QAM.

13
14 Chapter 3. Design of the proposed tool

initialize parameters
simulation
model

extract
numerology

standard
okay?

yes
set
modulation
yes
order

multicarrier
generation
& channel

simulation
chosen

power
BER plot options amplifier
simulation enabled simulation

analyse
multicarrier

present &
save results

more
simulation?

no

no
stop

Figure 3.1 Flow chart of the simulation script.

• Peak to Average Power Ratio (PAPR). We were discussing in the last chapter that PAPR was one of
the drawbacks of using OFDM as the modulation technique. This parameter set what is the maximum
PAPR value admissible. If it is higher, it will be reduced by peak clipping techniques.

• Seed for random number generation (SEED). This parameter intends to set the seed for random
number generation, commonly used for developers.

Additionally, there are also some boolean variables that the user can change, and that allows us to run one
specific simulation or just avoid the representation of figures to be more computationally efficient. Regarding
those parameters, they will be commented in detail in the next chapter where all simulations that have been
carried out are presented.

As was said a few lines ago, before starting the simulation, the tool checks that every parameter is defined
according to the standard. If that is not true, the simulation will stop, showing in the screen that the parameters
set are not defined according to the standard, and it shows how the format should be. An example is shown in
Fig. 3.2, where a certain number of RB of 1 is set, but the program returns a message specifying the range of
the RB that can be set for the given parameters.
 3.1 Software designed 15

Figure 3.2 Screenshot that shows and example of what happen when parameters specified does not match the
standard of 5G NR.

Once the parameters have been defined properly, the next step is to specify the modulation order. It is done
like this because there is a simulation that uses all the modulations to calculate some results and compare
them for every modulation. Hence, if it is defined here and not before, it is possible to repeat the simulation
only varying that parameter, and they can be allocated under a for a loop. After that, the set-up would be
complete, and the next step is to generate the 5G-NR signal and after that, receive it. All this process is
described in detail in the next section, where a block diagram that represents this action is commented. Lastly,
all results must be saved but additionally, some of then are printed and can be seen in the console log. If
more simulations are necessary to do, the software will be run again, otherwise, it will finish and we will be
able to see all the figures that have been generated.

3.1.2 Block diagram

Random ovs Spectrum

S/P P/S CP Peak
data Modulator stage I IFFT ovs stage II shaping
conversion conversion insertion clipping
generator filter

Figure 3.3 Diagram of the multicarrier generation stage (ovs stands for oversampling).

The block diagram of the transmitter stage is shown in Fig. 3.3. In this section, waveform generation is
discussed. For waveform reception, the steps defined are the same as in the generation stage, but backward.
For simplicity, we will only comment on the transmitter stage. There are some stages with more than one
arrow, which means that the information in those stages is processed in parallel. Each step is explained next.

Random data generator

First of all, a random sequence of bits must be generated. Therefore, it is necessary to calculate how many
bits are needed. It is important to remark that the bit sequence will be determined by the seed value already
commented in the last section. Taking into account the number of slots and resource blocks that have been
specified, it is possible to calculate the number of bits that must be generated for the transmission,

Nbits = (14 Nslots ) · (12 NRB ) · log2 (k), (3.1)

where Nslots , NRB and k are the number of slots, RBs and the order modulation, respectively. This equation
comes from the fact that we need to insert k bits into an specific subcarrier every symbol. 12 NRB is the
number of subcarrier per OFDM symbol and 14 Nslots the number of OFDM symbols specified.

To give some understanding of the equation defined above, it is important to remark that the number of
bits generated is influenced mainly by two variables: the number of slots and the number of resource blocks.
We will see in the next chapter that if we want to run a simulation we must generate a significant number of
bits, so it is important to take into account that there are two ways of doing that: time domain (slots) and
16 Chapter 3. Design of the proposed tool

frequency domain (resource blocks). The first one is unlimited since we can always generate more symbols
and therefore exploit the time domain. However, the number of RBs is limited at some point by the standard.

Modulator

Once that bits have been generated, the next step is to modulate them. For this purpose, it is necessary to
obtain the level of amplitude for each complex symbol in both in-phase (I) and quadrature (Q) components.
First, we get a constant that guarantees the desired energy per bit Eb in Joules,
s
3Eb log2 (M1 M2 )
A= , (3.2)
M12 + M22 − 2

where M1 and M2 are the order modulation

√ for I and Q components, respectively. In 5G NR, all modulation
schemes are squared, so M1 = M2 = M = 2k . The alphabet that represents every level of amplitude, AI and
AQ is obtained as follows [Proakis and Salehi, 2008],

AI = A(2i − M1 + 1)
(3.3)
AQ = A(2 j − M2 + 1),

where i and j = 0,1,...,M − 1. Then, a bit sequence must be associated with each symbol. The best way to
do that is by using a Gray codification, that ensures the minimum BER for a given symbol error rate (SER).
After that, the bit sequence has turned into a digital signal that contains the bits mapped into symbols and
can be represented into a constellation To show an example, Fig. 3.4 represents the constellation of every
symbol for a given modulation order. The energy per bit value has been set to 1 during all experiments.
Once the symbol has been calculated, the next step is to associate them with the subcarriers. Thanks to
the simplicity of OFDM modulation, this association is a direct operation because the value of each symbol
corresponds to the value of each subcarrier. It is important to remark that the symbol generated is represented
in the complex plane, as can be seen in Fig. 3.4. As a consequence, the subcarrier is also complex and if we
want to plot the subcarriers, we will have to represent the absolute value of them instead of the real or the
imaginary part. This is a concept that engineers that work with signals know very well.

Oversampling

To meet the requirements for other external applications, sometimes it is necessary to apply an oversampling
(ovs) to increase the sampling rate. The natural way of doing that is at the final stage, where an interpolation
may be performed to add/remove samples and change the sampling rate. However, this method distorts the
spectrum of the signal and therefore it is necessary to explore other techniques to change the signal sampling
frequency. One technique is to perform part of the oversampling in this stage. After modulating the subcarriers
individually, it is possible to add zeros to the left and the right of the subcarriers. As a consequence, the
number of samples will increase, but since we are adding zeros on both sides, the information stored will not
change.
It is important to remark that only an equal number of zeros must be added on both sides. That means that
only an oversampling with an integer value can be performed here. It follows that if an oversampling with
decimals is required, an additional fine oversampling to the final rate will be performed before the clipping
stage. But, at least, most of it will be performed here and, hence, the distortion is minimized.
The second part of the oversampling is performed at the end of the transmitter stage, after the CP insertion.
However, it is going to be explained here because we have already put into context this concept. As it was
said at the beginning of this section, if it is possible, the oversampling will be performed in the frequency
domain, adding zeros to the left and the right of the subcarriers if it is an upsampling, or removing zeroes if
it is downsampling. But sometimes the overall oversampling that must be performed will not be an integer
number and therefore we will need to explore other techniques to change the frequency sampling accordingly.
What has been done here is a technique that is called FFT interpolation [Selva, 2015].
Lastly, it must be clear that the purpose of the oversampling is to change the frequency sampling and an
oversampling technique that works in the frequency domain will be always preferable: the distortion of doing
a resampling is minimized if we face the problem in this domain.
 3.1 Software designed 17

&RQVWHOODWLRQXVHG &RQVWHOODWLRQXVHG










 
4

4
 

í í

í
í

í
í

í
í

í í í í      í í í í     

, ,

&RQVWHOODWLRQXVHG &RQVWHOODWLRQXVHG









4

4

 

í
í

í

í

í

í í    í í í    

, ,

Figure 3.4 Constellation for all the modulation order generated in 5G NR: QPSK, 16QAM, 64QAM and
256QAM.

IFFT

The next step is to perform the IFFT. It has also been considered that more than one OFDM symbol may be
processed at the same time, so it will be necessary to compute one IFFT per OFDM symbol. The IFFT was
defined previously on Eq. (2.2). It is important to remark one more time that N is the FFT size, which is the
next power of two of the number of active subcarriers. As it is commonly known, doing the calculation with
a power-of-two number of points is more efficient, and this technique has been adopted in every design. It is
important to mention that for doing these calculations the library numpy has been used, which provides a
really good performance in digital signal processing commands. What is more, the reader should remember
that the number of symbols is influenced by two terms: the number of active subcarriers, e.g, the total
resources in terms of symbols, but also by the number of slots, where 14 OFDM symbols represent 1 slot.
Therefore, now you can think that a good way to program this action is a for loop, where an IFFT is performed
for every OFDM symbol. But that is not the best way of doing that. Numpy allows to execute more than one
IFFT operation in parallel, we have to store the data in a matrix, where the number of columns specifies the
number of IFFT operations. This is helpful and improves the quality of the code since we avoid using for
loops that makes always this kind of software slower.

The output of this stage is the signal in the time domain, with the frequency sample required except for the
last part of the oversampling, if it applies. But the signal is not ready to be transmitted over the air yet. There
are still some steps that are explained next.
18 Chapter 3. Design of the proposed tool

Cyclic prefix

Once the signal has been transformed into the time domain, the cyclic prefix must be added. As it was already
specified in (2.5), the last NCP samples of every symbol must be extracted and added to the beginning of the
symbol. This has to be done for every symbol, taking into account that the number of samples extracted
depends on the symbol position as well as the specified numerology. Now we do have to operate recursively
because not every symbol has the same CP length. It was already commented in the last chapter, but in 5G
NR every 7 symbols, there is one with a shorter duration. Insert the CP requires only one action: take the last
samples of the signal in the time domain and put them at the beginning.
It is important to mention here that as we have commented, the oversampling changes the number of
subcarriers and therefore the number of points of the IFFT. It means that now the number of samples that the
cyclic prefix must add is different than the number if no oversampling is applied. The number of samples of
the cyclic prefix was originally defined in (2.5). Now we must add the same factor that changes the number
of subcarriers, e.g, the value of oversampling in frequency domain. If we do not do this, the percentage of
samples that contains the cyclic prefix will not meet the standard requirements.

Peak clipping

Flexible numerology that 5G introduces exhibits a high peak-to-average power ratio (PAPR), which in-
creases the demand on the performance of radio frequency (RF) power amplifiers [Gomes et al., 2018].
For this reason, a peak-clipping stage is introduced to reduce the PAPR of the signal before it is filtered
[Rateb and Labana, 2019, Levanen et al., 2017]. The peak clipping function is defined as
(
x(n), |x(n)| ≤ A
G(x(n)) = (3.4)
A · e j·∠(x(n)) , |x(n)| < A,

where x(n) is the input signal, and A a scale factor based on a PAPR target in decibels, PAPRtarget . The scale
factor is obtained as
A = 10(PAPRtarget −PAPRx )/20 , (3.5)
where PAPRx is the PAPR of the signal input x in decibels, that is defined as follow,

max(|x(n)|)
 
PAPRx = 20 log10 , (3.6)
rms(x)

being rms(x) the root mean square value of x(n).

The purpose of peak clipping is to reduce the amplitude of some samples in the time-domain signal. The
function does not disrupt the sample if the absolute value of the sample is lower than the parameter A. If it is
greater, it does not disrupt the phase, changing the module by the value of the parameter A. Doing that, the
phase remains the same and therefore the signal distortion is minimized.

Spectrum shaping filter

delay

k+1 ... M-2 M-1 M 0 1 ... k-1 k

y(n)

0 1 ... k-1 k k+1 ... M-2 M-1 M

Figure 3.5 Due to the filter delay, it is necessary to reorder the output signal.
 3.2 Code availability 19

The addition of the CP will distort the spectrum and it might now contain some spurious radiation beyond
the bandwidth channel. Therefore, it is necessary to filter the signal with a low-pass filter before modulating
the carrier and radiating the resulting RF signal over the air [Park et al., 2019]. This low-pass filter is aimed at
shaping the RF signal spectrum, that is why this stage is referred to as spectrum shaping [Cheng et al., 2016].
The low-pass filter has been designed using the Kaiser window h(n) [Oppenheim et al., 2001], that is
defined as

 I0 [β (1−[(n−α)α]2 )1/2 ]
, 0 ≤ n ≤ N,
h(n) = I0 (β ) (3.7)
0, otherwise,

where α = N/2, being N the length of the filter, i.e, number of taps, and I0 (·) represents the zeroth-order
modified Bessel function of the first kind. Lastly, β and N are design parameters that we have to obtain in
order to design our filter according to the specifications. As a design parameter, we want at least A = 100 dB
rejection in the stopband, being the cutoff frequency (Fc ) of the filter the bandwidth of the generated OFDM
signal divided by two (we are working with the equivalent low-pass signal), given by

BWsignal N ·∆f
Fc = = RB . (3.8)
2 2
The transition bandwidth is the difference between the channel bandwidth and the cutoff frequency, i.e.,
BWtr = BWch − Fc . The value of BWch for the proposed parameters is obtained from Table 2.2. Next step
is to get the parameters needed for the filtering, β and N, that are obtained with the following equations,

0.1102(A − 8.7),
 A > 50,
β = 0.5842(A − 21)0.4 + 0.07886(A − 21), 21 ≤ A ≤ 50, (3.9)

0, A ≤ 21,


and
10A/10 − 7.95
N= + 1. (3.10)
2.285(π · BWtr )

Once the desired filter has been attained, the next step is to calculate the filter output. We have used the
FFT to compute circular filtering. Hence,

Y ( f ) = X( f ) · H( f ). (3.11)

However, it is necessary to do one more step to get our filtered signal. The filter delay is k = N/2 samples,
which means that the first samples of the output are allocated from sample N/2 to the end. Moreover, since
we have applied circular filtering, the last samples will be allocated at the beginning of the obtained signal.
To solve that, it is necessary to reorder the samples, as shown in Fig. 3.5.
In practice, the choice of Fc must be done carefully because a low value of the cut-off frequency might
involve in a distortion of the received constellation. In other words, it is a trade-off between distortion in the
constellation and spectrum spurious due to the CP. It is possible illustrate this with an example. A low value
of Fc means that part of the signal will be outside the filter borders, as shown in the left pictures on Fig. 3.6.
In that case, the Fc was equal to the 97.5 % of the signal bandwidth. For the second case, the value of Fc was
set to the 102.5 % of the bandwidth signal. As it can be shown, that adjustment was doing manually, running
the simulation and observing which value was okay. Finally, the value that was set for the official version was
the second one, where the constellation received looks much better than in the first case, where it is shown
that the signal spectrum is completely distorted.

3.2 Code availability

One of the purposes of this project is to make the software publicly available. This work is not only intended
for private use, the tool must be available to download for everyone. If we are in an engineering context, it is
20 Chapter 3. Design of the proposed tool

)LOWHUHIIHFW )LOWHUHIIHFW

+I +I

 
;I ;I

<I <I

í í

36'>G%P+]@

36'>G%P+]@
í í

í í

í í

í í

           

IUHTXHQF\>0+]@ IUHTXHQF\>0+]@

&RQVWHOODWLRQUHFHLYHG &RQVWHOODWLRQUHFHLYHG

 

 
4

4
 

í í

í í

í í    í í   

, ,

Figure 3.6 Spectrum shaping filter: how important is the choice of cut-off frequency. 97.5 vs 102.5.

obvious that GitHub is the most used online repository. But as you know, this tool is developed in Python,
and we must not forget that there is an official repository for this programming language. Lastly, this work
is planned to be submitted as a paper in the IEEE Access 1 . There is also a tool for publishing code in a
research context called Codeocean. Without any doubt, this tool is also an important target that must be taken
into account.
Before start commenting on how the code is available in every platform, it is important to remark how the
code is structured into files. It consists of only 3 files:
• [Link] (See Code A.1). It includes all configuration variables commented during this chapter
and the simulations that can be performed that are discussed in the next chapter. This script can be
considered as an example of use, and it is not part of the tool itself.
• [Link] (See Code A.2). This file contains all the functions related to the multicarrier generation,
OFDM transmitter and receiver, and all elements of the block diagram that is represented on Fig. 3.3.
It is considered the main part of the project.
• [Link] (See Code A.3). Here is where all extra code that is necessary to run the main applications
is located. It contains also some functions related to modulation/demodulation, filtering, etc.
With all these ingredients, what was clear was that this software must be published. Finally, all of the
options were chosen and the code is publicly available in all of the platforms that we have commented. The
process to put the code available in every context and also how to set up the project are discussed next.

3.2.1 Codeocean

Codeocean is a platform where users can publish code for free. As they say on his website, their mission is to
make computational research easier, more collaborative, and durable. A big step was done when in 2017 they
partnered the IEEE. It enabled the author to enhance the visibility of their papers, going one step further and
1
https:// [Link]/
 3.2 Code availability 21

allowing everyone not only to discover the project reading the paper, but also trying a demo running code
and exploring how it is done in terms of code.
Since the first moment that this project was planned to be published, the intentions were to publish the
code also in this platform. Finally, we succeed in the publication and this project is publicly available since
June 2020 at Codeocean [Palomino et al., 2020]. The fact of upload a project to Codeocean means that a
Digital Object Identifier (DOI) number is assigned, meaning the project easier to be referenced. The DOI
number that was assigned to this project was: 10.24433/CO.2712496.v1.

Figure 3.7 Screenshot of my capsule at Codeocean.

Every project in Codeocean is associated with a capsule, that is a platform where the user can observe
the project from outside it: it is possible to run the code but can not modify anything. It is also possible to
open output files but the access is limited in some cases by the administrator. Our capsule in Codeocean is
available in the following link: https:// [Link]/ capsule/ 7434591/ tree/ v1. If you click on it, you will
see something similar like the screenshot represented in Fig. 3.7. On the left side of the screen, you will see
the working directory as well as some files under the folder called code. That folder is where the code is
placed and you can open the files and take a look at the Python scripts. However, if you try to edit the files
you will see that it is not possible, since it is blocked by the user. That is the concept of a capsule that was
commented above. On the right-top part of the web page, there is a blue button where the user will run the
app. This will generate some messages on the log console but also will create the output files that will appear
under the output folder. In this particular case, the user will find all waveforms generated and the results of
every simulation.
As was said, this project was approved on the 14th of June by the Codeocean reviewers and it is available
for everyone. As of the day, I am writing these lines, July 2020, this is the only project uploaded to Codeocean
related to 5G for educational purposes.

3.2.2 PyPI

The tool that is presented in this work would be ideally preferable to be used by other scripts or functions. In
the end, it is intended for the understanding of 5G-NR waveforms but it can be also helpful if you need to test
your application and need to generate 5G-NR signals. Therefore, it is important to look for something that
allows you to import the code into your application without any problems. The result of that is the necessity
of upload this project to a Python repository: the Python Package Index (PyPI).
It is estimated that more than 230 thousand packages are available to download in PyPI. It is considered
the official repository of this programming language. For that reason, it was considered a good idea to upload
this project to PyPI. The difference here is that now the reason why people would download this package is a
22 Chapter 3. Design of the proposed tool

different one than the reason they had with Codeocean. Now we are not talking about researching, but about
professional purposes. Therefore, file [Link] is not included here.
The project was successfully uploaded to PyPI in June 2020 and can be found on the following link:
https:// [Link]/ project/ NRlabdtsc/ . It is also important to mention that it can be installed easily in our
Python environment just by executing the following command:
>> pip3 install NRlabdtsc
Once the installation is complete, you can import the package with the following lines at the beginning of
your script:
>> import [Link] as NR
>> import [Link] as utils
After that, you can code whatever you want using this package. It is important to mention that once we
have downloaded the package, we can import it on the simulation script, [Link], instead of having every
file in the current folder. Now the package is imported in the whole system, and it is possible to check that
the package is installed by executing the following command:
>> pip freeze
This command will return every package installed in the system sorted alphabetically. The package NRlabdtsc
will be on the list, with the last version available when the user downloaded it.

3.2.3 Github

Lastly, this project is also uploaded to Github, the most famous repository for collaborative documents.
Github also supports Git, a free distributed version control system, so it is possible to track different versions
of the code, which is helpful to have everything well organized. The installation of the repository can be
done by executing the following command:
>> git clone [Link]
There is nothing much that can be said here since the author has worked before with this tool and has
experience on it, therefore no extra time has been spent on learning this tool. The repository is available at
the following webpage: https:// [Link]/ guipalloz/ NRlab-dtsc.
 4 Results and discussions

Scientists investigate that which already is; engineers create that

which has never been.

Albert Einstein

T he last chapter of this thesis covers all the simulation carried out, and their following results and
discussions. As it has been mentioned during this document, the purpose of all this work is to provide
a complete tool that allows the users to create their experiments. However, as an example, as well as part of
the validation process, some simulations have been developed that will be explained during this chapter. As
a consequence, with those simulations, some interesting results can be extracted. Most of the results have
indeed been already commented on during the whole document, but here we are describing, step by step,
which simulations are important to validate this work and how the user can get some inspiration on these
simulations and perform new ones.

Since this is the last chapter of this thesis, there is a section that comments a few lines about how was the
methodology that was followed and, most important, what are the conclusions of the works as well as some
future lines. This section can be found at the end of this chapter.

4.1 Simulations and results

To measure the system performance of the proposed tool, this section is organized as follows. Firstly, the
result of the multicarrier generation stage, explained in detail during chapter number 3, is analyzed. More
precisely, the 5G signal is generated in the time domain. Then, a simulation through an AWGN channel is
performed, and its result is presented and discussed. Here the main figures of merit of a classical digital
communication scheme will play an important role. We will calculate some parameters and it will be shown
how good our software works. Finally, the generated signal is sent to an external server that creates the
signal and tests it with a power amplifier located in the University of Chalmers, Gothenburg (Sweden). All
simulations explained during this chapter have been implemented and the code is included at the end of this
work (Code A.1).

4.1.1 Simulation number 1: Multicarrier generation stage

The first simulation that was performed was the direct output of the block diagram that was described in the
last section. As was commented, the output was a signal generated in the time domain. Hence, an analysis is
performed to study this output in detail. The problem of the PAPR in 5G is also addressed here, where a
study of how PAPR affects to our system and how good it is improved when peak-clipping techniques are
introduced (which also motivates and validates the assumption it was done by introducing this element) has
been performed and their results are included also next.

23
24 Chapter 4. Results and discussions

Analysing the signal in the time domain




|x(t)|





7;VLJQDO

6ORWV


      

7LPH>XV@

Figure 4.1 5G waveform for a subcarrier spacing equal to 60 kHz and 5 slots.

The output of the multicarrier generation stage is the generated 5G signal in the time domain, which was
shown in Fig. 4.1 for 20 RBs, a subcarrier spacing of 60 kHz and 5 slots. As was already mentioned, those
parameters are just an example, and the software allows us to change them. It is possible to appreciate how
the time duration of the signal corresponds with the parameters given in Fig. 2.1, where all numerology was
represented. To give an example, a subcarrier spacing equals 60 KHz refers to the third row on the figure. It
is possible to see on the picture that 56 symbols are equal to 1 subframe (1 ms), therefore the duration of
each symbol is 1/56 ms. 14 symbols are 1 slot, which means that in the simulation we have 14· = 70 OFDM
symbols. Lastly, the duration of every symbol is 1/56 ms · 70 = 1.25 ms that is exactly the duration of the
signal represented in Fig. 2.1.

100

80
Counts

60

40

20

0
11 11.5 12 12.5 13 13.5 14
PAPR [dB]

Figure 4.2 Histogram of the PAPR, obtained for 1000 signals generated.

Table 4.1 Numerology for PAPR simulation.

Number of RB ∆f PAPR target Number of simulations Seeds

250 15 KHz 10.5 dB 1000 0-999
 4.1 Simulations and results 25

14.0
Output
13.5 Input

13.0

12.5
PAPR (dB)

12.0

11.5

11.0

10.5

10.0
0 200 400 600 800 1000
Simulations

Figure 4.3 Input and output PAPR of the peak clipping stage along 1000 simulations.

Peak clipping stage motivation

Before finishing the transmitter stage, it was mentioned in the last chapter that a peak clipping was necessary
to reduce the PAPR. The next experiment is to calculate precisely that PAPR and motivate somehow the
fact of adding this element in the transmitter stage. The simulation generated a signal with the numerology
specified in table 4.1. It consisted of generating 1000 times a signal with the characteristics mentioned and
averaging the results of the PAPR obtained before and after the peak-clipping stage.
In Fig. 4.2 the histogram of the PAPR before peak clipping is shown. A histogram is defined as a
representation of the distribution of some data. In this case, the horizontal axis represents the values of the
PAPR found on the signal samples, and the vertical axis indicates the number of occurrences of those values
in every sample. Analyzing the figure, most of the obtained PAPR for each simulation is higher than 11 dB,
so it is evident that peak clipping is necessary [Ochiai and Imai, 2001]. Using the peak-clipping technique
defined in Section 3.1.2 with a PAPR target of 10.5 dB, the obtained normalized mean squared error (NMSE)
between the PAPR after peak clipping and the target value of 10.5 dB is equal to 0.023%. Besides, an average
of 15.1 samples for every simulation was affected by clipping. One last aspect that can be analyzed is the
one that is represented in Fig. 4.3, that is the PAPR calculated before and after peak clipping along 1000
simulations. It is shown how the improvement is performed, with an almost perfect output PAPR equals to
the target one, that was 10.5 dB. The script that performs all these techniques mentioned here can be found in
the Code A.4.
In this tool, a basic peak-clipping function has been implemented (see Section 3.1.2), but more advanced
techniques can be applied here, which reduces, even more, the PAPR and minimizing residual effects. In
parallel to this project, another thesis has done a deep study about peak clipping techniques [López, 2020],
therefore it could be interesting to have performed an analysis of how different techniques affects the waveform
generation. Without any doubt, that analysis will be useful to upgrade this software in the future.

4.1.2 Simulation number 2: AWGN channel and most typical figures of merit

The challenge of this second simulation that is presented in this chapter is to obtain the classic figures of
merit that are calculated when a digital wireless system is tested. The starting point will be the output of the
transmitter stage, as we did with the last simulation. Two measurements have been specified: Bit Error Rate
(BER) and Error Vector Magnitude (EVM). Both experiments were performed following the parameters
specified by the 5G-NR standard.
26 Chapter 4. Results and discussions

Bit Error Rate (BER)

To obtain the performance of the designed tool in terms of BER, the waveform generated was passed through
an additive white Gaussian noise (AWGN) channel, and then it was detected. It is important to perform a
BER simulation analysis in every digital system because those parameters show us how is the quality of the
full radio link in terms of bits. BER value is extracted for specific transmission conditions, i.e., for a given
value of Eb /N0 and a modulation order. It is possible to perform a Montecarlo simulation that iterates over a
range of values of those. More precisely, in this simulation, the values selected for Eb /N0 were from 0 dB to
15 dB. Additionally, several modulation techniques were implemented to evaluate the performance of the
proposed model. Precisely, modulations supported by 5G such as QPSK, 16QAM, 64QAM, and 256QAM
were chosen. The random bits were modulated and inserted in each subcarrier following the order modulation
specified. The result of the Montecarlo simulation is shown in Fig. 4.4. To calculate the BER after the
transmission and its posterior reception, it is just dividing the bits that were different after the detection by the
total number of bits transmitted. If we do that for every value of Eb /N0 a curve will be obtained. If we repeat
the same procedure but now for every other order modulation, a similar figure like the one represented here
will be obtained. But to know if the results are correct or need to be improved, it was also considered to plot
the theoretical curves. However, there does not exist a close expression of those curves. As a consequence,
the dashed lines represent the maximum level that the BER can be. In mathematical terms, those curves can
be approximated as

10−1

10−2

10−3
BER

10−4 QPSK (Th)

16QAM (Th)
64QAM (Th)
10−5 256QAM (Th)
5G - QPSK
5G - 16QAM
10−6 5G - 64 QAM
5G - 256 QAM
10−7
0 2 4 6 8 10 12 14
Eb
N0 [dB]

Figure 4.4 Average Bit Error Rate (BER) versus a sweep in the bit energy to noise power ratio.

r !
4 3Eb /No log2 M
BER ≈ Q . (4.1)
log2 M M−1

From Fig. 4.4 can be observed that the simulated curves are below the theoretical ones since these latter
were an approximation of the first ones, as was said in last above. Also, for a fixed value of Eb /N0 , the
simulation with a lower order modulation results in a lower BER. However, the number of bits that are
inserted into the subcarriers is also lower, so the bit rate is reduced. This means that the choice of the order
modulation relies on the channel conditions: a higher data rate will be preferable, but the BER value must be
suitable for the specific desired application.
 4.1 Simulations and results 27

Error Vector Magnitude (EVM)

Another typical figure of merit of a digital system is the error vector magnitude (EVM), which measures how
far are the constellation points from the ideal one. The EVM is calculated as follows,

q
n=0 Ierr (n) + Qerr (n)
∑N−1
1 2 2
N
% EVM = × 100%, (4.2)
EVM Reference
where n is the symbol index, N the number of symbols for the given transmission, and I and Q are the in-phase
and quadrature components of the error vector, where Ierr is defined as the difference between de original
points in the constellation and the received ones. Lastly, EVM Reference is a normalization factor that is
calculated as q
EVM Reference = max I2err (n) + Q2err (n) , (4.3)

i.e., the maximum value of the norm of every point of the constellation. The Montecarlo simulation that
was carried out was exactly the same it was done in last simulation but now calculating the EVM instead.
Therefore, we can extract a similar figure and it is expected that the system performance would improve with
higher values signal to noise ratio. The results of the simulations are shown in Fig. 4.5 and it is possible to
appreciate that the performance tendency is quite the same we had for the BER simulation.

QPSK
16QAM
60 64QAM
256QAM
50
EVM (%)

40

30

20

10

0
0 2 4 6 8 10 12 14
Eb
N0 [dB]

Figure 4.5 Error Vector Magnitude (EVM) versus a sweep in the bit energy to noise power ratio.

4.1.3 Simulation number 3: Sending the generated signal to an external RF WebLab

The last simulation consists in test our system in an external environment that is commonly used by the
research group that manages this project. This simulation is organized into two sections: firstly, the external
laboratory is introduced, and the simulation parameters are explained. Lastly, the results of the simulations
are presented.

RF Weblab at Chalmers University

One last simulation was performed, to give compatibility with the work performed by the Digital Pre-
Distorsion (DPD) research group that will use the tool. This simulation consists of sending the signal to an
RF WebLab located at Chalmers University of Technology [Landin et al., 2015]. It is possible to remotely
access the measurement system that was initially designed for hosting the IMS Digital PreDistortion Student
Design Competition in 2014 and 2015. According to the RF WebLab webpage, the intentions are "to give
DPD developers worldwide access to a common amplifier and measurement system, making the evaluation
28 Chapter 4. Results and discussions

of the performance of DPD algorithms easy, and enabling comparisons between various algorithms on a
common platform". The measurement setup consists of the following:
• A Vector Signal Transceiver (PXIe-5646R VST) with an instantaneous bandwidth equals to 200 MHz.
• A linear driver amplifier with a gain approximately equals to 40 dB.
• The DUT which is a nonlinear GaN power amplifier (Cree CGH40006-TB).
• The DUT has also a power source module (PXI-4130) that allows the users to know the power amplifier
current consumption.
However, there are also some limitations that the signal must meet before taking the measurements. To give
some examples, the signal that is interpreted as the input must have a PAPR lower than 20 dB, a maximum
number of samples of 1 million, and a maximum peak power of −8 dBm. It is also important to remark that
the peak output power level is limited to 6 W. Also, the sampling rate must be 200 MHz. If you have read
the last chapter, where all elements of the transmitter stage were explained, you will understand now how
important is to add the block that performs an oversampling that changes the sampling rate. Thanks to that,
we can easily generate a custom signal for every application.

40

30

20
Output power [dBm]

10

−10

−20
−50 −45 −40 −35 −30 −25 −20 −15 −10
Input power [dBm]

Figure 4.6 AM-AM characterization of the power amplifier.

Programming the script and simulation

RF Weblab provides some scripts in Matlab 1 that allows users to access the power amplifier. However, the
purpose of this simulation is to get the measurements without using any external actions. Hence, it was
convenient to explore other techniques that integrate those files in our simulations. Fortunately, Matlab has
an engine that allows us to run Matlab scripts with Python. To install the engine, all we need to do is to follow
these steps:
1. Open Matlab and execute the following command:
1
scripts available at: http:// [Link]/ rfweblab/ access-details/
 4.1 Simulations and results 29

65

60

55

50
Gain [dB]

45

40

35

30

25
−50 −45 −40 −35 −30 −25 −20 −15 −10
Input power [dBm]

Figure 4.7 Gain-AM characterization of the power amplifier.

>> matlabroot
The output will be the path where Matlab is installed. Navigate to that folder.
2. Now open a Windows prompt execute the following command:
>> cd matlabroot/extern/engines/python
where matlabroot is the output of the first step.
3. Lastly, you will find a Python script. Execute it running the following command:
>> python3 [Link] install
It will install the MatLab engine for Python.
4. After that, the Matlab engine for Python will be ready, and we can load it in Python with the following
line: eng = [Link].start_matlab(). It will return an object, and we can call every
Matlab command or function as a method of this object. It is also possible to use arguments, just add
them as parameters when the method is called.
The experiment setup is included in the simulation script already commented, located at the end of this work
(Code A.1). It is possible to appreciate how a flag called RunMatlabFunction enables this third simulation. If
the value is true, then some data are saved into a mat file that the MatLab script will use. The Matlab script
we are using is similar to the ones available at RF Weblab, but we are loading the data in Matlab instead of
creating them. More precisely, we are generating 10 different signals sweeping RMS values, from −30 dBm
to −21 dBm. The obtained results are then analyzed and the gain in decibels for each sample is obtained.
For every RMS value, several simulations are performed, and then the result is averaged. The results are
therefore with less noise but it will take more time to get the results. As always, it is a trade-off.
Once the Matlab script has finished and the measurement is obtained, it is possible to load again the output
data in Python. Furthermore, we obtain the input and the output of the RF Weblab (both real and imaginary
parts). And interesting and direct results is to calculate the AM-AM and Gain-AM characterization, very
useful when studying non-linear behaviors. The AM-AM characterization represents the amplitude variation
30 Chapter 4. Results and discussions

of the output respect to the input, whereas the Gain-AM represents the power amplifier gain (the output minus
the input, if measured in decibels) respect to the same input. Both characterizations have been calculated and
the resulting curves are represented in Figs. 4.6 and 4.7. The script that generates the figures is attached and
can be found in Code A.5.

4.2 Conclusions and future lines

The complete understanding of OFDM system, the processes involved in 5G waveform generation, and the
performance methods that are state of the art are concepts that need to be taught in Electrical Engineer
programs. To help Electrical Engineering students from all over the world to help to understand those
concepts, this work has been completely developed for waveform generation categorized as a FOSS platform,
which means that it is an opensource platform. This software allows the student to observe and study how
every block of the transmitter and reception stage works in 5G in terms of digital communication, looking at
them not only as a black-box but also being able to analyze how they are implemented. Additionally, the
platform gives the students the opportunity of modifying design parameters and see how they affect the
transmission in terms of PAPR, BER, and EVM, which are the most typical figures of merit. The students
observe a practical example that links with the theoretical contents already studied in the program, hence it
helps them to acquire the desired knowledge.
Thinking about myself, this work has been useful to reinforce my knowledge in mobile communications. I
had previous knowledge in 5G NR, but developing this tool in a real environment and generating waveforms
according to the standard gave me practical experience in this field. What is more, it is always good to develop
an open-source platform, because everyone can read what you have done and if they want, they can modify
and improve your work. That is how engineering works and how humanity has reached impossible goals:
helping each other. Additionally, this master thesis somehow concentrates all work last year. This document
reflects how was the process of this project, starting from the theory, and finishing coding and testing.
But this project does not end here. This thesis is just the first step that summarized the first version of this
software. After the project dissertation, this tool will be sent to the department responsible for this project
and they will be able to continue improving this tool. Furthermore, this work belongs to a tendency observed
during the last years: the migration of all research projects to Python. As a consequence, this is not the only
thesis that presents a work programmed in Python, and in the future, there will be an integration of all of
them.
As it has been commented previously, the intentions with this project were to submit a paper for publication
on IEEE Access magazine, so the works do not end here. If the request is accepted, a reviewing process will
start, therefore it is probable that the project will be modified in the future.
To sum up, the author feels satisfied with this project. He has developed a complete tool from nothing
to what is published on Codeocean, and he has given to everyone the possibility to study and explore the
different waveforms that can be generated according to the standard that will define the way that cellphones
will share data in the next generation of mobile communication.
 Appendix A

Implementation of the project

A.1 Simulation script that uses the toolbox presented in this work

Code A.1 Simulation script that uses the toolbox presented in this work ([Link]).
his#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Author: Guillermo Palomino Lozano
Organization: University of Seville
Date: May 2020
"""
import numpy as np
import os
import [Link] as plt
import [Link] as sio
#import [Link]
from [Link] import ScalarFormatter
import warnings

import [Link] as NR
import [Link] as utils

#------------------------------------------------------------
[Link]('all')

NRB = 150 # Number of resource blocks

Df = 30e3 # Subcarrier Spacing
OvS = True # Oversampling
Fs = 200e6 # Only important when oversampling is true
Nslots = 1
FR = 1 # Frecuency range
M = 64 #Modulation order
PAPR = 10.5 #PAPR desired
SEED = 1000 # Seed for random number generation
PlotOptions = True #True for active plots (Subcarriers Tx/Rx, Constellation,
Spectrum, etc )
#-----------------------------------------------Pl-------------
# Simulations available:

31
32 Chapter A. Implementation of the project

BERsimulation = False
Amplifiersimulation = False
RunMatlabFunction = False
Filtering = True

#------------------------------------------------------------
[Link]('ignore')
if (BERsimulation == True):
#Modulation = [Link]([4])
Modulation = [Link]([4, 16, 64, 256]) # Modulation order
EbNo = [Link](0,15,1)
BER = [Link]((len(EbNo),len(Modulation)))
errorteorico = [Link]((len(EbNo),len(Modulation)))
EVM = [Link]((len(EbNo),len(Modulation)))
NMSE = [Link]((len(EbNo),len(Modulation)))
else:
Modulation = [Link]([M]) # Modulation order
EbNo = [Link]([float('inf')])
BER = 0
for jj in range(len(Modulation)):
M = Modulation[jj]
k = np.log2(M)
MyStructure = [Link](Df,FR)
print("\n------------- STARTING TX -------------")
print("\n M = \%d" \%(M))
x, Bn, FFTSize, Alphabet, alpha, Eb, const_points, Fs, Fo, BwCh, PAPRx,
PAPRy = [Link](MyStructure, NRB, OvS, Fs, Nslots, M,
PAPR, SEED, PlotOptions, Filtering)
#Ready for real measurements!
if (RunMatlabFunction == True):
#RunBERsimulation.m implementation goes here
Nmeas = 3
Pin = [Link](-30,-20)
#Pin = [Link]([-30,-20])
X_amp = [Link](([Link],[Link]),dtype=np.complex128)
Y_amp = [Link](([Link],[Link]),dtype=np.complex128)
currpath = [Link]()
[Link]("Chalmers")
print("\n------------- STARTING Power Amplifier stage -------------")
print("Starting Matlab engine...")
eng = [Link].start_matlab()
for ii in range(len(Pin)):
RMSin = Pin[ii]
print("\n Saving signal for real measurements...")
[Link]('NR_signal.mat', {'x':x,'RMSin':RMSin,'Nmeas':Nmeas})
print(" Running real measurement in the RF WebLab for Pin = \%d dBm
... it will take around \%d minute(s) ..." \% (int(RMSin), int(
Nmeas)))
#[Link]("[Link]")
[Link](nargout=0)
print(" done!")
mat = [Link]('NR_signal_Output.mat')
aux = mat["x"][:,0]
X_amp[ii,:].real = mat["x"][:,0].real
X_amp[ii,:].imag = mat["x"][:,0].imag
Y_amp[ii,:].real = mat["y"][:,0].real
Y_amp[ii,:].imag = mat["y"][:,0].imag
 A.1 Simulation script that uses the toolbox presented in this work 33

#data = mat["data"]
#Save the results in a file
[Link]()
[Link]("../")
[Link]('[Link]',X_amp=X_amp,Y_amp=Y_amp,Pin=Pin)
y_amp = Y_amp[1,:]
print("\n------------- FINISHING Power Amplifier stage -------------")
if (BERsimulation == False):
print("\nGenerated signal:")
print(x)
print("\nBits transmitted:")
print(Bn)
print("\n------------- STARTING RX -------------")
for ii in range(len(EbNo)):
if (Amplifiersimulation == True):
data = [Link]('[Link]')
Xamp = data['X_amp']
Yamp = data['Y_amp']
Pin = data['Pin']
Nsim, Nsamples = [Link]
for jj in range(Nsim):
print("\n Analyzing measurement for Pin = \%.2f dBm" \% Pin[jj])
y = [Link](Yamp[jj,:], EbNo[ii], Chtype="AWGN")
rx_Bn, rx_const_points, ACPR, ACPR2 = [Link](
NRB, MyStructure, y, int(Nslots), Alphabet, alpha, int(
FFTSize), Fs, Fo, PlotOptions, BERsimulation,BwCh, Filtering)

[Link](const_points.flatten(), rx_const_points.
flatten(), NUMPOINTS = 500)
else:
SNRdB=EbNo[ii]+10*np.log10(k)+10*np.log10(NRB*12/FFTSize)
y = [Link](x, SNRdB, Chtype="AWGN")
if (BERsimulation == False):
print("Received signal:")
print(x)
rx_Bn, rx_const_points, ACPR, ACPR2 = [Link](NRB,
MyStructure, y, int(Nslots), Alphabet, alpha, int(FFTSize), Fs,
Fo, PlotOptions, BERsimulation,BwCh, Filtering)
if (PlotOptions == True):
[Link](const_points.flatten(), rx_const_points.
flatten(),NUMPOINTS = 500)
if (BERsimulation == False):
print("Received sequence:")
print(rx_Bn)

if (BERsimulation == True):
print(" EbNo = \%d dB" \% (EbNo[ii]))
BER[ii,jj] = [Link]((Bn != rx_Bn))/[Link]
#print("\nBER para SNR = \%f dB: \%f" \% (EbNo[ii], BER[ii,jj]))
ebno = 10**(EbNo[ii]/10)
errorteorico[ii,jj] = 4/k*[Link]([Link](3*ebno*k/(M-1)))
#arg = [Link](3*k/(M-1)/No)
#errorteorico[ii,jj] = (4/k)*(1-1/[Link](M))*Qfunct(arg)
EVM[ii,jj] = utils.EVM_Calculation(const_points.flatten(), rx_
const_points.flatten())
NMSE[ii,jj] = utils.NMSE_Calculation(x,y)
print("\n------------- FINISHING RX -------------")
34 Chapter A. Implementation of the project

if (BERsimulation == True):
[Link]()
[Link](EbNo, errorteorico[:,0], '-',label='M = 4 (Th)', color='blue')
[Link](EbNo, errorteorico[:,1], '-',label='M = 16 (Th)', color='red')
[Link](EbNo, errorteorico[:,2], '-',label='M = 64 (Th)', color='green
')
[Link](EbNo, errorteorico[:,3], '-',label='M = 256 (Th)', color='
black')
[Link](EbNo, BER[:,0],'-o', label='M = 4', color='blue')
[Link](EbNo, BER[:,1],'-o', label='M = 16', color='red')
[Link](EbNo, BER[:,2],'-o', label='M = 64', color='green')
[Link](EbNo, BER[:,3],'-o', label='M = 256', color='black')
[Link]('EbNo [dB]'); [Link]('BER');
[Link](True,which="both",ls="-")
[Link]()
axes = [Link]()
axes.set_ylim([10**(-7),0.5])
[Link]('5G NR System performance for AWGN channel')
[Link]('../results/[Link]')

[Link]()
fig, ax = [Link]()
for [Link] in [[Link], [Link]]:
[Link].set_major_formatter(ScalarFormatter())
[Link](EbNo, EVM[:,0],'-o',label='M = 4', color='blue')
[Link](EbNo, EVM[:,1],'-o',label='M = 16', color='red')
[Link](EbNo, EVM[:,2],'-o',label='M = 64', color='green')
[Link](EbNo, EVM[:,3],'-o',label='M = 256', color='black')
[Link]('EbNo [dB]'); [Link]('EVM (\%)');
[Link](True,which="both",ls="-")
[Link]()
[Link]('5G NR System performance for AWGN channel')
[Link]('../results/[Link]')

[Link]()
fig, ax = [Link]()
for [Link] in [[Link], [Link]]:
[Link].set_major_formatter(ScalarFormatter())
[Link](EbNo, NMSE[:,0],'-o',label='M = 4', color='blue')
[Link](EbNo, NMSE[:,1],'-o',label='M = 16', color='red')
[Link](EbNo, NMSE[:,2],'-o',label='M = 64', color='green')
[Link](EbNo, NMSE[:,3],'-o',label='M = 256', color='black')
[Link]('EbNo [dB]'); [Link]('NMSE');
[Link](True,which="both",ls="-")
[Link]()
[Link]('5G NR System performance for AWGN channel')
[Link]('../results/[Link]')
else:
EVM = utils.EVM_Calculation(const_points.flatten(), rx_const_points.flatten
())
print("EVM: \%.2f" \% EVM)
 A.2 Main 5G-NR toolbox script for waveform generation 35

A.2 Main 5G-NR toolbox script for waveform generation

Code A.2 Main 5GNR toolbox script for waveform generation ([Link]).
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Author: Guillermo Palomino Lozano
Organization: University of Seville
Date: May 2020
"""

import sys
import os
import numpy as np
import [Link] as plt
import [Link] as sio
import NRutils as utils

[Link]["[Link]"] = "Times New Roman"

def Numerology(Df, FR):

"""
Description
----------
Specify the numerology according to the standard of 5GNR

Parameters
----------
Df : float
subcarrier spacing.
FR : int
frequency range.

Returns
-------
MyStructure : dictionary
information about the numerology specified according to the standard.

"""
TimeDomainStructure = [{"DF": 15e3,"Tslots": 1e-3, "FR": 1, "Range": [25,
270], "mu": 0},
{"DF": 30e3,"Tslots": 0.5e-3, "FR": 1, "Range": [11,
273], "mu": 1},
{"DF": 60e3,"Tslots": 0.25e-3, "FR": 1, "Range": [11,
135], "mu": 2},
{"DF": 60e3,"Tslots": 0.25e-3, "FR": 2, "Range": [66,
264], "mu": 2},
{"DF": 120e3,"Tslots": 0.125e-3, "FR": 2, "Range": [32,
264], "mu": 3},
{"DF": 240e3,"Tslots": 0.0625e-3, "FR": 2, "Range":
[32, 264], "mu": 4}]
Dfs = [15e3, 30e3, 60e3, 60e3, 120e3, 240e3]
#Time-Domain structure definition:
if Df == 60e3: #According to the standard: 2 possibilities with Df = 60 Khz
36 Chapter A. Implementation of the project

if FR == 1:
MyStructure = TimeDomainStructure[2]
else:
MyStructure = TimeDomainStructure[3]
else:
MyStructure = TimeDomainStructure[[Link](Df)]
return MyStructure

def MulticarrierGeneration(MyStructure, NRB, OvS, Fs, Nslots, M, PAPRd, SEED,

PlotOptions, Filtering):

Ncarr_act = NRB * 12 #Number of active carriers: Each RB means 12 carriers

FFTSize = int(2**[Link](np.log2(Ncarr_act))) #Next power of 2
if OvS == False:
OvS = 1
OvS_freq = 1
Fs = float(FFTSize * MyStructure["DF"])
Fo = Fs
else:
Fo = float(FFTSize * MyStructure["DF"])
OvS = Fs / Fo
if (OvS < 1):
print("Error. OverSampling lower than 1")
[Link](0)
OvS_freq = [Link](OvS)
OvS_freq = 1
# Fs is the value given at the input
print(" Oversampling will be performanced: \%.2f" \% OvS)
print(" From Fo = \%.2f MHz to Fs = \%.2f MHz " \% (float(FFTSize *
MyStructure["DF"])/1e6, Fs/1e6))

x, Bn, Alphabet, alpha, Eb, const_points, BwCh = tx_OFDM(NRB, Nslots,

MyStructure, M, FFTSize, Ncarr_act, Fo, Fs, OvS_freq, SEED, PlotOptions
)
x = x - [Link](x);
x = x / [Link]([Link](x));

PAPRx = 20*np.log10([Link]([Link](x))/[Link]([Link](x))) # 10*log10(max(

abs(x))/rms(x))
print(" PAPR before clipping: \%f dB" \% (PAPRx))
clip = 10**((PAPRd - PAPRx) / 20)
idx = [Link]([Link](x) > clip)
print(" \%d Samples will be affected by clipping" \% ([Link](idx)))
x[idx] = clip*[Link](1j*[Link](x[idx]))
PAPRy = 20*np.log10([Link]([Link](x))/[Link]([Link](x)))
print(" PAPR after clipping: \%f dB" \% (PAPRy))
if (Filtering == True):
#do filtering stage here...
x = utils.Filter_signal(Fs, BwCh*1e6, Ncarr_act*MyStructure["DF"], x
)
if (PlotOptions == True):
[Link]()
n, bins, patches = [Link](x=[Link](x), bins=20, color='#607c8e',alpha
=0.7, rwidth=0.85)
[Link](axis='y', alpha=0.75)
[Link]('Absolute value of the x(n) sample')
 A.2 Main 5G-NR toolbox script for waveform generation 37

[Link]('Counts')
[Link]('Histogram of absolute value samples of x(n)')
[Link]('../results/[Link]')
t = [Link]([Link])* 1/Fs
[Link](figsize=(10,4))
[Link](t*10**6,[Link](x), label='TX signal')
[Link]('Time [us]'); [Link]('$|x(t)|$');
[Link](True);

for xc in range(0,Nslots+1):
[Link](x=(xc * MyStructure["Tslots"] * 1e6),linewidth=2, color
='r', label='Slots')
if xc == 0:
[Link](fontsize=10, loc='lower left')
[Link]('../results/[Link]')
f, Pxx = [Link](x, Fs, FFTSize)
Pxx = 10*np.log10(Pxx) + 30
[Link](figsize=(5, 4))
[Link](f,Pxx)
[Link]('frequency [MHz]')
[Link]('PSD [dBm/Hz]')
[Link]()
[Link](True);
[Link]('Power Spectral Density of the signal transmitted')
[Link]('../results/[Link]')
print(" Generated OFDM signal has \%d samples" \% ([Link]))
print(" FFTSize: \%d samples" \% (FFTSize))
print("\n------------- FINISHING TX -------------")
return x, Bn, FFTSize, Alphabet, alpha, Eb, const_points, Fs, Fo, BwCh,
PAPRx, PAPRy

def tx_OFDM(NRB, Nslots, MyStructure, M, FFTSize, Ncarr_act, Fo, Fs, OvS_freq,

SEED, PlotOptions):
#Range of Channel Bandwidths for the Different Numerologies
Df = MyStructure["DF"]

if MyStructure["FR"] == 1:
BwCh = [Link]([5, 10, 15, 20, 25, 30, 40, 50, 60, 70, 80, 90, 100])
else:
BwCh = [Link]([50, 100, 200, 400])
#Checking that our NRB is in range
if (NRB < MyStructure["Range"][0]) or (NRB > MyStructure["Range"][1]):
print("The number of resource blocks does not match the defined
numerology\n")
print("Desired range: \%d-\%d \n" \% ( MyStructure["Range"][0],
MyStructure["Range"][1]))
print("Your value: \%d \n" \% (NRB))
[Link](0)

Ncarr_act = NRB * 12 #Number of active carriers: Each RB means 12 carriers

print(" Number of active subcarriers: \%d" \% Ncarr_act)
BwEff = Df * Ncarr_act
BwEff = (BwEff + 1/2*(BwEff - NRB*12*Df-Df))/1e6 #Eq (18.1)
#And now we allocate this BW into one of the possibilities defined above
idxBw = [Link](BwCh > BwEff)[0]
Bw = BwCh[idxBw[0]] #Now we have set our final Bandwidth
print(" Channel Bandwidth: \%d MHz" \% int(Bw))
38 Chapter A. Implementation of the project

print(" Bandwidth occupied: \%d MHz" \% int(Ncarr_act*Df/1e6))

NsymbOFDM = Nslots * 14 #14 OFDM symbols every time slot (see Figure 7.2)
Nres = NsymbOFDM * Ncarr_act #Total resources in terms of symbols
k = int(np.log2(M)) #Bits per symbol
Nbits = int(Nres*k) #Total number of bits
print(" \%d bits were generated" \% (Nbits))
[Link](SEED)
Bn = [Link](0,2,Nbits)
Eb = 1

#Standard: QPSK, 16QAM, 64QAM and 256QAM

if M != 4 and M != 16 and M != 64 and M!= 256:

print("Invalid modulation. Formats used within 5G-NR mobile
communications system: QPSK, 16QAM, 64QAM and 256QAM.")
[Link](0)

[BnI,BnQ,AnI,AnQ,AI,AQ] = [Link](Bn, Eb, [Link](M),[Link](M))

if (PlotOptions == True):
[Link](figsize=(5,5))
for ii in range(0,[Link]):
for jj in range(0,[Link]):
[Link](AI[ii],AQ[jj], 'ro', label='Constellation')
[Link]('I'); [Link]('Q');
[Link](True);
[Link]('Constellation used')
[Link]('../results/[Link]')
Alphabet = AI
const_points = [Link]((1,[Link]),dtype=np.complex_)
const_points.real = AnI
const_points.imag = AnQ
const_points = [Link](const_points,(Ncarr_act,NsymbOFDM))
alpha = [Link]([Link](const_points))

ofdm_freq = [Link]((FFTSize,NsymbOFDM));
ofdm_freq = ofdm_freq.astype([Link])
#Adding information
aux = 0
for ii in range(0,NsymbOFDM):
for jj in range((int)((FFTSize/2)-(Ncarr_act/2)),(int)((FFTSize/2)+(
Ncarr_act/2))):
ofdm_freq[jj,ii] = const_points[aux,ii]
aux = aux + 1
aux = 0
ofdm_freq = utils.OverSampling_Frequency(ofdm_freq,OvS_freq, True)
if (PlotOptions == True):
[Link]()
[Link]([Link](ofdm_freq[:,0]), use_line_collection = True)
[Link]('Subcarriers')
[Link](True)
[Link]('Subcarriers Tx after Oversampling')
[Link]('../results/[Link]')

ofdm_freq = [Link](ofdm_freq)
ofdm_time = [Link](ofdm_freq.shape,dtype=[Link])
for i in range(0,NsymbOFDM):
 A.2 Main 5G-NR toolbox script for waveform generation 39

ofdm_time[:,i] = [Link](ofdm_freq[:,i])

def addCP(OFDM_time,NCP):
cp = OFDM_time[-NCP:] # take the last CP samples ...
return [Link]([cp, OFDM_time]) # ... and add them to the beginning

x = [Link]([])

#Cyclic Prefix:

NCP0 = 160/2048*FFTSize *OvS_freq #Overhead - Oversampling done before must

be taken into account
NCP = 144/2048*FFTSize *OvS_freq #Overhead 7\%

for ii in range(0,Nslots*2):
x = [Link]([x,addCP(ofdm_time[:,ii*7],(int)(NCP0))])
for jj in range(7*ii+1,7*ii+7):
x = [Link]([x,addCP(ofdm_time[:,jj],(int)(NCP))])
x = [Link](x,Fs,Fo)

return x, Bn, Alphabet, alpha, Eb, const_points, Bw

def rx_OFDM(Ncarr_act, NsymbOFDM, y, Alphabet, alpha, FFTSize, OvS_freq,

PlotOptions, Simulation, BwCh, Fs):
Nsamples = int(FFTSize*OvS_freq) #Nsamples -> We have an oversampling
that will be fixed in frequency domain
rx_ofdm_time = [Link]((Nsamples, NsymbOFDM),dtype=np.complex_)
y = [Link](y,(int([Link]/(int(NsymbOFDM/14*2))),int(NsymbOFDM/14*2))
,order='F')
#Cyclic Prefix:
NCP0 = int(160/2048*Nsamples) #Overhead
NCP = int(144/2048*Nsamples) #Overhead 7
for ii in range(0,int(NsymbOFDM/14*2)):
rx_ofdm_time[:,ii*7] = y[NCP0:(NCP0+Nsamples),ii]
for jj in range(1,7):
rx_ofdm_time[:,jj+ii*7] = y[NCP0+Nsamples+(NCP+Nsamples)*(jj-1)+
NCP:NCP0+Nsamples+(NCP+Nsamples)*(jj),ii]
ACPR, ACPR2 = utils.ACPR_Calculation(rx_ofdm_time, Fs, FFTSize, BwCh,
OvS_freq)
print("Adjacent 1st Channel Power Ratio (ACPR): \%.2f dB" \% ACPR)
if OvS_freq > 5:
print("Adjacent 2nd Channel Power Ratio (ACPR): \%.2f dB" \% ACPR2)
rx_ofdm_freq = [Link]((Nsamples, NsymbOFDM),dtype=np.complex_)
for i in range(0,NsymbOFDM):
rx_ofdm_freq[:,i] = [Link](rx_ofdm_time[:,i])
rx_ofdm_freq = [Link](rx_ofdm_freq)
if (PlotOptions == True):
[Link]()
[Link]([Link](rx_ofdm_freq[:,0]), use_line_collection = True)
[Link]('Subcarriers')
[Link](True)
[Link]('Subcarriers Rx')
[Link]('../results/[Link]')

#Downsampling: We want to recover our original signal (Fo MHz instead

of Fs MHz)
40 Chapter A. Implementation of the project

rx_ofdm_freq = utils.OverSampling_Frequency(rx_ofdm_freq,OvS_freq,False)
#False means we want downsampling (remove zeros)
rx_const_points = [Link]((Ncarr_act,NsymbOFDM),dtype=np.complex_)
aux = 0
for ii in range(0,NsymbOFDM):
for jj in range((int)((FFTSize/2)-(Ncarr_act/2)),(int)((FFTSize/2)+(
Ncarr_act/2))):
rx_const_points[aux,ii] = rx_ofdm_freq[jj,ii]
aux = aux + 1
aux = 0
# From subcarriers to symbols...
beta = [Link]([Link](rx_const_points))
rx_const_points = rx_const_points * (alpha/beta)
rx_AnI = rx_const_points.real
rx_AnQ = rx_const_points.imag
rx_AnI = rx_AnI.flatten()
rx_AnQ = rx_AnQ.flatten()
est_AnI = [Link](rx_AnI, Alphabet)
est_AnQ = [Link](rx_AnQ, Alphabet)
rx_BnI = utils.symb2bit(est_AnI, Alphabet)
rx_BnQ = utils.symb2bit(est_AnQ, Alphabet)
k = int(np.log2([Link]))
rx_Bn = [Link]([],dtype=[Link])
# ... and now bits (parallel to series conversion included here)
for ii in range(0,int(rx_BnI.size/k)):
rx_Bn = [Link]([rx_Bn,rx_BnI[ii*k:ii*k+k]])
rx_Bn = [Link]([rx_Bn,rx_BnQ[ii*k:ii*k+k]])
return rx_Bn, rx_const_points, ACPR, ACPR2

def Channel(x, SNRdb, Chtype):

"""
Description
----------
Channel model of the simulated environment

Parameters
----------
x : numpy of array complex
input signal.
SNRdb : float
signal to noise ratio specified in decibels.
Chtype : string
Channel type.

Returns
-------
y : numpy of array complex
output signal of the channel stage.

"""
if (Chtype == "AWGN"):
nsr = 10**(-SNRdb/10)
Ps = [Link]([Link](x**2))/[Link]
n = [Link](len(x)*2).view(np.complex128)/[Link](2) #Noise
with power 1
n = [Link](Ps*nsr)*n #We set noise power to meet SNR specifications
y = x + n
 A.2 Main 5G-NR toolbox script for waveform generation 41

return y

def MulticarrierReceiver(NRB, MyStructure, y, Nslots, Alphabet, alpha, FFTSize,

Fs, Fo, PlotOptions, Simulation, BwCh, Filtering):
Ncarr_act = NRB * 12
NsymbOFDM = Nslots*14
OvS = Fs / Fo
OvS_freq = [Link](OvS)
f, Pyy = [Link](y, Fs, FFTSize)
Pyy = 10*np.log10(Pyy) + 30
if (PlotOptions == True):
[Link](figsize=(5, 4))
[Link](f,Pyy)
[Link]('frequency [MHz]')
[Link]('PSD [dBm/Hz]')
[Link]('Power Spectral Density of the signal received')
[Link]()
[Link](True);
[Link]('../results/[Link]')

#Interpolation in time domain (Downsampling)

y = [Link](y,Fo*OvS_freq,Fs)

rx_Bn, rx_const_points, ACPR, ACPR2 = rx_OFDM(int(Ncarr_act), int(NsymbOFDM

), y, Alphabet, alpha,FFTSize, OvS_freq, PlotOptions, Simulation, BwCh,
Fs)
return rx_Bn, rx_const_points, ACPR, ACPR2
42 Chapter A. Implementation of the project

A.3 Some other util functions

Code A.3 Some other util functions ([Link]).

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Author: Guillermo Palomino Lozano
Organization: University of Seville
Date: May 2020
"""
import sys
import os
import numpy as np
import [Link] as plt
from math import pi
from [Link] import kaiserord, firwin, freqz, welch
from [Link] import erfc
from [Link] import GrayCode

def PSD(y, Fs, FFTSize):

"""
Description
-----------
Compute the power spectral density for a given signal with a frequency
sampling Fs and using FFTSize points

Parameters
----------
y : numpy complex array
input signal.
Fs : float
frecuency sampling.
FFTSize : int
number of points required for the computing of the fast fourier
transform algorithm.

Returns
-------
f : numpy float array
frequency grid.
Pxx : numpy complex array
power spectral density of the given signal y.

"""

f_aux, Pxx_aux = welch(y, Fs, nperseg=FFTSize)

f = [Link]([])
Pxx = [Link]([])
length = len(f_aux)
f=[Link]((f,f_aux[int(FFTSize/2):length]))
f=[Link]((f,f_aux[0:int(FFTSize/2)]))
f = f/1e6
Pxx=[Link]((Pxx,Pxx_aux[int(FFTSize/2):length]))
 A.3 Some other util functions 43

Pxx=[Link]((Pxx,Pxx_aux[0:int(FFTSize/2)]))
return f, Pxx

def ACPR_Calculation(y, Fs, FFTSize, BwCh, OvS):

"""
Description
----------
Compute the adjacent channel power ratio (ACPR) for a given signal y

Parameters
----------
y : numpy complex array
input signal.
Fs : float
frecuency sampling.
FFTSize : int
number of points required for the computing of the fast fourier
transform algorithm.
BwCh : int
Channel bandwidth.
OvS : float
Oversampling needed for meet the requirement of the frequency sampling
desired.

Returns
-------
ACPR : float
ACPR of the first adjacent channel.
ACPR2 : float
ACPR of the second adjacent channel.

"""

Nsamples, Nsymb = [Link]

Pchann = [Link](Nsymb)
Pabov = [Link](Nsymb)
Pbelw = [Link](Nsymb)

Pabov2 = [Link](Nsymb)
Pbelw2 = [Link](Nsymb)
for ii in range(Nsymb):
yaux = y[:,ii]

f, psdy = PSD(yaux, Fs, FFTSize)

idx_chann = (f > -(BwCh/2)) * (f < (BwCh/2))

idx_abov = (f > (BwCh/2)) * (f < ((BwCh/2) + BwCh))
idx_belw = (f < (-BwCh/2)) * (f > ((-BwCh/2) - BwCh))

Pchann[ii] = [Link](psdy[idx_chann])*BwCh*1e6
Pabov[ii] = [Link](psdy[idx_abov])*BwCh*1e6
Pbelw[ii] = [Link](psdy[idx_belw])*BwCh*1e6

if OvS > 5:
#Also the second adjacent channel

idx_abov2 = (f > (BwCh/2 + BwCh)) * (f < ((BwCh/2) + 2*BwCh))

44 Chapter A. Implementation of the project

idx_belw2 = (f < (-BwCh/2) - BwCh) * (f > ((-BwCh/2) - 2*BwCh))

Pabov2[ii] = [Link](psdy[idx_abov2])*BwCh
Pbelw2[ii] = [Link](psdy[idx_belw2])*BwCh

ACPR = 10*np.log10([Link]([Link]([[Link](Pabov),[Link](Pbelw)]))/np.
mean(Pchann))
ACPR2 = False
if OvS > 5:
ACPR2 = 10*np.log10([Link]([Link]([[Link](Pabov2),[Link](Pbelw2)]))
/[Link](Pchann))
return ACPR, ACPR2
def EVM_Calculation(const_points, rx_const_points):
"""
Description
-----------
Compute the error vector magnitude (EVM) between the original and resulting
point of the constellation

Parameters
----------
const_points : numpy array complex
constellation points transmitted.
rx_const_points : numpy array complex
constellation points received.

Returns
-------
EVM : float
EVM performance.

"""
EVM_Reference = [Link]([Link](const_points.real**2+const_points.imag**2))
err = const_points - rx_const_points;
EVM = [Link](1/[Link]*[Link]([Link]**2 + [Link]**2))/EVM_Reference *
100
return EVM

def NMSE_Calculation(x,y):
"""
Description
----------
Compute the normalized mean square error for to given values x and y

Parameters
----------
x : numpy array of complex
signal 1.
y : numpy array of complex
sigtnal 2.

Returns
-------
float
NMSE.
 A.3 Some other util functions 45

"""
return ([Link]([Link]((x - y))**2))/([Link]([Link](x)**2))

def DrawConstellation(const_points, rx_const_points, NUMPOINTS):

"""
Description
----------
Draw the constellation with tx and rx points, for a given number of points
NUMPOINTS

Parameters
----------
const_points : numpy array complex
constellation points transmitted.
rx_const_points : numpy array complex
constellation points received.
NUMPOINTS : int
Number of points desired for plotting the constellation.

Returns
-------
None.

"""
[Link](figsize=(5,5))
index = [Link](0, high=len(rx_const_points), size=NUMPOINTS)
for ii in range(len(index)):
[Link](rx_const_points.real[ii],rx_const_points.imag[ii],'-o',color='
blue')
[Link](const_points.real[ii],const_points.imag[ii],'-o',color='red')
[Link]('I'); [Link]('Q');
[Link](True);
[Link]('Constellation received')
[Link]('../results/[Link]')

def OverSampling_Frequency(X,OvS,Upsampling):
"""
Description
----------
Perform the oversampling in frequency domain by adding/removing zeros

Parameters
----------
X : numpy array complex
input signal.
OvS : float
Oversampling factor.
Upsampling : boolean
True if upsampling must be performed, false if downsampling.

Returns
-------
Y : numpy array complex
input signal after oversampling.

"""
#Zero padding OvS times the length of X -> we change fs
46 Chapter A. Implementation of the project

if (OvS != 1):
(Nx,Nsymb) = [Link](X)
if (Upsampling == True):
#Upsampling: Add zeros
Ny = [Link](Nx*OvS)
Y = [Link]((int(Ny),Nsymb),dtype=np.complex_)
if Nx \% 2 == 0:
[Link][int(Ny/2 - [Link](Nx/2)):int(Ny/2 + [Link](Nx/2)),:]
= [Link]
[Link][int(Ny/2 - [Link](Nx/2)):int(Ny/2 + [Link](Nx/2)),:]
= [Link]
else:
[Link][int(Ny/2 - [Link](Nx/2)):int(Ny/2 + [Link](Nx/2)+1)
,:] = [Link]
[Link][int(Ny/2 - [Link](Nx/2)):int(Ny/2 + [Link](Nx/2)+1)
,:] = [Link]
else:
#Downsampling: Remove zeros
Ny = [Link](Nx/OvS)
Y = [Link]((int(Ny),Nsymb),dtype=np.complex_)
if Nx \% 2 == 0:
[Link] = [Link][int(Nx/2 - [Link](Ny/2)):int(Nx/2 + [Link](
Ny/2)),:]
[Link] = [Link][int(Nx/2 - [Link](Ny/2)):int(Nx/2 + [Link](
Ny/2)),:]
else:
[Link] = [Link][int(Nx/2 - [Link](Ny/2)):int(Nx/2 + [Link](
Ny/2)+1),:]
[Link] = [Link][int(Nx/2 - [Link](Ny/2)):int(Nx/2 + [Link](
Ny/2)+1),:]
else:
Y = X
return Y

def RootRaisedCos(Nfilt,B=0,T=1):
"""
Description
----------
Define root raised cosine function

Parameters
----------
Nfilt : int
Filter length.
B : float, optional
Roll-off factor of root raised cosine function. The default is 0.
T : float, optional
Reciprocal of the symbol-rate. The default is 1.

Returns
-------
TYPE
DESCRIPTION.

"""
x = [Link](-Nfilt/2, Nfilt/2, Nfilt+1)
 A.3 Some other util functions 47

return (1/[Link](T))*([Link](pi*x/T*(1-B))+4*B*x/T*[Link](pi*x/T*(1+B)))/(
pi*x/T*(1-(4*B*x/T)**2))

def primeFactors(n):
"""
Description
----------
Compute the prime factors of a number n

Parameters
----------
n : int
input number.

Returns
-------
result : array of float
prime factors of number n.

"""

result = [Link]([],dtype=[Link])
# Print the number of two's that divide n
while n \% 2 == 0:
result = [Link](result,2)
n = n / 2
# n must be odd at this point
# so a skip of 2 ( i = i + 2) can be used
for i in range(3,int([Link](n))+1,2):

# while i divides n , print i ad divide n

while n \% i== 0:
result = [Link](result,i)
n = n / i

# Condition if n is a prime
# number greater than 2
if n > 2:
result = [Link](result,n)

return result

def resample_quotients(fs1, fs2):

"""
Description
----------
Compute the P and Q resampling coefficients to be used in FFTinterpolate

Parameters
----------
fs1 : int
frequency sampling 1.
fs2 : int
frequency sampling 2.

Returns
-------
48 Chapter A. Implementation of the project

P : float
P coefficient.
Q : float
Q coefficient.

"""
v1 = primeFactors(fs1)
v2 = primeFactors(fs2)
total_ind = [Link]([])
for k in range(0,[Link]):
#If we can find element k of v1 in v2
if [Link](v1[k],v2):
#Find first index in v2 where it can be found
ind = [Link](v1[k] == v2)[0][0]
#Remove the value at index k from v1
total_ind = [Link](total_ind,k)
#Remove the value at index ind from v2
v2 = [Link](v2,ind)

P = [Link](v1[np.setdiff1d([Link](0,[Link]),total_ind)])
Q = [Link](v2)
return P,Q

def FFTinterpolate(x,fs_y,fs_u):
"""
Description
----------
Interpolation performed by FFt algorithm

Parameters
----------
x : numpy array of complex
input signal.
fs_y : float
sampling rate of the output signal.
fs_u : TYPE
sampling rate of the input signal.

Returns
-------
y : TYPE
DESCRIPTION.

"""
if (fs_u != fs_y):
N = [Link]
P, Q = resample_quotients(fs_y, fs_u)
Nn = float(N*P/Q)
U = [Link](x)
Y = [Link]((int(Nn)),dtype=np.complex_)
if ([Link](Nn) == Nn):
Nn = int(Nn)
Y[Nn-1] = (1j)*10**(-16)
if (P > Q):
#Upsampling
if (Nn \% 2) == 0:
if (N \% 2) == 0:
 A.3 Some other util functions 49

#Even number of samples in u

#Easy to put back
N = int(N)
Y[0:int(N/2)] = U[0:int(N/2)]
Y[Nn-int(N/2):Nn] = U[int(N/2):N]
else:
Y[0:int([Link](N/2))] = U[0:int([Link](N/2))]
Y[Nn-int([Link](N/2)):Nn] = U[int([Link](N/2)):N]
else:
print("Not implemented")
[Link](0)
else:
#Downsampling
Y[0:int([Link](Nn/2))] = U[0:int([Link](Nn/2))]
Y[Nn-int([Link](Nn/2)):Nn] = U[N-int([Link](Nn/2)):N]
else:
print("Not an integer number of samples. Use some other method")
[Link](0)
y = [Link](Y)
else:
y = x
return y

def Filter_signal(Fs, BwCh, Bw, x):

"""
Description
----------
Function that filter an input signal using the Kaiser window method.
Parameters
----------
Fs : float
frequency sampling.
BwCh : float
Channel bandwidth available.
Bw : float
bandwidth of the input signal.
x : numpy array of complex
input signal.

Returns
-------
y : numpy array of complex
signal filtered.

"""
# The Nyquist rate of the signal.
nyq_freq = Fs / 2.0

Fend = (BwCh / 2) / nyq_freq

# The cutoff frequency of the filter.
Fc = (Bw*1.025 / 2) / nyq_freq

# The desired width of the transition from pass to stop,

# relative to the Nyquist rate. We'll design the filter
# with a 5 Hz transition width.
width = Fend - Fc
# The desired attenuation in the stop band, in dB.
50 Chapter A. Implementation of the project

ripple_db = 100

# Compute the order and Kaiser parameter for the FIR filter.
N, beta = kaiserord(ripple_db, width)
if (N \% 2 == 0):
N = N+1
# Use firwin with a Kaiser window to create a lowpass FIR filter.
taps = firwin(N, Fc , window=('kaiser', beta))
delay = int(0.5 * (N))
# Use lfilter to filter x with the FIR filter.
#yz = lfilter(taps, 1.0, xz)
#y = yz[delay:]
yaux = [Link]([Link](x)*[Link](taps,[Link]))
y = [Link]([])
y = [Link](y,yaux[delay:])
y = [Link](y,yaux[0:delay])
#------------------------------------------------
# Plot the magnitude response of the filter.
#------------------------------------------------

[Link](2)
[Link]()
w, h = freqz(taps, worN=8000)
[Link]((w/pi)*nyq_freq/1e6, 10*np.log10([Link](h)), linewidth=2, label="H
(f)")
f, Pxx = PSD(x, Fs, 8000)
Pxx = 10*np.log10(Pxx) + 30
f2, Pyy = PSD(yaux, Fs, 8000)
Pyy = 10*np.log10(Pyy) + 30
[Link](f,Pxx - [Link](Pxx),label="X(f)")
[Link](f2,Pyy- [Link](Pyy),label="Y(f)")
[Link]('Frequency (MHz)')
[Link]('Gain')
[Link]('Frequency Response')
[Link](loc="upper right")
[Link](True)
[Link]('frequency [MHz]')
[Link]('PSD [dBm/Hz]')
[Link]()
[Link](True);
[Link]('Filter effect')
[Link](True)
[Link]('../results/[Link]')
return y

def split(Bn, M1, M2):

k1=int(np.log2(M1))
k2=int(np.log2(M2))
k=k1+k2
Nb=len(Bn)
W=[Link](Bn,[int(Nb/k),k])
BnI=[Link](W[:,:k1],[k1*int(Nb/k)])
BnQ=[Link](W[:,k1:],[k2*int(Nb/k)])
return BnI,BnQ

def de2gray(d,n):
 A.3 Some other util functions 51

gray_list_str = list(GrayCode(n).generate_gray())
gray_list = list(map(lambda ind: [Link](list(gray_list_str[ind]), dtype=
[Link]), range(len(gray_list_str))))
g = list(map(lambda ind: gray_list[int(d[ind])], range(0,len(d))))

return [Link](g)

def detectsymb(rn, alfabeto):

N = len(rn)

An = [Link](N)

for i in range(N):
ind = [Link](abs(rn[i]-alfabeto) == [Link](abs(rn[i]-alfabeto)))
An[i] = alfabeto[ind]

return An

def gray2de(b):
"""
Description
----------
Gray to decimal converter

Parameters
----------
b : numpy array of int
input bits.

Returns
-------
d : numpy array of int
output bits.

"""
from numpy import zeros_like, logical_xor, fliplr, shape, reshape, matrix,
arange, array
b=matrix(b)
c=zeros_like(b)
c[:,0] = b[:,0];

for i in range(1,shape(b)[1]):
c[:,i] = logical_xor(c[:,i-1], b[:,i])

c=fliplr(c);

[n,m] = shape(c)
if min([m,n]) < 1:
d = [];
return
elif min([n,m]) == 1:
m = max([n,m])
n = 1
c = reshape(c,[n,m])
52 Chapter A. Implementation of the project

d = array((c * matrix(2**arange(m)).T).T).squeeze();

return d

def Qfunct(x):
"""
Calculation of Q function for BER performance
----------

Parameters
----------
x : numpy array of complex
input signal.

Returns
-------
res : float
Q(x).

"""
res=(1/2)*erfc(x/[Link](2))
return res

def symb2bit(An,Alphabet):
"""
Description
----------
Symbol to bit converter

Parameters
----------
An : numpy array of float
array of symbols.
Alphabet : numpy array of float
alphabet of the QAM modulation.

Returns
-------
Bn : numpy array of int
sequence of bits corresponding to the input symbols.

"""
k = np.log2(len(Alphabet))

if k>1:
dist = abs(Alphabet[0]-Alphabet[1])
idx = [Link]((An-Alphabet[0])/dist)
Bn = [Link](de2gray(idx,k),[int(k*len(An))])
else:
Bn = ((An/max(Alphabet))+1)/2

return Bn

def qam(Bn, Eb, M1, M2):

"""
 A.3 Some other util functions 53

Description
-----------
QAM modulator

Parameters
----------
Bn : numpy array of int
sequence of bits to transmit.
Eb : float
Energy per bit.
M1 : int
Modulation order 1.
M2 : int
Modulation order 2.

Returns
-------
BnI : numpy array of int
bits sequence of in-phase component.
BnQ : numpy array of int
bit sequence of quadrature component.
AnI : numpy array of float
transmitted levels in in-phase component.
AnQ : numpy array of float
transmitted levels in quadrature component.
AI : numpy array of float
used levels used in in-phase component.
AQ : numpy array of float
used levels in quadrature component.

"""

k1=int([Link](np.log2(M1)))
M1=2**(k1)
k2=int([Link](np.log2(M2)))
M2=2**(k2)

k=k1+k2
Nb=len(Bn)
Bn=np.r_[Bn, [Link](int(k*[Link](Nb/k)-Nb), dtype=int)]

A= [Link](3*Eb*np.log2(M1*M2)/(M1**2+M2**2-2))

AI=A*(2*[Link](M1)-M1+1)
AQ=A*(2*[Link](M2)-M2+1)

BnI,BnQ=split(Bn,M1,M2)

NbI=len(BnI)
NbQ=len(BnQ)

if M1>2:
AnI=AI[gray2de([Link](BnI,[int(NbI/k1),k1]))]
else:
AnI=AI[BnI]
54 Chapter A. Implementation of the project

if M2>2:
AnQ=AQ[gray2de([Link](BnQ,[int(NbQ/k2),k2]))]
else:
AnQ=AQ[BnQ]

return BnI,BnQ,AnI,AnQ,AI,AQ
 A.4 Extra code for some figures generation 55

A.4 Extra code for some figures generation

Code A.4 Script that generates Fig.4 4.2 and Fig. 4.3.
# -*- coding: utf-8 -*-
"""
Author: Guillermo Palomino Lozano
Organization: University of Seville
Date: May 2020
"""
import numpy as np
import os
import [Link] as plt
import [Link] as sio
import tikzplotlib
import mc5g as NR
import ccdd as ccdd
import [Link]
from [Link] import ScalarFormatter
import warnings
#------------------------------------------------------------
[Link]('all')

NRB = 250 # Number of resource blocks

Df = 15e3 # Subcarrier Spacing
OvS = True # Oversampling
Fs = 100e6 # Only important when oversampling is true
Nslots = 1
FR = 1 # Frecuency range
PAPR = 10.5 #PAPR desiredPlotOptions = False #True for active plots (
Subcarriers Tx/Rx, Constellation, Spectrum, etc )
#-----------------------------------------------Pl-------------
# Simulations available:
Simulation = False
#------------------------------------------------------------
[Link]('ignore')
NUMSIM = 1000
PAPRx = [Link](NUMSIM)
PAPRy = [Link](NUMSIM)
Nsamples = [Link](NUMSIM)
M = [Link]([16])
k = np.log2(M)
MyStructure = [Link](Df,FR)
print("\n------------- STARTING TX -------------")
if Simulation == True:
for jj in range(0,1000):
SEED = jj
print("Simulation number %d" % jj)
x, Bn, FFTSize, Alphabet, alpha, Eb, const_points, Fs, Fo, BwCh, PAPRx[
jj], PAPRy[jj], Nsamples[jj] = [Link](MyStructure
, NRB, OvS, Fs, Nslots, M, PAPR, SEED, False, False)
[Link]('[Link]',PAPRx=PAPRx,PAPRy=PAPRy, Nsamples=Nsamples)
else:
data = [Link]('[Link]')
PAPRx = data['PAPRx']
56 Chapter A. Implementation of the project

PAPRy = data['PAPRy']
[Link]()
n, bins, patches = [Link](x=PAPRx, bins=30, color='#607c8e',alpha=0.7, rwidth
=0.85)
[Link](axis='y', alpha=0.75)
[Link]('PAPR (dB)')
[Link]('Counts')
[Link]("[Link]")

[Link]()
[Link]([Link](0,1000),PAPRy, label='Output')
[Link]([Link](0,1000),PAPRx, label='Input')
[Link]()
[Link]('Simulations')
[Link]('PAPR (dB)')
[Link]()
[Link]("[Link]")

error = [Link]([Link](10.5 - PAPRy)) * 100

print("PAPR error after peak clipping: %.5f" % (error))
print("Average of samples affected: %.2f " % ([Link](Nsamples)))
 A.4 Extra code for some figures generation 57

Code A.5 Script that generates Fig. 4.7 and Fig. 4.6.
# -*- coding: utf-8 -*-
"""
Author: Guillermo Palomino Lozano
Organization: University of Seville
Date: May 2020
"""

import numpy as np
import [Link] as plt
import tikzplotlib
[Link]('all')
data = [Link]('[Link]')
Xamp = data['X_amp']
Yamp = data['Y_amp']
Pin = data['Pin']
XdB = 10*np.log10([Link](Xamp)**2/50)+30
YdB = 10*np.log10([Link](Yamp)**2/50)+30
G = YdB - XdB
leg=["-30dBm","-29dBm","-28dBm","-27dBm","-26dBm","-25dBm","-24dBm","-23dBm
","-22dBm","-21dBm"]
[Link]()
for ii in range(len(Pin)):
[Link](XdB[len(Pin) - ii - 1,:], G[len(Pin) - ii - 1,:],s=0.5, label=
leg[len(Pin) - ii - 1])
[Link]('Input Power')
[Link]('Gain')
[Link]('Gain / AM')
[Link](-50, -8)
[Link](25, 65)
[Link]()
for ii in range(len(Pin)):
[Link](XdB[len(Pin) - ii - 1,:], YdB[len(Pin) - ii - 1,:],s=0.5,label=
leg[len(Pin) - ii - 1])
[Link]('Input Power')
[Link]('Output Power')
[Link]('AM / AM')
[Link]((-30, 65))
 List of Figures

1.1 Road to 5G [Remmert, 2019] 2

1.2 Gantt diagram associated to the project planning 4

2.1 Slots in 5G according to subcarrier spacings of 15, 30, 600, 120, and 240 kHz 10

3.1 Flow chart of the simulation script 14

3.2 Screenshot that shows and example of what happen when parameters specified
does not match the standard of 5G NR 15
3.3 Diagram of the multicarrier generation stage (ovs stands for oversampling) 15
3.4 Constellation for all the modulation order generated in 5G NR: QPSK, 16QAM,
64QAM and 256QAM 17
3.5 Due to the filter delay, it is necessary to reorder the output signal 18
3.6 Spectrum shaping filter: how important is the choice of cut-off frequency. 97.5 vs 102.5 20
3.7 Screenshot of my capsule at Codeocean 21

4.1 5G waveform for a subcarrier spacing equal to 60 kHz and 5 slots 24

4.2 Histogram of the PAPR, obtained for 1000 signals generated 24
4.3 Input and output PAPR of the peak clipping stage along 1000 simulations 25
4.4 Average Bit Error Rate (BER) versus a sweep in the bit energy to noise power ratio 26
4.5 Error Vector Magnitude (EVM) versus a sweep in the bit energy to noise power ratio 27
4.6 AM-AM characterization of the power amplifier 28
4.7 Gain-AM characterization of the power amplifier 29

59
 List of Tables

2.1 Numerology in 5G NR 8
2.2 Frequency range related to numerology in 5G 11

4.1 Numerology for PAPR simulation 24

61
 List of Codes

A.1 Simulation script that uses the toolbox presented in this work ([Link]) 33
A.2 Main 5GNR toolbox script for waveform generation ([Link]) 37
A.3 Some other util functions ([Link]) 44
A.4 Script that generates Fig.4 4.2 and Fig. 4.3 57
A.5 Script that generates Fig. 4.7 and Fig. 4.6 59

63
 Bibliography

[ETS, 2018] (2018). Physical channels and modulation, 3GPP TS 38.211 version 15.2.0 Release 15.
[Cis, 2020] (2020). Cisco annual internet report (2018–2023). Technical report, White paper, Cisco public.
[ROA, 2020] (2020). Release timeline 5G. 3GPP, ETSI.
[Ari and Ustazhanov, 2014] Ari, N. and Ustazhanov, M. (2014). Matplotlib in Python. 2014 11th Interna-
tional Conference on Electronics, Computer and Computation (ICECCO), pages 1–6.
[Cheng et al., 2016] Cheng, X., He, Y., Ge, B., and He, C. (2016). A filtered OFDM using FIR filter based
on window function method. In 2016 IEEE 83rd Vehicular Technology Conference (VTC Spring). IEEE.
[Dahlman et al., 2018] Dahlman, E., Parkvall, S., and Sköld, J. (2018). Overall transmission structure. In
5G NR: the Next Generation Wireless Access Technology, pages 103–131. Elsevier.
[Fangohr, 2004] Fangohr, H. (2004). A comparison of C, MATLAB, and Python as teaching languages in
engineering. In Computational Science - ICCS 2004, pages 1210–1217. Springer Berlin Heidelberg.
[Ghosh et al., 2019] Ghosh, A., Maeder, A., Baker, M., and Chandramouli, D. (2019). 5G evolution: A view
on 5G cellular technology beyond 3GPP release 15. IEEE Access, 7:127639–127651.
[GMB, 2006] GMB, N. R. (2006). Technology of high speed packet access (hspa).
[Gomes et al., 2018] Gomes, R., Sismeiro, L., Ribeiro, C., Fernandes, T. R., Sanchez, M. G., Hammoudeh,
A., and Caldeirinha, R. F. S. (2018). Will COTS RF front-ends really cope with 5G requirements at
mmWave? IEEE Access, 6:38745–38769.
[Guan et al., 2017] Guan, P., Wu, D., Tian, T., Zhou, J., Zhang, X., Gu, L., Benjebbour, A., Iwabuchi, M.,
and Kishiyama, Y. (2017). 5G field trials: OFDM-based waveforms and mixed numerologies. IEEE J. Sel.
Areas Commun., 35(6):1234–1243.
[Gupta and Jha, 2015] Gupta, A. and Jha, R. K. (2015). A survey of 5G network: Architecture and emerging
technologies. IEEE Access, 3:1206–1232.
[Henry et al., 2020] Henry, S., Alsohaily, A., and Sousa, E. S. (2020). 5G is real: Evaluating the compliance
of the 3GPP 5G new radio system with the ITU IMT-2020 requirements. IEEE Access, 8:42828–42840.
[Ijaz et al., 2016] Ijaz, A., Zhang, L., Grau, M., Mohamed, A., Vural, S., Quddus, A. U., Imran, M. A., Foh,
C. H., and Tafazolli, R. (2016). Enabling massive IoT in 5G and beyond systems: PHY radio frame design
considerations. IEEE Access, 4:3322–3339.
[ITU-R, 2015] ITU-R (2015). Framework and overall objectives of the future development of IMT2020 for
2020 and beyond. Recommendation ITU-R M.2083.
[Kaltenberger et al., 2019] Kaltenberger, F., d. Souza, G., Knopp, R., and Wang, H. (2019). The openairin-
terface 5G new radio implementation: Current status and roadmap. In WSA 2019; 23rd International ITG
Workshop on Smart Antennas, pages 1–5.

65
66 Bibliography

[Landin et al., 2015] Landin, P. N., Gustafsson, S., Fager, C., and Eriksson, T. (2015). WebLab: A Web-
Based Setup for PA Digital Predistortion and Characterization [Application Notes]. IEEE Microw. Mag.,
16(1):138–140.
[Levanen et al., 2017] Levanen, T., Kaikkonen, J., Nielsen, S., Pajukoski, K., Renfors, M., and Valkama,
M. (2017). 5G new radio UL coverage with peak clipping. In 2017 IEEE 86th Vehicular Technology
Conference (VTC-Fall). IEEE.
[Lin et al., 2017] Lin, P.-C., Huang, S.-L., and Li, X.-Y. (2017). Teaching and learning next generation
mobile communication networks through open source openAirInterface testbeds. In 2017 Asia-Pacific
Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC). IEEE.
[López, 2020] López, D. (2020). Técnicas de reducción del factor de cresta aplicadas a señales 5g new radio.
Master’s thesis, Universidad de Sevilla.
[Marcus, 2015] Marcus, M. J. (2015). 5G and "IMT for 2020 and beyond" [Spectrum Policy and Regulatory
Issues]. IEEE Wireless Commun., 22(4):2–3.
[Ochiai and Imai, 2001] Ochiai, H. and Imai, H. (2001). On the distribution of the peak-to-average power
ratio in OFDM signals. IEEE Trans. Commun., 49:282 – 289.
[Oliphant, 2004] Oliphant, T. E. (2004). Scipy tutorial.
[Oliphant, 2007] Oliphant, T. E. (2007). Python for scientific computing. Comput. Sci. Eng., 9(3):10–20.
[Oppenheim et al., 2001] Oppenheim, A. V., Buck, J. R., and Schafer, R. W. (2001). Discrete-time signal
processing. Vol. 2. Upper Saddle River, NJ: Prentice Hall.
[Osseiran et al., 2014] Osseiran, A., Boccardi, F., Braun, V., Kusume, K., Marsch, P., Maternia, M., Queseth,
O., Schellmann, M., Schotten, H., Taoka, H., Tullberg, H., Uusitalo, M. A., Timus, B., and Fallgren, M.
(2014). Scenarios for 5G mobile and wireless communications: the vision of the METIS project. IEEE
Commun. Mag., 52(5):26–35.
[Ozgur et al., 2017] Ozgur, C., Colliau, T., Rogers, G., Hughes, Z., and Bennie, E. (2017). MatLab vs.
Python vs. R. Journal of Data Science: JDS, 15:355–372.
[Palomino et al., 2020] Palomino, G., Pérez-Hernández, A., Madero-Ayora, M. J., and Becerra, J. A. (2020).
An open source project for learning 5G waveforms.
[Park et al., 2019] Park, J., Lee, E., Park, S.-H., Raymond, S.-S., Pyo, S., and Jo, H.-S. (2019). Modeling and
analysis on radio interference of OFDM waveforms for coexistence study. IEEE Access, 7:35132–35147.
[Proakis and Salehi, 2008] Proakis, J. and Salehi, M. (2008). Digital Communications. McGraw-Hill.
[Rateb and Labana, 2019] Rateb, A. M. and Labana, M. (2019). An optimal low complexity PAPR reduction
technique for next generation OFDM systems. IEEE Access, 7:16406–16420.
[Remmert, 2019] Remmert, H. (2019). What is 5G? part 1 - evolution and the next generation.
[Salama and Elmesalawy, 2019] Salama, A. I. and Elmesalawy, M. M. (2019). Flexible and adaptive testbed
for 5G experimentations. In 2019 Novel Intelligent and Leading Emerging Sciences Conference (NILES).
IEEE.
[Selva, 2015] Selva, J. (2015). FFT Interpolation from nonuniform samples lying in a regular grid. IEEE
Trans. Signal Process., 63(11):2826–2834.
[Truong et al., 2014] Truong, T.-A., Arzel, M., Lin, H., Jahan, B., and Jézéquel, M. (2014). Dft precoded
ofdm—an alternative candidate for next generation pons. J. Lightwave Technol., 32(6):1228–1238.
[van der Walt et al., 2011] van der Walt, S., Colbert, S. C., and Varoquaux, G. (2011). The numpy array: A
structure for efficient numerical computation. Comput. Sci. Eng., 13:22–30.
[Zaidi, 2017] Zaidi, A. (2017). Three design principles of 5G New Radio.