• Nenhum resultado encontrado

Digital pre-distortion for 5G transmission over high-capacity optical fronthaul fiber links

N/A
N/A
Protected

Academic year: 2021

Share "Digital pre-distortion for 5G transmission over high-capacity optical fronthaul fiber links"

Copied!
78
0
0

Texto

(1)

Universidade de Aveiro Departamento deElectr´onica, Telecomunica¸c˜oes e Inform´atica, 2019

Marco Andr´

e

Tavares Fernandes

Digital Pre-Distortion for 5G Transmission over

High-Capacity Optical Fronthaul Fiber Links

Pr´

e-Distor¸

ao Digital para Transmiss˜

ao de Sinais

5G em Redes ´

Oticas de Grande Capacidade

(2)
(3)

Universidade de Aveiro Departamento deElectr´onica, Telecomunica¸c˜oes e Inform´atica, 2019

Marco Andr´

e

Tavares Fernandes

Digital Pre-Distortion for 5G Transmission over

High-Capacity Optical Fronthaul Fiber Links

Disserta¸c˜ao apresentada `a Universidade de Aveiro para cumprimento dos requesitos necess´arios `a obten¸c˜ao do grau de Mestre em Eletr´onica e Tele-comuni¸c˜oes, realizada sob a orienta¸c˜ao cient´ıfica do Dr. Paulo Miguel Nepo-muceno Pereira Monteiro, Professor do Departamento de Eletr´onica Tele-comunica¸c˜oes e Inform´atica da Universidade de Aveiro e sob a coorienta¸c˜ao cient´ıfica do Doutor Fernando Pedro Pereira Guiomar, Investigador Auxiliar do Instituto de Telecomunica¸c˜oes de Aveiro.

(4)
(5)

O J´uri

Presidente Prof. Dr. Anibal Manuel de Oliveira Duarte

Professor Catedr´atico do Departamento de Eletr´onica Telecomunica¸c˜oes e In-form´atica da Universidade de Aveiro

Orientador Prof. Dr. Paulo Miguel Nepomuceno Pereira Monteiro

Professor associado do Departamento de Eletr´onica Telecomunica¸c˜oes e In-form´atica da Universidade de Aveiro

Arguente Prof. Dra. Maria do Carmo Raposo de Medeiros

Professora Associada da Faculdade de Ciˆencias e Engenharia da Universidade de Coimbra

(6)
(7)

agradecimentos / acknowledgements

This thesis is dedicated to all my family and friends who were always by my side during my academic career. I would like to thank also to Prof. Dr. Paulo Monteiro, Dr. Fernando Guiomar and Dr. Abel Lorences-Riesgo for the guidance and help during this thesis. Lastly I want to thank the Uni-versity of Aveiro, the IT-Aveiro and the ORCIP project for having provided all the conditions required to the realization of this thesis.

(8)
(9)

Palavras-Chave 5G, LTE, Compensa¸c˜ao N˜ao Linear, R´adio sobre Fibra Anal´ogico

Resumo A nova gera¸c˜ao de comunica¸c˜oes r´adio traz muitos desafios para o trans-porte destes mesmos sinais. Enquanto que ´e exigida maior capacidade, os requisitos para complexidade, custo, consumo de potˆencia e latˆencia s˜ao mais estritos. Esta disserta¸c˜ao explora a possibilidade de usar transmiss˜ao anal´ogica de sinais r´adio sobre fibra em redes de acesso centralizadas. ´E proposto e estudado o uso de transmissores ´oticos de baixo custo baseados em transmissores SFP adaptados para transportar sinais anal´ogicos. A performance dos transmissores ´e caracterizada experimentalmente com sinais 4G e 5G. Os campos de estudo desta disserta¸c˜ao incluem o impacto da frequˆencia da portadora, a potˆencia RF transmitida, a potˆencia ´

otica recebida e o comprimento da fibra. Adicionalmente, os benef´ıcios de usar um polinˆomio com mem´oria para compensa¸c˜ao n˜ao linear s˜ao avaliados. Esta compensa¸c˜ao n˜ao linear permite opera¸c˜oes 5G de elevada performance obtendo um EVM inferior a 3.5% para sinais na regi˜ao de frequˆencias 1 com 100 MHz de largura de banda, o que ´e inferior aos limites estabelecidos pelo 3GPP para transmiss˜ao 256QAM. Adicionalmente uma redu¸c˜ao de EVM de 5.7% para 5.4% ´e verificada quando se transmite um sinal 5G FR2 com 400 MHz de largura de banda por 20 km de fibra.

(10)
(11)

Keywords 5G, LTE, Nonlinear compensation, Analog-Radio over Fiber

Abstract The new generations of radio communications bring many challenges for the fronthaul transport. While demanding larger capacity than previous generations, low cost, low complexity, low power consumption and low latency are required characteristics. This thesis explores the possibil-ity of using analog radio-over-fiber (RoF) fronthaul transmission in a centralized-radio access network. We propose and study the use of low-cost transceivers based on commercially available small form-factor pluggable (SFP) transceivers adapted to transport analog signals.

The transceiver performance is experimentally characterized both with 4G LTE and 5G signals. The studies of this thesis include the impact of the carrier frequency, transmitted RF power, received optical power as well as distance of the fiber link. Additionally, the benefits of a low-complexity memory polynomial-based nonlinear compensation are evaluated. This nonlinear compensation enables high-performance 5G operation yielding an EVM < 3.5% in frequency region 1 (FR1) with 100 MHz bandwidth, which is below the limits specified by 3GPP for 256QAM transmission . In addition it also reduces the EVM from 5.7% to 5.4% when transmitting over 20 km a FR2 400 MHz 5G signal at an intermediate carrier frequency of 3.5 GHz.

(12)
(13)

Contents

Contents i

List of Figures iii

List of Tables v Acronyms vii 1 Introduction 1 1.1 Background . . . 1 1.2 Objectives . . . 3 1.3 Document Structure . . . 3 1.4 Contributions . . . 4

2 State of the Art 5 2.1 Network Architecture . . . 5

2.2 Fronthaul Interface . . . 6

2.2.1 Digital-Radio over Fiber . . . 7

2.2.2 Analog-Radio over Fiber . . . 8

2.3 Radio Access Technologies . . . 8

2.4 Digital Pre-Distortion . . . 9 2.4.1 Look-Up Table . . . 10 2.4.2 Memory Polynomial . . . 10 2.4.3 Volterra Series . . . 11 3 Experimental Setup 13 3.1 Experimental Setup . . . 13 3.2 Optical Transceivers . . . 15 3.2.1 Crosstalk Measurements . . . 18

3.3 Variable Optical Attenuator . . . 20

4 Experimental Results with 4G Signal 23 4.1 4G-LTE Signal Parameters . . . 23

4.2 Electrical B2B Analysis . . . 24

4.3 Transceiver Comparison . . . 26

4.3.1 Bandwidth Influence . . . 30

(14)

5 5G Digital Pre-Distortion 33 5.1 Experimental Setup . . . 33 5.1.1 RF signal . . . 34 5.1.2 Procedure . . . 35 5.2 Results with 100 MHz . . . 35 5.2.1 DPD models comparison . . . 36 5.2.2 Optical B2B analysis . . . 36 5.2.3 20 km analysis . . . 38

5.2.4 Photodiode sensitivity analysis . . . 40

5.3 Results with 400 MHz 256QAM . . . 41

5.3.1 IF analysis . . . 42

5.3.2 Optical B2B analysis . . . 42

5.3.3 20 km analysis . . . 43

5.4 Results with 400 MHz 64QAM . . . 45

6 Conclusions and Future Work 49 6.1 Conclusions . . . 49

6.2 Future Work . . . 50

Bibliography 53

(15)

List of Figures

1.1 Architecture of ORCIP infrastructure. . . 2

2.1 C-RAN architecture [1]. . . 6

2.2 Digital RoF for a C-RAN network. . . 7

2.3 Analog RoF for a C-RAN network. . . 8

2.4 Example structure of a LUT. . . 10

3.1 Schematic representation of the A-RoF transmission system implemented in the laboratory. . . 14

3.2 Picture of the actual testbench for RoF transmission implemented in the optical communications laboratory of Instituto de Telecomunica¸c˜oes. . . 14

3.3 Schematic of the board used to drive the TOSA for A-RoF transmission. . . . 15

3.4 Schematic of the board used to drive the ROSA for A-RoF reception. . . 16

3.5 Final appearance of the first adaptation of the digital transceiver to analog. . 16

3.6 Adaptation of the transceiver laser circuit in order to perform A-RoF. . . 17

3.7 Adaptation of the transceiver photodiode circuit in order to perform A-RoF. 17 3.8 Modified CWDM transceiver, in (a) there is the alterations performed, and in (b) the transceiver is already in the original package. . . 18

3.9 Diagram of the setup used to measure the transceiver crosstalk. . . 18

3.10 Dependency of the crosstalk with the RF power for a carrier frequency of 2.5 GHz. 19 3.11 Dependency of the crosstalk with the RF frequency with a RF power of 0 dBm. 20 3.12 Characterization of the electro-optic VOA and definition of a linear region of operation for received power tuning in the photodiode. . . 21

4.1 Example constellation of an LTE signal after transmission over an analog op-tical fronthaul. . . 24

4.2 Measured attenuation in the electrical B2B setup composed only of the RF measurement instrumentation (AWG, VSG and VSA). . . 25

4.3 Measured EVM in the electrical B2B setup composed only of the RF measure-ment instrumeasure-mentation (AWG, VSG and VSA). . . 26

4.4 RF attenuation imposed by the optical fronthaul, (a) corresponds to the sce-nario where we used the analog 1310 nm transceiver and (b) corresponds to the 1530 nm transceiver. . . 27

4.5 Measured EVM with the optical transceivers in an OB2B configuration, (a) corresponds to the scenario where we used the analog 1310 nm transceiver and (b) corresponds to the 1530 nm transceiver. . . 27

(16)

4.6 Best EVM measured for each transceiver (after RF power optimization) and

comparison with the electrical B2B performance. . . 28

4.7 EVM penalty introduced by the optical transceivers. . . 29

4.8 RF power range over which the EVM is degraded less than 1% with regard to the minimum EVM, (a) corresponds to the analog 1310 nm transceiver and (b) corresponds to the 1530 nm transceiver. . . 30

4.9 Best EVM measured for a LTE signal of 5, 10 and 20 MHz. . . 31

4.10 Comparison between B2B, 10 km and 20 km EVM for different attenuations in the SMF link. . . 32

5.1 Experimental fronthaul setup. . . 34

5.2 Example of a received constellation for a 5G signal. . . 35

5.3 Measured EVM in B2B using different DPD models. . . 36

5.4 Measured EVM in B2B for the cases of no DPD, linear DPD and different nonlinear DPD based on memory polynomials. . . 37

5.5 Measured EVM after 20 km SMF for the cases of no DPD and different non-linear DPD based on memory polynomials. . . 38

5.6 Measured spectra for the optimal case with DPD (3 dBm) before and after applying the model. . . 39

5.7 Measured spectra for the optimal case without DPD (0 dBm) and with DPD (3 dBm). . . 40

5.8 Measured EVM after 20 km SMF as a function of the received optical power. 41 5.9 Measured EVM for each RF carrier. . . 42

5.10 Measured EVM in B2B for the cases of no DPD, linear DPD and different nonlinear DPD based on memory polynomials. . . 43

5.11 Measured EVM after 20 km SMF for the cases of no DPD, linear DPD and different nonlinear DPD based on memory polynomials. . . 44

5.12 Measured EVM with and without DPD for the B2B scenario and the scenario with 20 km SMF. . . 44

5.13 Measured EVM in B2B for the cases of no DPD, linear DPD and different nonlinear DPD based on memory polynomials. . . 45

5.14 Measured EVM after 20 km SMF for the cases of no DPD, linear DPD and different nonlinear DPD based on memory polynomials. . . 46

5.15 Measured EVM with and without DPD for the B2B scenario and the scenario with 20 km SMF. . . 47

(17)

List of Tables

(18)
(19)

Acronyms

3GPP 3rd Generation Partnership Project 5G Fifth Generation

A-RoF Analog-Radio over Fiber ADC Analog-Digital Converter AWG Arbitary Waveform Generator B2B Back-to-Back

BBU Base Band Unit BS Base Station

C-RAN Centralized-Radio Access Network CO Central Office

CPRI Common Public Radio Interface

CWDM Coarse Wavelength Division Multiplexing D-RoF Digital-Radio over Fiber

DAC Digital-Analog Converter DML Directly Modulated Laser DMT Discrete Multi-Tone DPD Digital Pre Distortion DSP Digital Signal Processing DU Digital Unit

E/O Electrical/Optical

EML External Modulated Laser EVM Error Vector Magnitude FR1 Frequency Range 1

(20)

FR2 Frequency Range 2 H2H Human to Human H2M Human to Machine IF Intermediate Frequency LTE Long Term Evolution LUT Look-Up Table

M-MIMO Massive-Multiple Input Multiple Output M2M Machine to Machine

O/E Optical/Electrical

OBSAI Open Base Station Architecture Initiative OOK On-Off Keying

ORCIP Optical Radio Convergence Infrastructure for Communications and Power delivering ORI Open Radio equipment Interface

PAM Pulse-Amplitude Modulation QAM Quadrature Amplitude Modulation RAN Radio Access Network

RAT Radio Access Technology RF Radio Frequency

ROSA Receiver Optical Sub-Assembly RU Radio Unit

SFP Small-Form Pluggable SMF Single Mode Fiber SNR Signal to Noise Ratio TIA Trans-Impedance Amplifier

TOSA Transmitter Optical Sub-Assembly VOA Variable Optical Attenuator

VSA Vector Signal Analyzer X2X Everything to Everything

(21)

Chapter 1

Introduction

1.1

Background

With the new generations of mobile communications in mind, Fifth Generation (5G) and beyond, it continues the adoption of networks architectures like Centralized-Radio Access Network (C-RAN) [2]. These architectures have the information flowing between the antenna and a centralized Base Band Unit (BBU) through an optical link, following the specifica-tions of the Common Public Radio Interface (CPRI) standard [3]. Since the CPRI standard specifies that the data is transported digitally based on simple On-Off Keying (OOK) mod-ulation, bandwidth multiplication depending on the number of quantification bits is required to transport the data. This results in a very poor spectral efficiency since the number of quantification bits is typically beyond 10 [1]. 5G signals, with specified bandwidth up to 400 MHz, will demand higher bit rates than the current Long Term Evolution (LTE) signals whose maximum specified bandwidth is only 20 MHz. Therefore there is an urge to overcome the challenge of transporting 5G signals.

One possible way to transport 5G signal without requiring such a large bandwidth would be to increase the spectral efficiency of the modulation format. Keeping direct detection in the receiver due to the low complexity requirements of this application, the signal can be modulated in a M-Pulse-Amplitude Modulation (PAM) format. However, this type of modulation is not recommended for high values of M due to its lower resilience towards noise and nonlinearities [4]. Higher modulation and complex formats like M-Quadrature Amplitude Modulation (QAM) are still available, the adversity of this is that it requires more complex and costly hardware, specifically it needs to use external modulation in the optical transmitter and to have a coherent receiver, solutions that are robust but very costly [5]. Other alternative consist in the Discrete Multi-Tone (DMT) method where the frequency spectrum is divided in multiple sub-carriers each one transmitting a QAM signal [6, 7]. DMT however requires a more sophisticated Digital Signal Processing (DSP).

An alternative that is becoming more attractive consists in using an analog channel to transmit the signals. With analog transmission, there is the possibility of using the Radio Frequency (RF) signals to directly modulate the optical lasers, eliminating the need of an external modulator, and have a simple receiver based on just a photodiode and a Trans-Impedance Amplifier (TIA). Due to bandwidth limitations, when the RF carrier is high beyond the bandwidth of the opto-electronic components, the data could be transported in an Intermediate Frequency (IF) [8]. Another advantage consists in being able to simplify the

(22)

infrastructures removing the high-resolution Analog-Digital Converter (ADC) and Digital-Analog Converter (DAC) in the Radio Unit (RU), enabling the development of several RU per architecture since the cost is drastically reduced. Additionally, it will reduce the latency since DSP is avoided.

On the other hand, a reliable communication is challenging due to the lower immunity to noise of analog transmission as well as the presence of nonlinearities in both lasers and photodiodes. This is the main problem that has to be overcome to make analog transmission a valid solution for C-RAN architectures. In order to tackle the issues of analog transmission, Digital Pre Distortion (DPD) has been proposed to increase the transmitted power, i.e. having a larger Signal to Noise Ratio (SNR), without having distortions caused by nonlinearities.

Most of this work is still theoretical due to the lack of the existence of a dynamic testbed where all these different methods can be implemented and tested. The Optical Radio Con-vergence Infrastructure for Communications and Power delivering (ORCIP) project aims to deploy this testbed developing an infrastructure for 5G based on Cloud-Radio Access Net-work (RAN). The project consists in having several RUs deployed around University of Aveiro Campus and connected to a central location by an optical link.

Figure 1.1: Architecture of ORCIP infrastructure.

An overview of the ORCIP infrastructure can be seen in Figure 1.1. Here it is possible to see the topics mentioned before, namely the distributed RUs, the optical fronthaul and the Central Office (CO).

This thesis is a result of a collaboration with ORCIP and the resulting work will later be integrated in the infrastructure.

(23)

1.2

Objectives

The main goals for this thesis are the following:

• Develop and optimize a low-cost analog optical fronthaul based on the adaptation of low-cost SFP transceivers to analog transceivers;

• Characterize the performance in terms of frequency band, bandwidth, RF signal power driving the laser, optical receiver power and link distance with both 4G and 5G signals; • Evaluate the benefits of using low-complexity nonlinear compensation algorithms in

order to reduce the EVM degradation provided by the transceivers; • Integrate the work developed in the ORCIP infrastructure.

1.3

Document Structure

The remaining of this document is organized as follows:

• Chapter 2: Here we present the state of the art regarding current mobile network ar-chitectures, and a detail vision of the different methods of transporting radio signals through optical fiber is given. The challenges brought by the new generations of Ra-dio Access Technology (RAT) are also discussed. Lastly, a general overview of DPD algorithms are introduced;

• Chapter 3: In this chapter we describe the experimental setup used to emulate a real op-tical fronthaul, including its composing devices and components. Moreover, we explain in detail how to transform a digital commercial available SFP to an analog transceiver, being doing a new board to drive the transceiver or adapting the already existing one; • Chapter 4: This chapter shows the optimization of the setup presented in the chapter

before, for a real 20 MHz 64QAM LTE signal;

• Chapter 5: Here we discuss the procedure required to extract and apply a DPD model. The results obtained with 100 MHz and 400 MHz 5G signals are presented;

• Chapter 6: The conclusions of the thesis and future work are presented here;

• Appendix: Some additional information is presented, regarding the calculus of the EVM and how we can relate it with the SNR.

(24)

1.4

Contributions

The main contributions for the work here deployed are:

• Analysis of the concept behind the future architectures for mobile networks together with the need of evolve towards these networks;

• Analysis of the limitations that the Digital-Radio over Fiber (D-RoF) will have with future communications generations, and the advantages that the Analog-Radio over Fiber (A-RoF) can bring for high-capacity 4G and/or 5G communication;

• Automation of the measurements using MATLAB to control the laboratory instruments; • Characterization of the developed optical transceivers;

• Implementation of a simple DPD model based on memory polynomial and demonstra-tion of its performance gain with 5G signals;

• Paper published as first author to the conference ConfTele2019 to be held in the city of Lisbon with the title “Optimization of low-cost analog optical transceiver for a C-RAN outdoor testbed”;

• Paper submitted as co-author to the conference IMOC 2019 to be held in the city of Aveiro with the title “Experimental Transmission of LTE Signal Using Visible Light Communications”;

(25)

Chapter 2

State of the Art

In this chapter the state of the art of the fields of interest for this thesis is presented, starting with the current and future architectures for a radio communications system and then focusing in the optical fronthaul where concepts like A-RoF and D-RoF will be addressed, with the main focus on the A-RoF. A brief overview regarding the evolution of RAT is discussed. Lastly various DPD algorithms are described.

2.1

Network Architecture

In a traditional RF communication network architecture, each antenna has associated a Base Station (BS) with a BBU and a RU. The BBU is responsible for processing the digital signal and the RU for the exchange of the RF signals with the antenna. Since these signals are normally transported over a coaxial cable, this type of transmission requires a lot of power because the coaxial cable introduces high loss in the system [1].

With the new generation of radio communications, mobile operators are challenged not only to increase the network capacity but also they have to be able to support larger band-width signals. Network coverage is also a aspect that has always to be in the mind of mobile operators. These requirements introduce a new set of problems, namely the lack of phys-ical space, the high cost of building infrastructures that support those conditions and the incapability of having hardware that supports multiple wireless network standards [9].

In order to overcome these new challenges, a new architecture has been proposed based on a C-RAN that consists in several RUs gathering information. The whole information is then processed by the same BBU. This BBU is moved to a CO that can be several km from the antenna site, being RU the only thing that stays attached to the antenna reducing sub-stantially the power consumption. The interface between these two infrastructures is defined as fronthaul and the most common physical layer is the optical fiber [1]. More details about this interface will be given in the next section. Figure 2.1 shows a typical C-RAN architec-ture with the characteristics specified before, the BBU is here represented as Digital Unit (DU). The fronthaul interface is most commonly described as following the CPRI standards, however it is not the only solution for this interface.

The adoption of C-RAN architectures brings several advantages with regard to a typical architecture. Since the RU is the only equipment on the antenna site, the power consumption reduces significantly. Moreover the installation near the antenna makes possible to be air cooled not needing its own cooling system. Another power reduction advents from the use of

(26)

Figure 2.1: C-RAN architecture [1].

optical fiber to connect the RU to the DU, which has a lot less attenuation than a coaxial cable. This interface also brings another advantage that is the low latency between the RU and BBU. C-RAN also enables the coordination between adjacent cells increasing the channel capacity. Another advantage is related to security: in a traditional architecture some protocol is needed to ensure security in the backhaul, now since the BBU is physically located in a secure site this is no longer necessary [1].

There are obviously some limitations with traditional C-RAN, especially regarding net-work expansion and densification, i.e., the RU can have only a certain limited number of antennas associated. In the same way, the BBU can also only process the data of a certain number of RUs. These problems are already being addressed by the scientific community [10, 11, 12].

It is expected that a Centralized-RAN evolves to an architecture Cloud-RAN, where there is a BBU pool with various BBU nodes, each one with the capacities of a traditional BBU. The main difference consists in the fact that a BBU is dynamically allocated in real-time to fulfill the needs, ensuring that the resources are not being wasted which lead to a reduction of power consumption and an increase of network speed [13].

2.2

Fronthaul Interface

In the C-RAN architectures mentioned in the previous section, the interface between the RU and the BBU is named fronthaul and its physical layer is normally optical fiber. There are two general ways of transporting the radio signals in the fiber, D-RoF and A-RoF. Typically the method most used for a C-RAN network is the D-RoF, however with the new challenges that the new generations of radio communications will bring A-RoF has been studied to be

(27)

2.2.1 Digital-Radio over Fiber

A general overview of D-RoF can be seen in Figure 2.2. Here it is possible to see that in the downlink case the digital signal in the BBU is directly converted to the optical domain and then sent to the fiber. In the RU side, the signal is converted again to the electric domain and only then it goes through a DAC for conversion to the analog domain and prepared to be sent to the antenna. In the uplink the received analog signal in the antenna is amplified and filtered, then it is converted to the digital domain and sent through the fronthaul. The optical signal is then received in the BBU by a photodiode. Since the transmission is digital, the signal can already be processed.

Digital Signal Processing AMP AMP DAC ADC

Figure 2.2: Digital RoF for a C-RAN network.

This interface has been defined by initiatives such as the Open Base Station Architecture Initiative (OBSAI), CPRI and Open Radio equipment Interface (ORI).

• OBSAI specifies the architecture of the network, the function of a group of modules that have to fulfill a minimum of requirements in order to support one or more of the current standardized RAT, and it aims to guarantee compatibility and interoperability inside the network [15].

• CPRI is the standard most used by mobile operators, it defines the specifications for the interface between the BBU and the RU. It is a constant bit rate interface either for the downlink or the uplink [16].

• ORI was defined to develop a new standard based on CPRI. It aims to specify the inter-face between the BBU and RU and achieve interoperability between different vendors components [9].

The transmission of radio signals using D-RoF makes the communication reliable, the sampled signal is relatively immune to the nonlinear effects caused by the Electrical/Optical (E/O) and Optical/Electrical (O/E) conversions and also to noisy channels. Moreover this method also provides the implementation of low-cost optical transceivers. Therefore the implementation of D-RoF in a C-RAN system leads to a simpler CO [2].

D-RoF has inherently some associated problems, e.g. the requirement of bandpass sam-pling, and the problems brought by the ADC and DAC such as aliasing, quantization and jitter noise. With the implementation of this method in a C-RAN architecture other problems emerge. In this case the RU has to be more complex to perform the ADC and DAC opera-tions, which also leads to higher power consumption in the RU, so larger batteries have to be installed near the antenna. The implementation of new generations of communications, such 5G and beyond, will impose that a D-RoF link should support several Gbps signals since they introduce concepts such as carrier aggregation and Massive-Multiple Input Multiple Output (M-MIMO) which will lead to serious bandwidth problems in this link [2, 17].

(28)

2.2.2 Analog-Radio over Fiber

The alternative to D-RoF transmission is to send analog signals, originating the A-RoF method. Figure 2.3 shows a simple A-RoF scheme for a C-RAN network. With regard to the D-RoF scheme, the ADC and DAC are removed from the RU and passed to the BBU. Taking again the downlink case as an example, the digital signal in the BBU is converted to analog domain and then filtered to directly modulate the optical laser. This signal travels through the fiber and then in the RU it is simply detected by a photodiode, amplified, filtered and sent to the antenna. The uplink signal is received in the RU by the antenna, amplified, filtered and then used to directly modulate the optical laser. After the fiber link the optical signal is received in the BBU and converted to the electric domain. It is then filtered and converted to digital in order to be processed.

Digital Signal Processing DAC ADC AMP AMP

Figure 2.3: Analog RoF for a C-RAN network.

Implementing A-RoF solves the bandwidth problems associated to D-RoF, since in the analog case the required bandwidth roughly matches that of the transmitted signal, i.e. the bandwidth expansion caused by D-RoF is avoided. An advantage that is immediately visible by inspecting Figure 2.3 is the simplicity of the RU in contrast to the one required for D-RoF. Once again it is possible to use low-cost transceivers where a Directly Modulated Laser (DML) can be used if the RF frequency is within its bandwidth. For higher frequencies External Modulated Laser (EML) is typically needed, however DML continue to be a possibility if the RF signal is first downconverted to an IF [2].

However, an A-RoF signal is more prone to suffer from physical layer impairments. Since a digital signal is basically composed of ones and zeros, the only consequence of noise/distortion in the signal, is a bad bit decision in the receiver. On the other hand, in an analog signal any noise degrades its quality. Fiber impairments such attenuation and nonlinearites are worse in this type of transmission [2, 18, 19]. The transceiver quality is also more important in this case, and therefore the conversion E/O, O/E needs to be performed with very few losses and linearly [20, 21].

In order to mitigate these impairments and make communications possible using A-RoF, several techniques of DPD are currently being studied [22, 23, 24]. Other possibility consist in having a hybrid system that implements A-RoF and D-RoF combined [18].

2.3

Radio Access Technologies

In the current radio generations the main services required are to establish Human to Human (H2H) connection, and the existent technologies are designed in order to meet this requirements. However, the evolution trend indicates that the connection will have to be Everything to Everything (X2X), i.e. besides satisfying the contact between humans there

(29)

With applications such as, the Internet of Things, smart houses and smart cities the demands to a traditional RAT is huge. It urges the need to develop a new generation of technologies such as 5G, which is able to address and overcome these evolving challenges. One of the main problems consists in the exponential traffic data growth: by 2020 it is projected that a user will demand a data rate from 10 to 100 times higher than the average user requested in 2010 [10]. This will raise the minimum requirements in a RAN, not only in terms of data throughput but also in terms of bandwidth capacity. Besides, emerging applications such as 3D gaming and augmented reality will require a reduction of latency by a factor of 5 to 10.

Until this moment the wireless part of a RAN was the main obstacle that mobile operators had to address, when the optical fronthaul was expected to have capacity to support any requirements. However, with these new requirements the amount of effort introduced in optical infrastructure will increase and create new challenges that need to be solved [10].

The most recent recommendations from the 3rd Generation Partnership Project (3GPP) to 5G provide another challenge to the optical fronthaul, mainly because of the strict imposed limits for reliability (> 99.9999 %) and latency (1 ms). There are some challenges regarding the signal quality, recommended to be measured through Error Vector Magnitude (EVM). This value should be calculated using the following equation and have to meet the limits shown in Table 2.1. EV M (%) = s Perror Psignal × 100, (2.1)

Modulation Format EVM limit (%)

QPSK 17.5

16QAM 12.5

64QAM 8

256QAM 3.5

Table 2.1: 3GPP limits for each modulation format

Another challenge inherent to any generation advance, is that the new infrastructures not only has to support the newer technology standards, but also has to coexist with the already existent ones. This means that the new network architectures will have to support and be able to combine/separate signals from different generations, namely 4G and 5G. One architecture that has shown potential to meet the 5G requirements is the Cloud-RAN, explained in Section 2.1.

2.4

Digital Pre-Distortion

In order to improve and obtain the desired performance of the transmission system, sig-nal compensation techniques can be applied. This compensation can be performed before transmission (pre-compensation) or after transmission (post-compensation). A system hav-ing both, pre- and post-compensation, is another possible solution.

As the name suggests, DPD is a technique of pre-compensation, in which the signal is deliberately distorted before being sent to the system. This pre-distortion needs to be the inverse of the one introduced by the system, so that the output signal is ideally not distorted.

(30)

It is needless to say that to introduce this controlled distortion, one must first infer the distortion that the system introduces. In order to apply DPD various techniques can be used, among which we will address the following:

1. Look-Up Table (LUT) 2. Memory Polynomials 3. Volterra Series

Choosing the best model is not an immediate or easy task. It will always be a compromise between complexity, robustness and time to extract the model. Moreover there are distortions that can be better identified using a specific technique.

2.4.1 Look-Up Table

The Look-Up Table (LUT) model consists in dividing the system response in a number of subparts and then obtain a different model for each one. As depicted in Figure 2.4, the entry of the LUT is selected according to the instantaneous phase and/or amplitude of the input signal. | . | . . . . . . Phase LUT Gain LUT y(n) | y(n) | ejϕ|y(n)| G|y(n)| z(n)

Figure 2.4: Example structure of a LUT.

Figure 2.4 shows one possible architecture for a LUT-based DPD model, where y is the signal to be pre-distorted and z is the signal after the model is applied. This structure implements two different LUTs for the same signal, one that is responsible to compensate its phase and another that compensates the amplitude. The LUT considered in this work is a memoryless model. Although it is possible to use LUTs with memory, in those cases it is better to use a more robust model like a memory polynomial or a Volterra Series.

2.4.2 Memory Polynomial

(31)

be expressed by the following equation, z(n) = K−1 X k=0 Q−1 X q=0 akq· y(n − q) · |y(n − q)|k, (2.2)

where y is the transmitted signal without pre-distortion, z the pre-distorted signal, K the nonlinear order and Q is the memory depth. One of the main problems with memory polyno-mials is extracting the model, which in this case consists in getting the coefficients akq. The

most common method used to obtain this values is the Least-Square Method [25].

In order to fine-tune the compromise between performance and complexity, it is important to find the best architecture for the memory polynomial, i.e. find the best values for Q and K. Considering the system under test, it is possible to have a coarse estimation for these values, which can be refined by performing a brute-force optimization, i.e. by experimentally testing different values of Q and K around the expected optimum ones. When increasing the memory depth and the nonlinear order, the number of coefficients to be determined increases significantly and the time for their calculations becomes to long to be sustainable.

Comparing to the first presented method, LUT, memory polynomials can best model a highly nonlinear system. Nevertheless, despite of its straightforward application, it is a model that requires and additional effort on its optimization.

2.4.3 Volterra Series

When having complex nonlinear systems, Volterra series is a powerful tool for high-performance DPD [26]. Although providing an improved accuracy, this model also comes with the cost of higher complexity and time of extraction of the coefficients.

The Volterra series model can be generally written as,

z(n) = h0+ P X p=1   M1−1 X m1=0 ... Mp−1 X mp=0 hp(m1, m2, ..., mp) · mp Y j=1 y(n − mj)  , (2.3)

where once again y and z represent the signal before and after DPD respectively, P the polynomial degree, M is a vector with memory values that can be different for each degree and h are the kernels of the Volterra series. It is the values of the various kernels, hp, that

have to be obtained when extracting the model.

It is easy to see that for high values of P and Mp the number of coefficients increases

rapidly. This number can be reduced by removing the redundant terms that appear because of the commutative property of the multiplication. Although the number of non redundant coefficients to be determined for a p kernel can be obtained using equation (2.4) this value can still be too high when increasing M and p.

Mp(Mp− 1)...(Mp− (p − 1)), (2.4)

The Volterra series is one of the most robust methods to extract the model of a given nonlinear system. However, as previously mentioned, the price to pay is the time of extraction of the model, which can easily become too high.

(32)
(33)

Chapter 3

Experimental Setup

This chapter comprehensively describes the experimental setup implemented to emulate a low-cost analog optical fronthaul. The procedure of converting a typical Small-Form Pluggable (SFP) digital transceiver into an analog transceiver suitable for A-RoF transmission will be described in detail. Some inherent problems brought by this transformation are presented as well as some crosstalk measures. Lastly, the calibration procedure of the VOA utilized to set the received optical power will be presented.

3.1

Experimental Setup

In order to emulate a real analog optical fronthaul we have implemented the experimental setup shown in Figure 3.1. The setup is composed by an Arbitary Waveform Generator (AWG) responsible for the generation of the RF baseband signal. This AWG has two differential output channels corresponding to the I and Q waveforms. This signal is then up-converted by the IQ mixer before modulating the analog optical transmitter. The optical signal then travels through a given length of Single Mode Fiber (SMF). Before the optical receiver, there is a Variable Optical Attenuator (VOA), which enables to control the power at the photodiode. The optical receiver performs the O/E conversion and this electrical signal is received in the Vector Signal Analyzer (VSA). The VSA down-converts the signal and demodulates it using standard compensation algorithms. A picture of the actual experimental setup in the optical communications laboratory of Instituto de Telecomunica¸c˜oes can be seen in Figure 3.2.

In Figure 3.1 there is another setup configuration composed of only RF components: the aforementioned AWG, IQ mixer, VSA, and with the introduction of a 20 dB RF attenuator. This last component is introduced in order to maintain similar RF power at the VSA input with this setup as with the case of having the optical fronthaul. Some preliminary tests with this setup showed that for frequencies < 3.3 GHz the received RF power in the VSA was about 20 dB lower than the power driving the laser, justifying the value of the RF attenuator.

This RF setup provides the possibility of measuring the performance that can be obtained without the optic link, which enables to quantify the exact penalty introduced by the optical fronthaul and guarantee that the RF equipment is not a limitation.

As shown in Figure 3.1 we are able to control the equipment using MATLAB, which enabled to perform extensive autonomous tests. It is worth noticing that the setup is able to characterize any physically compatible optical transceiver, therefore providing a generic testbed for A-RoF test and optimization. The MATLAB code developed can be consulted

(34)

AWG + IQ mixer Analog Optical Transmitter VOA Analog Optical Receiver VSA 20 dB RF attenuator SMF MATLAB

Figure 3.1: Schematic representation of the A-RoF transmission system implemented in the laboratory.

and used, with previous request, in [27].

Figure 3.2: Picture of the actual testbench for RoF transmission implemented in the optical communications laboratory of Instituto de Telecomunica¸c˜oes.

(35)

3.2

Optical Transceivers

We used low-cost optical transceivers with an approximate cost of 50AC each, which were bought as standard digital SFP transceivers and then modified in the laboratory in order to allow for analog transmission.

These commercial SFP transceivers include a Transmitter Optical Sub-Assembly (TOSA)/ Receiver Optical Sub-Assembly (ROSA) pair that is independent of the type of transmission performed. Our aim is to, keeping these components, modify the surrounding electronics enabling them to send and receive analog signals, respectively.

In order to perform the aforementioned conversion, we have initially designed a PCB capable of receiving an RF signal and use it to directly modulate the laser. We have also designed the receiver circuit whose take the signal resulting of the photodiode and then splits it in two RF outputs. These circuits are shown in Figure 3.3 and in Figure 3.4, the first corresponding to the laser and the second the photodiode one.

(36)

Figure 3.4: Schematic of the board used to drive the ROSA for A-RoF reception. The final result of this adaptation is visible in Figure 3.5. Performing this adaptation brings the advantage of being able to change the emitted laser power using the potentiometer P1. On the down side, this adaptation makes the transceiver physically incompatible with a standard SFP evaluation board.

(37)

After the successful transformation presented before, we proceed with the aim of modifying the transceiver for analog transmission while keeping the original package. Figure 3.6 and Figure 3.7 show the modifications for the emitter circuit and the receiver, respectively.

Figure 3.6: Adaptation of the transceiver laser circuit in order to perform A-RoF.

Figure 3.7: Adaptation of the transceiver photodiode circuit in order to perform A-RoF. The modifications here presented consisted manly in bypassing the digital part of the board and connecting directly the input data pins to the output ones. In the figures the applied modifications are represented in red. The remaining components of the transceiver were maintained doing their original function. With these small modifications it was possible to obtain an analog low-cost optical transceiver and keep it in the original package as can be seen in Figure 3.8.

We modified and then tested two different transceivers, supporting different operating wavelengths and RF cutoff frequencies. One transceiver transmits at an optical wavelength of 1310 nm and has a cutoff frequency of 2.5 GHz while the other has a wavelength of 1530 nm and a maximum RF frequency of 10 GHz. It is worth noticing that the values of frequency here presented were found in the datasheets of the transceivers. Given that the transceivers were physically altered, there can be some fluctuations in these values.

(38)

(a) (b)

Figure 3.8: Modified CWDM transceiver, in (a) there is the alterations performed, and in (b) the transceiver is already in the original package.

3.2.1 Crosstalk Measurements

Some initial tests showed that, after the performed modifications, the transceivers were prone to have crosstalk problems. With this in mind, we performed a simple test consisting in measure the received power with the laser on and off. With the laser off it is not expected to have any received power but due to crosstalk problems the signal appeared in the receiver. Figure 3.9 shows the setup used to measure these powers. When we have the laser on, the optical link is established (blue in the figure) and the received power correspond to the attenuation due to optical losses. For measuring the power with the laser off, the link is interrupted and the power that is measured in the VSA is all due to crosstalk problems.

Figure 3.9: Diagram of the setup used to measure the transceiver crosstalk.

We defined crosstalk as the ratio between the power received with the laser off and the one with laser on, described in the following equation,

∆P = PLaser Of f PLaser On

, (3.1)

(39)

signal power also increases. After a RF power of 0 dBm this increase is more accentuated.

Figure 3.10: Dependency of the crosstalk with the RF power for a carrier frequency of 2.5 GHz. Fixing the RF power in 0 dBm we measured again the value of ∆P sweeping the RF frequency between 1 GHz and 10 GHz with steps of 500 MHz. These results are presented in Figure 3.11, which shows a massive dependency between the crosstalk and the RF carrier frequency. For higher frequencies (after 7.5 GHz), the crosstalk was to high, if we take the example of 10 GHz, ∆P = 1, which means that the RF power received when the optical link is broken is the same that when there is the optical components.

The results of measured crosstalk render the joint utilization of the transmitter and re-ceiver in the transre-ceiver more challenging, running into the risk of degrading the overall transmission performance. To avoid this problem and to ensure that the transceiver crosstalk is avoided, in all of the remaining tests we have decided to use TOSA and ROSA from different SFPs packages. Although this choice implies some hardware inefficiency (one TOSA/ROSA is unutilized per SFP pair), it guarantees a crosstalk-free operation for the remaining of this thesis. It is worth emphasizing that, while the prevention and/or mitigation of crosstalk in the modified SFP transceivers is a challenging and interesting topic for future work, in this thesis we will focus our attention on the optimization of A-RoF performance in the absence of transceiver crosstalk, namely through the use of advanced DPD algorithms.

(40)

Figure 3.11: Dependency of the crosstalk with the RF frequency with a RF power of 0 dBm.

3.3

Variable Optical Attenuator

In order to adjust the optical power at the input of the receiver photodiode, in this work we utilize an electro-optic VOA, whose basic working principle is to respond with a variable optical attenuation to a given applied input voltage. Since two different wavelengths were used (1310 nm and 1530 nm), it was necessary to characterize the VOA for both wavelengths. This characterization was performed by connecting one end of the VOA to an optical laser with constant output power and the other end to an optical power meter. Using this setup, the attenuation is measured in the reference power meter for a wide range of voltage values, i.e. from minimum to maximum attenuation achieved by the device.

The results obtained from this characterization for each wavelength are shown in Fig-ure 3.12. It is important to verify that the region where the VOA will be used is linear. The maximum required attenuation for the VOA is defined by the minimum link loss of the transmission system, which corresponds to the case where the setup is mounted in optical Back-to-Back (B2B), while the minimum required attenuation corresponds to the scenario with longest optical fronthaul of 20 km of fiber between the laser and the photodiode. There-fore in order to determine if this VOA can be used these values have to be calculated. Since the transceivers have an emitted power of -3 dBm and the required optical power at the pho-todiode is -12 dBm, it become apparent that the value for maximum required attenuation is 9 dB. The fiber associated losses are approximately ∼ 20 × 0.2 = 4dB, resulting in a required attenuation of 5 dB in the VOA. Both these values are marked in Figure 3.12, which clearly shows that we will be operating in the linear region of the component.

(41)

Figure 3.12: Characterization of the electro-optic VOA and definition of a linear region of operation for received power tuning in the photodiode.

(42)
(43)

Chapter 4

Experimental Results with 4G

Signal

This chapter provides a comprehensive experimental assessment and optimization of A-RoF transmission utilizing the experimental setup and optical transceiver previously described in Chapter 3. As previously mentioned in Section 2.2.2, the optical transceiver is one of the most important components for A-RoF transmission, in order to ensure low signal degra-dation. Taking this aspect into account, this chapter includes the experimental assessment of two distinct low-cost transceivers: a fixed wavelength SFP operating at 1310 nm and a CWDM-compatible SFP operating at 1550 nm. After optimizing both transceivers, the best performing transceiver has been selected for a comprehensive experimental test using 4G-LTE signals, including the assessment of key system parameters, such as the impact of signal bandwidth and fronthaul distance in the overall transmission performance.

4.1

4G-LTE Signal Parameters

Since the goal of this chapter consists in emulating and optimizing a real optical fronthaul in a C-RAN architecture, it makes sense that the chosen radio signal is compatible with this scenario, so it was decided to use a downlink LTE signal with all the standarized channels, namely:

• Primary Synchronization Signal (P-SS) - Z-Chu • Secondary Synchronization Signal (S-SS) - BPSK • Physical Broadcast Channel (PBCH) - QPSK

• Physical Control Format Indicator Channel (PCFICH) - QPSK • Physical Hybrid-ARQ Indicator Channel (PHICH) - BPSK • Physical Downlink Shared Channel (PDSCH) - 64QAM

Associated to the LTE signal, two main characteristics had to be set before starting the characterization: the signal bandwidth and the modulation format of the data channel (PDSCH). Considering that the objective was to analyze the performance for the worst case,

(44)

we have chosen to use the maximum values for both parameters: 20 MHz bandwidth and 64QAM.

One example of a received constellation can be seen in Figure 4.1. Here it is visible the different modulation formats between the channels.

Figure 4.1: Example constellation of an LTE signal after transmission over an analog optical fronthaul.

4.2

Electrical B2B Analysis

As mentioned in Section 3.1, it is important to ensure that the performance limitation does not come from the RF equipment itself. Therefore, we start by measuring the electrical B2B performance using the experimental setup signaled in red in Figure 3.1. These results also allow to calculate the exact penalty of introducing the optical fronthaul later on.

These electrical B2B tests consisted in analyzing the EVM and the received RF power when changing the power and frequency of the RF signal. The range of tested frequencies was between 300 MHz and 8 GHz with steps of 100 MHz, while the RF power was varied from -10 dBm to 10 dBm with steps of 1 dB.

Figure 4.2 shows the measured attenuation of the RF setup. As expected, the attenuation remained quite constant for all the measurements, being always around 22 dB, corresponding to the 20 dB of the attenuator plus some additional losses in cables and connectors.

Unlike the measured results of attenuation, the EVM results brought unexpected results. These results are shown in Figure 4.3 and it can immediately be seen that there are two zones where the results show some unexpected penalty. When having frequencies around 3.5 GHz and RF powers between -10 dBm and -3 dBm the performance degraded substantially. After some investigation it was concluded that this was a problem in the IQ mixer that involved automatic changes of hardware inside the equipment. After discovering this problem we have decided to work around it and not address it in this thesis. The second zone of incoherent

(45)

Figure 4.2: Measured attenuation in the electrical B2B setup composed only of the RF measurement instrumentation (AWG, VSG and VSA).

problem of IQ imbalance between the channels connecting the AWG and the IQ mixer. It has been then confirmed that, with proper IQ balancing the EVM could be reduced in all of the search space down to approximately 0.8%.

Overall, despite some exceptional cases, the electrical B2B performance can be considered as being independent of the RF power. We can also safely conclude that the RF measurement equipment will not create any performance limitation on the A-RoF transmission system.

(46)

Figure 4.3: Measured EVM in the electrical B2B setup composed only of the RF measurement instrumentation (AWG, VSG and VSA).

4.3

Transceiver Comparison

The tests proceed by adding the A-RoF transceivers in an optical B2B scenario. As previously mentioned, both transceivers were tested separately in the same conditions using always the VOA to maintain a constant optical power of -12 dBm at the photodiode.

The first test was similar to the one performed in the previous section, with the goal of finding how the performance changes with the RF power and frequency. Here due to the optical transceivers, it is expected to observe a greater performance dependence on the signal power and frequency.

Similarly to the electrical B2B scenario, we start by analyzing the attenuation of the link and then the EVM for each transceiver. The attenuation results are plotted in Figure 4.4, whereas the EVM performance is shown in Figure 4.5.

From the analysis of Figure 4.4, it is possible to see that there is a dependence of the atten-uation with the RF frequency. Increasing the carrier frequency lead to higher attenatten-uation due to the bandwidth limitations for the transceiver. This corresponds to the expectable behavior of a typical transceiver with limited bandwidth. No dependence between the attenuation and the RF power is verified. For frequencies below 3.5 GHz the attenuation is roughly 20 dB. This also justified the value for the RF attenuator in the previous setup.

Figure 4.5 gives a general overview of each transceiver performance. For both of them it exists a dependence on the frequency and power to obtain the best EVM, i.e., increasing the frequency demands a high power to have the lowest EVM. We can also observe that the 1310 nm transceiver had a worst behavior for high frequencies, which is expected since consulting the datasheet of this transceiver we can see that it has a cutoff frequency around

(47)

(a) (b)

Figure 4.4: RF attenuation imposed by the optical fronthaul, (a) corresponds to the scenario where we used the analog 1310 nm transceiver and (b) corresponds to the 1530 nm transceiver.

(a) (b)

Figure 4.5: Measured EVM with the optical transceivers in an OB2B configuration, (a) corresponds to the scenario where we used the analog 1310 nm transceiver and (b) corresponds to the 1530 nm transceiver.

To gain insight into the performance of the transceivers, we analyzed the best EVM that could be obtained for each frequency shown in Figure 4.6. Readers should note that the values here do not take in account the RF power, i.e., we show the lower EVM obtained from all the RF powers tested for each frequency. In this figure it is also shown the comparison with the electrical B2B scenario.

Analyzing the results it is possible to see that up to 5 GHz both transceivers had similar performances, being pretty much matched up to 2 GHz, while for higher frequencies the 1530 nm seems to have a smoother behavior. The main difference shows up for frequencies higher than 5 GHz, where the performance of the 1310 nm transceiver degrades rapidly. This behavior for the 1530 nm only appears after 7 GHz. The large bandwidth of the SFP+ is then confirmed.

(48)

Figure 4.6: Best EVM measured for each transceiver (after RF power optimization) and comparison with the electrical B2B performance.

Both transceivers show a good performance, being possible to obtain EVM values lower than the limit established from the 3GPP for the 64QAM (8%).

The problem around 3.3 GHz that was analyzed for the electrical B2B scenario popped up again with the optical transceivers, which might mislead the analysis of the transceiver performance in this frequency region. In order to have a corrected representation of the penalty introduced by the optic transceivers, we have decided to calibrate out the EVM measured in electrical B2B using the following expression,

∆ EVM = q

EVM2optic− EVM2

RF, (4.1)

The ∆ EVM value is the value that would be measured if the RF equipment was ideal. We explain in Appendix A the deduction of this equation.

The results of ∆ EVM are then shown Figure 4.7. Although we may still observe a considerable degradation of performance for frequencies around 3.5 GHz, it is interesting to notice that for lower frequencies the EVM penalty of introducing the optical transceivers was no greater than 1.5%, which is a remarkable result for such low-cost transceivers.

Up to this point the EVM analysis has been performed for the best RF power, but in a real scenario there will always be inevitable power fluctuations, and therefore it is important for the transceivers to have tolerance to these deviations. Figure 4.8 shows, for both transceivers, how much can the RF power be deviated from its optimal point while maintaining an EVM penalty below 1%. Analyzing first the best RF power for each frequency, it is possible to see that, once again, the 1530 nm transceiver has a smoother behavior. In turn, the 1310 nm has a lot of ups and downs in its optimal RF power which is not desirable in scenarios were carrier aggregation is performed, there is large bandwidth signals or when RF frequency can change.

(49)

Figure 4.7: EVM penalty introduced by the optical transceivers.

provides a large power margin: for some cases the power can even be either of the extremes tested and the EVM penalty is still < 1%. It is for higher frequencies that this margin is considerable reduced. On the other hand the 1530 nm shows an approximately margin for all frequencies.

This last analysis concludes the comparison between the transceivers. With the exception of some punctual cases the 1530 nm showed to be better, sometimes not in terms of lower EVM but it has always the most steady and predictable behavior. This together with being Coarse Wavelength Division Multiplexing (CWDM) compatible, make the 1530 nm transceiver to be chosen to carry the remaining tests and will from now on be referred simply as transceiver.

(50)

(a)

(b)

Figure 4.8: RF power range over which the EVM is degraded less than 1% with regard to the minimum EVM, (a) corresponds to the analog 1310 nm transceiver and (b) corresponds to the 1530 nm transceiver.

4.3.1 Bandwidth Influence

As previously mentioned in Section 2.2, the signal bandwidth is a increasingly limiting factor for the design of the optical fronthaul. Due to the adoption of A-RoF transmission, this

(51)

theless, due to the inherent bandwidth limitations of the utilized transceivers, it is important to measure in term of EVM, how much the signal bandwidth affects the performance. Since the transceiver is DML, it is expected that the bandwidth has a higher influence than if an external modulator was used.

In order to measure the impact of signal bandwidth on the transceiver performance, we have applied the same methodology of the previous section but with LTE signals with bandwidths of 5 MHz, 10 MHz and 20 MHz. To that end, Figure 4.9 shows the best EVM over frequency (after optimizing the transmitted RF power) achieved with different signal bandwidths.

Figure 4.9: Best EVM measured for a LTE signal of 5, 10 and 20 MHz.

From the analysis of Figure 4.9 it is possible to see that the measured results are within the expected. The signal with smaller bandwidth (5 MHz) is the one that presents the lowest EVM. Comparing the 10 MHz and 20 MHz signals it is interesting to notice that for frequencies above 3.5 GHz the performance is roughly the same. From the obtained results we may conclude that, despite having some minor influence, the signal bandwidths allowed by 4G are not a key limiting factor for the performance of the adapted A-RoF transceivers. This allows also to conclude that the transceiver is still operating inside its work range and that it is not introducing nonlinear effects due to excessive bandwidth. It is also worth emphasizing the measured EVM is always well below the 3GPP limit for 64QAM transmission within the sub-6 GHz region, which clearly shows that A-RoF can be considered as a robust and much more bandwidth-efficient alternative to typical D-RoF deployments.

(52)

4.3.2 Impact of Fronthaul Length

As previously mentioned in Section 2.2, a critical issue associated with A-RoF transmis-sion is the impairments introduced by long distances of fiber, such as the dispertransmis-sion or the attenuation. Another aspect that can also be of great importance in this kind of transmission is the optical power at the photodiode: i) if it is too low, it affects the reliability of the O/E conversion; ii) if the power is too high, then the photodiode can introduce high nonlinearities in the system.

Both of the previous characteristics were tested simultaneously with a 20 MHz LTE signal. The EVM was measured when varying the optical received power between -15.5 dBm and -9 dBm with a step of 0.5 dB with no fiber, and then with 10 km and 20 km. Due to the fiber loss and the high insertion loss of the VOA introduced in Section 3.3, we need to use another VOA with lower insertion loss. This one was mechanical and the measurements could not be automatized.

Figure 4.10: Comparison between B2B, 10 km and 20 km EVM for different attenuations in the SMF link.

The results of this analysis are shown in Figure 4.10. Regarding the optical link we can observe that the performance is indeed slightly worse with long lengths of fiber, however the deterioration is almost negligible, being lower than 0.5%. The results also show that the best power to have at the photodiode is around -13 dBm, which is 1 dB higher that the value that was being used in all the tests before. Increasing the received optical power to values larger then -10.5 dBm results in a large performance deterioration. Nevertheless it is worth noticing that for powers between -15.5 dBm and -10.5 dBm the EVM measured is below the 3GPP limit no matter the length of the optic link, which provides some extra decibels of power budget for the A-RoF transmission system, at the cost of a small performance penalty.

(53)

Chapter 5

5G Digital Pre-Distortion

In this chapter, we study the transmission of 5G signals using the optimized setup pre-sented in the previous chapter. The significant bandwidth increase over legacy 4G-LTE signals brings new challenges that have to be overcome. To improve the transmission performance, we study the use of DPD.

This chapter is organized as follows. First, the 5G signal is described. Then, we explain the procedure to extract and apply the DPD model. Finally, the results obtained with both 100 MHz and 400 MHz 5G signals are presented.

5.1

Experimental Setup

After the characterization and optimization of a low-cost analog optical fronthaul to trans-mit a 20 MHz LTE signal, it was decided to subtrans-mit the setup to the challenges that new radio generations will bring. In this chapter, the system will be tested with a 5G signal, which suf-fers from larger performance degradation due to its larger bandwidth. This together with the fact that 256QAM modulation should be achievable, for which the 3GPP EVM requirements are more strict (3.5 %), demand the use of compensation techniques. It was decided to use a DPD model, namely a memory polynomial in order to mitigate the nonlinear effects and increase the performance.

Figure 5.1 shows the used setup capable of taking the signal received at the VSA and use it to calculate the coefficients akq of equation (2.2). Afterwards the signal is pre-distorted and

sent again through the fronthaul. The setup is pretty similar to the one in Figure 3.1, with the main difference being that now we need to establish a bridge between the VSA and the AWG.

(54)

AWG + IQ mixer Analog Optical Transmitter VOA Analog Optical Receiver VSA Training  DPD model Pre-Distort Waveform 5G signal 256-QAM SMF

Figure 5.1: Experimental fronthaul setup.

5.1.1 RF signal

Once again, we aim to test the setup with the most demanding scenario. Regarding to the modulation format, the most advanced option for 5G is using a 256QAM data channel. In terms of bandwidth, the maximum specified values are 100 MHz for Frequency Range 1 (FR1), and 400 MHz for Frequency Range 2 (FR2). Therefore, tests were performed for both bandwidth configurations.

Analogously to the LTE case, we have considered a 5G downlink signal with all the standarized channels, namely:

• Primary Synchronization Signal (PSS) - BPSK; • Secondary Synchronization Signal (SSS) - BPSK; • Physical Broadcast Channel (PBCH) - QPSK;

• PBCH Demodulation Reference Signal (DMRS) - QPSK; • Physical Data Shared Channel (PDSCH) - 256-QAM; • PDSCH DMRS - QPSK.

One example of a received signal constellation can be seen in Figure 5.2, which depicts all the different modulations formats corresponding to each channel.

(55)

Figure 5.2: Example of a received constellation for a 5G signal.

5.1.2 Procedure

As previously mentioned in Section 2.4, applying digital pre-distortion requires an initial procedure of extracting the system model. To do this, the 5G signal without pre-distortion is transmitted, and the model is extracted from the received signal. It is important in this stage to measure the performance in order to be able to compare it with the results with DPD.

After extracting the system model, the next step consists in applying it to the initial wave-form and send it again through the fronthaul. Finally, we measured the benefits of using the DPD model. The model was obtained with only one iteration, however this procedure could be repeated any number of times for an increased accuracy. Nevertheless, our preliminary tests revealed that the DPD performance tends to saturate already after the first iteration, and therefore, for the sake of simplicity and complexity, we have decided for a single-iteration approach.

5.2

Results with 100 MHz

For the 100 MHz signal the RF carrier was fixed at 3.5 GHz, within the frequency band that is most likely to be adopted in European countries [28]. Tests were divided in four parts, always with the goal to compare the results with and without DPD. These parts were:

• DPD models comparison; • Optical B2B analysis; • Analysis with 20 km SMF;

(56)

5.2.1 DPD models comparison

As previously mentioned in Section 2.4, choosing a DPD model is not an immediate task. Some preliminary tests had to be done in order to make this choice. The test consisted in taking several EVM measures at a fixed frequency of 3.5 GHz while sweeping the transmitted RF power between -2 dBm and 6 dBm, with and without DPD. The models tested were the ones enumerated in Section 2.4, namely:

• A memoryless LUT with 128 entries;

• A memory polynomial with K = 3 and Q = 6; • A Volterra Series with P = 3 and M = 6.

In order for the model extraction to be quicker the tests were done with a 64QAM 100 MHz 5G signal. The obtained results can be seen in Figure 5.3.

Figure 5.3: Measured EVM in B2B using different DPD models.

From the results analysis it is clear that the model presenting the best performance is the memory polynomial. Given its high-performance and low-complexity, we have chosen to proceed our DPD analysis based the memory polynomial model, which from this point onwards will be simply referred to as DPD model.

5.2.2 Optical B2B analysis

The first test consisted in analyzing the performance of the fronthaul in a simple scenario without optical fiber. Once again the VOA was used in order to set the received optical power

(57)

This test consisted in optimizing the RF power for the 256QAM 5G signal. These op-timization was done for various DPD models varying the values of K and Q described in equation (2.2).

The first model implemented model consisted only of a linear compensation with one sample of memory. After it was tested a DPD model without memory (Q = 1), while increasing the nonlinear order of the model (K) until obtaining the best performance. With the optimized value of K, we have then increased the memory of the model until finding its best value. The obtained results are in Figure 5.4.

Figure 5.4: Measured EVM in B2B for the cases of no DPD, linear DPD and different nonlinear DPD based on memory polynomials.

As can be seen without DPD the best EVM obtained is 4.3% and it is obtained for a RF power of 0 dBm. Note that this value is above the pretended 3.5% limit established by 3GPP. The linear compensation corresponding to the blue dashed line in the figure presented similar results to those obtained without DPD. With all the nonlinear models implemented it was possible to bring the measured EVM below the 3GPP limit for 256QAM. In this case the best DPD model was obtained with Q = 1 and K = 5, yielding an EVM of 3.2%. This model shows also to be better in terms of RF power margin where the EVM is below the 3.5%, providing approximately a 2 dB tolerance for transmitted power detuning. Not shown here for brevity but increasing the nonlinear order for values higher than 5 resulted in a progressive loss of performance, likely triggered by a less accurate model extraction of higher-order nonlinearities.

It is interesting to observe that the best model was obtained for a higher power than the others (3 dBm), which is a well-known advantage of nonlinear compensation in general: by mitigating nonlinearities, higher powers can be launched into the transmission system, thereby resulting in an improved SNR and/or power-budget.

(58)

It is worth noticing that, all the considered DPD models in Figure 5.10 enable to success-fully transmit a 5G 256QAM signal with an EVM below the 3GPP limit.

5.2.3 20 km analysis

After achieving the transmission in a B2B scenario we have added 20 km of SMF in order to get a more realistic setup. With this setup there were two main goals. First, to verify if the model obtained in B2B is accurate enough to achieve the best transmission performance with a fiber fronthaul. This would bring a great advantage in practical terms, enabling to optimize DPD only for the optical transceivers in a controlled laboratory environment and then directly apply it to any deployed fronthaul network. The other main goal of this test, which is also inherently related with the first, is to test whether a long fronthaul link composed of 20 km SMF would introduce memory effects in the system.

With these goals in mind we have proceeded the tests using the best DPD model obtained in B2B (Q = 1, K = 5) and with Q = 2 and K = 5. For the memoryless model (Q = 1, K = 5), we have measured the performance applying only the model obtained in B2B and then extracting a new model from scratch, introducing the 20 km fiber in the system. It is worth noticing that the insertion of the fiber link required to increase the power in the photodiode to -10.5 dBm in order to achieve the best performance. The obtained results are in Figure 5.5.

Figure 5.5: Measured EVM after 20 km SMF for the cases of no DPD and different nonlinear DPD based on memory polynomials.

Without DPD the performance remains similar to the B2B case, with minimum EVM of 4.3% when the RF power is 0 dBm. Through the analysis of Figure 5.5 it is possible to conclude that the model obtained in B2B is still valid up to a RF power of 2 dBm, achieving

Referências

Documentos relacionados

foliar para a amoreira-preta. Embrapa Clima temperado, Pelotas, RS, 2012.. Embrapa Clima Temperado, Pelotas 2012.. Tabela 4A) Análise de variância para produção por planta em

- O nível de formação especializada dos educadores e a sua formação ao nível do desenvolvimento infantil e educação da 1ª infância – estas duas variáveis de uma forma

Foi abordado um método para o planeamento de trajetória no espaço de configurações, segundo o qual, se procede à interpolação de posições angulares de cada uma das

NÓS E OS OUTROS EM/COM QUEM HABITAMOS Na atual sociedade do conhecimento, pautada por encontros inesperados com as mudanças da era digital e cimentada por múltiplas incertezas,

„ Formalismo um pouco confuso pela recursividade Formalismo um pouco confuso pela recursividade.. Para cada grupo Para cada grupo é é escolhida a expressão que possui o menor

Uma vez que os tribunais do acidente médico são estabelecidos nos tribunais populares locais em vários níveis, e não existindo uma lei orgânica especial em vigor para os

O intercâmbio de informação sobre experiências e discussões de cariz científico sobre diversos problemas clínicos através da Internet, tendo como intervenientes os médicos,

FEDORA is a network that gathers European philanthropists of opera and ballet, while federating opera houses and festivals, foundations, their friends associations and