Analysis of the Performance of Different DWDM FilterTechnologies for Mobile Fronthaul Applications

IN

DEGREE PROJECT

ELECTRICAL ENGINEERING,

SECOND CYCLE, 30 CREDITS

,

STOCKHOLM SWEDEN 2016

Analysis of the Performance of

Different DWDM Filter

Technologies for Mobile Fronthaul

Applications

FREDRIK AHLBOM

Abstract

In recent years, several studies and simulations have been made on changing the current Radio Access Network (RAN) architecture into a more centralized access network where the base band processing is done in a central office (CO) instead of out by the antenna site. This new architecture is denoted as the mobile fronthaul and is planned to be in use for the coming 5G network. The studies that have been made so far suggest that the new ar-chitecture can reduce cost, power usage and latency which are important factors regarding environmental, economical and data transmission issues. Furthermore, the new architec-ture allows a smarter distribution of data for each sector covered by the antennas, reducing redundant data transmission and thus increases the data efficiency. The disadvantage or challenge however is that some of the optical components will be transferred from the cur-rently controlled environment in the CO to an uncontrolled outdoor environment at the antenna site, which may generate risks as these components may be sensitive to especially changes in temperature.

In this master thesis, the optical performance of four different passive filter setups, using a thin film filter (TFF), an arrayed waveguide grating (AWG) and an interleaver, has been studied and compared in order to find the most suitable filter setup for the mobile fron-thaul. These optical parameters include insertion loss, isolation, crosstalk, 3 dB passband, center wavelength drift and also bit error-rate (BER) which have all been measured over a temperature interval of -40-85◦C. Moreover, the measurement results have been compared with results from simulations done with VPItransmissionmaker.

From the measurement results, the TFF had a better optical performance and reliability compared to the AWG mainly due to a higher isolation and a lower BER penalty of 0.2 dB compared to 0.5-1.5 dB for the AWG. Considering data capacity and economical aspects for a more realistic mobile fronthaul scenario with 80 channels using dense wavelength division multiplexing (DWDM) however, the AWG connected to the interleaver is more beneficial without risking negative affects on traffic performance.

Sammanfattning

Under senare ˚ar har flera studier och simulationer utf¨orts med syfte att ¨andra arkitektu-ren p˚a dagens radioaccess-n¨atverk till ett mer centraliserat n¨atverk d¨ar basbandsprocesse-ringen sker i en central nod ist¨allet f¨or ute vid antennen och radiomasterna. Denna nya arkitektur kallas mobile fronthaul och planeras att realiseras till 5G-n¨atet. De studier som har gjorts hittills indikerar p˚a att den nya arkitekturen kan minska ekonomiska kostnader, elanv¨andningen och latens vilka ¨ar viktiga faktorer som bland annat r¨or milj¨o-, ekonomi-och kapacitetrelaterade omr˚aden. Dessutom kan data f¨ordelas p˚a ett smartare s¨att ¨over alla delomr˚aden som antennerna t¨acker vilket minskar redundant datatrafik och d¨armed ¨okar den effektiva m¨angden data som skickas ut. Problemet eller utmaningen ¨ar att vissa optis-ka komponenter beh¨over flyttas fr˚an en nuvarande kontrollerad milj¨o till en okontrollerad utemilj¨o vid radiomasterna vilket kan medf¨ora risker d˚a dessa komponenter fr¨amst kan vara v¨aldigt temperaturk¨ansliga.

Inom detta examensarbete har optisk prestanda studerats, analyserats och j¨amf¨orts mel-lan fyra olika filterkonstellationer best˚aende av ett tunnfilmsfilter, ett AWG-filter och en interleaver med syfte att finna vilken konstellation som passar b¨ast f¨or mobile fronthaul-arkitekturen. De optiska parametrarna best˚ar av insertionsf¨orluster, isolation, ¨overh¨ ornings-interferens, 3 dB-passband, centerv˚agl¨angdsdrift samt bitfelsgrad vilka alla har blivit un-ders¨okta ¨over ett temperaturintervall p˚a -40-85 ◦C. Ut¨over detta s˚a har m¨atresultaten j¨amf¨orts med simulationer gjorda med VPItransmissionmaker.

Utifr˚an m¨atresultaten kunde det konstateras att tunnfilmsfiltret hade b¨attre optiska egen-skaper och ¨aven h¨ogre trov¨ardighet j¨amf¨ort med AWG-filtret fr¨amst p˚a grund av en h¨ogre isolation och l¨agre bitfelsgradsstraff p˚a 0.2 dB j¨amf¨ort med 0.5-1.5 dB f¨or AWG-filtret. Om en endast avv¨ager datakapacitet och ekonomiska aspekter f¨or ett mer realistiskt scenario f¨or mobile fronthaul med 80 DWDM-kanaler s˚a ¨ar AWG-filtret tillsammans med interleavern mer f¨ordelaktig att v¨alja utan att riskera n˚agra negativa p˚averkningar p˚a trafikprestandan.

Acknowledgment

I would first of all like to thank both my supervisor at Infinera, Einar In de Betou, and my supervisor at KTH, Richard Schatz, for their great enthusiasm and supporting knowledge in all areas of the thesis. You have throughout the thesis given me quick and helpful responses when I have been in need of it and you have motivated me to work even harder. I would also like to thank my examiner, Sergei Popov, for all the help with the administration of the thesis. Lastly, but not least, I would like to thank Magnus Olson and Infinera for allowing me to do my master thesis, and thus finish my studies, at Infinera. I’ve found the thesis very interesting and joyful and it has further increased my interest for fiber optics and ICT. Moreover, I must also mention how grateful I am for all the free coffee and ”Fredags-bulle” which has supported and motivated me throughout the semester. The thesis just wouldn’t have been the same without it.

This master thesis has been administrated together with KTH, the department of Materials and Nanophysics, with Richard Schatz as my supervisor and Sergei Popov as my examiner. The thesis is a course worth 30 ECTS credits, corresponding to 800 hours of work in 20 weeks according to the course plan [1]. The project has taken place at the premises of Infinera, where results of measurement data and simulations have been collected and analyzed with the help of my Infinera supervisor Einar In de Betou. Infinera together with KTH, through the European GRIFFON project gr.342391, act as sponsors for this thesis. This study may be considered as a research project, as the purpose is to deliver a report examining the feasibility for usage of these filters in the mobile fronthaul. The project has been planned to be performed during a period of 20 weeks, spanning from January 25th 2016 to June 17th 2016.

Contents

3.1.1 Optical network overview . . . 5

3.1.2 Network topologies . . . 5

3.1.3 The new mobile fronthaul architecture . . . 7

3.2 Transmission characteristics of filters . . . 8

3.2.1 Insertion loss and isolation . . . 9

3.2.2 3 dB passband . . . 9

3.2.3 Center wavelength drift . . . 10

3.2.4 Crosstalk . . . 10 3.3 Receivers . . . 11 3.3.1 Bit-Error Rate . . . 11 3.3.2 FEC . . . 12 3.4 Filters . . . 13 3.4.1 AWG . . . 13 3.4.2 TFF . . . 15 3.4.3 Interleaver . . . 16

4 Methods and measurements 19 4.1 Filters . . . 19 4.1.1 TFF . . . 19 4.1.2 AWG . . . 19 4.1.3 Interleaver . . . 19 4.2 Lab setups . . . 20 4.2.1 Filter characterization . . . 21 4.2.2 BER measurements . . . 24 4.2.3 Temperature control . . . 26 4.3 Simulations . . . 28

5 Results and analysis 30

5.1 Filter characterization . . . 30

5.1.1 Insertion loss and isolation . . . 30

5.1.2 Crosstalk . . . 39

5.1.3 Filter functions . . . 40

5.1.4 3 dB Passband . . . 42

5.1.5 Center wavelength drift . . . 44

5.2 BER measurements . . . 46

5.2.1 Crosstalk from adjacent channels . . . 46

5.2.2 Controlling setup with a BERT . . . 47

5.2.3 BER . . . 47 5.2.4 Simulation results . . . 54 6 Discussion 57 6.1 Technical aspects . . . 57 6.1.1 Simulations . . . 59 6.1.2 Power budget . . . 60 6.2 Measurement errors . . . 60 6.3 Ethical aspects . . . 62 6.4 Implementations . . . 62

7 Conclusions and future work 64 7.1 Conclusions . . . 64

1

Terminology

APD - Avalanche Photo Diode

ASE - Amplified Spontaneous Emission AWG - Arrayed Waveguide Grating BBU - Base Band Unit

BER(T) - Bit Error Rate (Tester) CDMA - Code Division Multiple Access CO - Central Office

CPRI - Common Public Radio Interface C-RAN - Cloud-Radio Access Network CW - Continuous Wave

DAS - Distributed Antenna System DFB - Distributed Feedback Laser D-Rof - Digital-Radio over Fiber

DWDM - Dense Wavelength Division Multiplexing FEC - Forward Error Correction

FPR - Free Propagation Region FSR - Free Spectral Range

FWHM - Full Width Half Maximum

GSM - Global System for Mobile communication GTE - Gires-Tournois Etalon

ITU-T - International Telecommunication Union-Standardization Sector LAN - Local Area Network

MAN - Metropolitan Access Network NRZ - Non Return to Zero

OADM - Optical Add-Drop Multiplexer PAM - Pulse Amplitude Modulation PDL - Polarization Dependent Losses PLC - Planar Lightwave Circuit PMD - Polarization Mode Dispersion PRBS - PseudoRandom Binary Sequence QAM - Quadrature Amplitude Modulation RAN - Radio Access Network

RRH - Remote Radio Head SMF - Single Mode Fiber

TDM - Time Division Multiplexing TFF - Thin Film Filter

VOA - Variable Optical Attenuator WAN - Wide Access Network

2

Introduction

2.1

Background

The technology used for communication between individuals and devices has changed dras-tically during the last couple of decades. Today, the demand for data capacity is record high and still increasing due to several factors. More people are getting connected as technology constantly improves and many third world countries are catching up. Not only the number of users are increasing, but also the data usage per person. Operators are constantly look-ing for improvements to meet the market’s growlook-ing demands. Recent studies [2]-[5] suggest replacing the current Radio Access Network (RAN) where each station has its own Base Band Unit (BBU) connected to the Remote Radio Head (RRH) via a coaxial cable with a new convention called the mobile fronthaul. This new architecture will replace several base band stations with a centralized Cloud-Radio Access Network (C-RAN), placing the BBUs at a Central Office (CO) that are connected to the antennas via fiber instead. The benefits of this new architecture is firstly a reduction of cost and power usage as the traffic across fiber doesn’t require the same power to run as with a coaxial cable. Secondly, replacing the coaxial cable with fiber also reduces the latency which enables a better transmission per-formance. Lastly, the C-RAN architecture provides an opportunity to control and manage output power from the antennas in a smarter way using a concept known as Distributed Antenna Systems (DAS) that may also reduce unnecessary power usage and cost.

One of the disadvantages or challenges with the C-RAN architecture however is that the passive optical filters have to be transferred from the current controlled environment in the CO to the antenna in uncontrolled outdoor conditions[6]. A large change in temperature may put a lot of stress on the relatively sensitive passive filters which may affect their performance negatively. One such problem that may appear is a center wavelength drift of the filter’s transfer function, leading to more crosstalk and a higher Bit Error Rate (BER) which eventually could cause a traffic shutdown. Thus, before implementing the C-RAN architecture, several parameters must be studied, leading in to the scope and purpose of this master thesis. The objectives of this master thesis will be to examine and compare four different filter technologies regarding usage in the new mobile fronthaul architecture. These filters will be tested for BER, insertion loss, isolation, crosstalk, 3-dB passband width and center wavelength drift over a temperature interval using lab equipment provided by Infinera. Finally, an evaluation and a comparison will be done considering technical performance, reliability and complexity for application of these different filter technologies.

2.2

Purpose

The purpose of this master thesis is to find the most suitable dense wavelength division multiplexing (DWDM) filter technology for usage in a new fronthaul architecture for the fiber-optic network. The study will include four different filter setups from two different vendors, denoted A and B, provided by Infinera which are to be evaluated and compared regarding measurement data on several different parameters, specification data and usage of the filter technologies. The main focus will be to test the filters optical performance for a wider temperature range.

2.3

Scope

The following points describe the scope of the master thesis. These are described in more detail in section 4 Methods and measurements.

• What DWDM filter technology is most suitable for use in extended temperature con-ditions?

• What center wavelength drift over temperature is expected from different filter tech-nologies?

• How are other filter parameters affected by operation in extended temperature? • How is the reliability of the filter affected by operation in extended temperature? • What is the impact of DWDM signal transmission from operation in extended

tem-perature?

• How is BER affected by temperature and by adjacent channels for each of the filter setups?

• While 10 Gb/s line rates and a 100 GHz DWDM grid may seem enough to meet capacity requirements in the short term (40 x 10 Gb/s = 400 Gb/s per fiber), scalability to a denser 50 GHz grid and 100 Gb/s wavelengths may be needed in the future (80 x 100 Gb/s = 8 Tb/s per fiber). This should be considered in this study.

2.4

Limitations

The following table describes the limitations of this thesis. These limitations may however be discussed based on simple assumptions.

• A cost analysis of the different filters will not be regarded in this study, but may be commented on. The main objective is instead to analyze the optical parameters and draw conclusions from these.

• As the filters only will be tested in a lab environment, the study will not include the physical installation and operation in the actual mobile fronthaul.

• The study will only concern signals at frequencies within the DWDM band.

• The study will not concern attenuation or polarization dependent losses (PDL) or polarization mode dispersion (PMD) in fiber.

• The study will not concern different modulation formats of the traffic signals for BER measurements.

• Power usage and efficiency will not be taken into consideration in the study.

2.5

Prior art

The term ”mobile fronthaul” is a relatively new expression within telecom and is just about to be implemented for the coming 5G-network. Thus, before jumping into this study, it is relevant to present some studies that have been done before within the same field.

In a paper by N. Carapellese, et al. [7], several different aggregation infrastructures, such as current systems and the new C-RAN, have been tested for power usage with the goal to find how much power can be saved with the C-RAN architecture. In the study, they concluded that the fronthaul based solutions was always better than the current backhaul systems with approximately 40-50% of savings in energy, claiming that ”the enhanced flexibility of C-RAN promises not only significant energy consumption savings, but also improved radio perfor-mance, reduced capital and operational expenditures, footprint and installation/intervention times. To fully exploit such a potential, a drastic re-engineering of the access infrastructure is needed, which will be one key of future fixed/mobile converged networks.” [7].

An invited paper by Thomas Pfeiffer [8] discusses the employment of Common Public Radio Interface (CPRI) transmission for certain setups of multi-sector and multi-antenna config-urations, which resulted in a theoretical bit rate of 148 Gb/s. The author also discusses

some potential technologies for mobile fronthaul and the future 5G network, advocating for point-to-point DWDM-based systems. Also the economical and capacity-related benefits are highlighted in the study.

3

Theory

3.1

The optical network and the mobile fronthaul

3.1.1 Optical network overview

Looking back only a few years in our history, the so-called public network was almost completely operated by telephone companies. These operators or carriers as they are also called have changed focus drastically with the technology enhancement that has been going on in recent years and are today providing a large variety of services, not only related to telephone services. This public network mainly consists of central offices (CO) that are also called nodes which are connected through fiber links. The whole network is usually divided into three different areas, one spanning at short distances between buildings and devices called a local-area network (LAN) or an access network, one spanning at a bigger area or within a metropolitan area, typically a size of ten or hundred km, called a metropolitan-area network (MAN) or an interoffice, and one that spans from hundreds to thousands of kilometers, usually sub-sea systems and longer connections between cities and countries, called a wide-area network (WAN) or interchange network. Figure 1 demonstrates the whole network system. For this thesis, regarding the mobile fronthaul, focus will typically be on the access network. [9]

Figure 1: An overview of the architecture of the network system that is in use today.

3.1.2 Network topologies

Looking further into a network, it is often required to be able to distribute information across several subscribers, for example among a group of houses or in between computers at an office, rather than just point-to-point communication. When designing the access network, one can use several different topologies where the most common are shown in figure 2. For the bus topology, the channels are carried via a single fiber and are dropped along the way using optical taps or optical add-drop multiplexers (OADMs) where a fraction of the light diverts from the fiber. This introduces a disadvantage however as the signal is attenuated after each tap. Along the fiber line, the losses increase exponentially which limits how many subscribers that can be reached. With a transmitted power PT, the power PN of the N th

PN = PTC[(1 − δ)(1 − C)]N −1 (1)

where C is the fraction of the power transmitted further to each node and δ is the insertion loss at each tap. For this example, C and δ are assumed to be the same for each tap. For the hub topology, the distribution of channels is acquired at a certain location called ”hub” which is done in the electrical domain before transmitting the signal further over fiber. This topology is typically used in bigger cities connecting several nodes and offices together. Both the bus and hub topologies are usually used for larger networks such as in MANs. A disadvantage with them however is that if there is a fiber outage in the link, a large number of the nodes may get affected. [23]

(a) Bus. (b) Hub.

(c) Ring. (d) Star.

Figure 2: Depicting networks with four different topologies.[23]

For shorter distances and smaller networks, typically within a LAN, ring and star topologies are more common. For the ring topology, several nodes are connected via point-to-point fiber in a closed route, where each node receives and transmits data further which then can work as a repeater. [10] For a star topology, all nodes are connected to a central hub via point-to-point fiber. This center hub can be either a passive or active unit. For an active hub, the optical signals are first converted to electrical signals before being processed, while for the passive hub, the distribution amongst the nodes is done optically via couplers. As for the case with the bus topology, the number of subscribers connected to the star will be limited by the distribution losses. Assuming an ideal N × N coupler, each node will receive a power of PT/N from a signal with transmitted power PT. By introducing insertion losses δ to the

nodes, the power received at each node, PN, can be written as

Here, the log2N comes from the interconnections between the 3 dB couplers that are needed

to achieve an N × N coupler. [23]

3.1.3 The new mobile fronthaul architecture

All of the common communication standards such as 2G, 3G, 4G, Wi-Fi, GSM and CDMA amongst others are operating by using a Radio Access Network (RAN) consisting of a cell configuration that controls power, cooling, voltage converting and information transmission within a base station (BS)[3]. Each BS is located by the antenna which is fed through a coaxial cable. This network is illustrated in figure 3a. The current RAN that is in use today is however bulky, complex, ineffective and expensive. Each BS is only connected to a fixed amount of sector antennas that cover relatively small areas and can only handle the traffic within that area. Also, these BS are built on large platforms that may require site rental, site support and a possibly problematic management support. Also, as these BSs are constructed to be able to support traffic at its peak at any time, they have a very low utilization rate as traffic may peak only a few times per day and power cannot be shared amongst other BSs [11]. The current RAN will thus be insufficient for the upcoming 5G system. However, installing the new C-RAN architecture is looking promising and aims to enable support for the future demands with a better control, higher data rate, more efficient power usage and to a lower price [12].

The main idea of changing the design of the current RAN by introducing a mobile fronthaul is to centralize the base band processing among several Remote Radio Heads (RRH). Instead of having all BBUs placed at each BS, they will instead operate at a Central Office (CO) or a concentration point. Each CO will then reach out to multiple BS using the concept of Digital Radio over Fiber (D-RoF), which corresponds to the new convention ”mobile fronthaul”. Instead of connecting the CO with a coaxial cable to the antenna device, the new transport segment will consist of fibers connecting the BBU with the RRH. The RRH will in addition be connected to the antenna through a coaxial jump [2]. Figure 3a shows the current model of a RAN and figure 3b shows the model of the new C-RAN.

(a) The current model of a RAN with the BBU placed by each radio unit in connection with the RRH through a coaxial cable [2].

(b) The model of the new C-RAN with the BBUs placed at a centralized point connected to the RRH through fiber [2].

Figure 3: An illustration of (a) the current RAN-model today and (b) the new C-RAN-model.

The following list shows the possible benefits of using a C-RAN compared to the current model [4][13]:

• Replacing the coaxial cable with fiber reduces the power consumption by firstly elimi-nating the electrical losses in the coaxial cable but also by moving the power amplifier from the base station to the RRH.

• With a DAS, it is possible to drive several antenna systems from one central office. The transmitted power can thus be adjusted to spread to the antennas covering the areas with the highest traffic at that moment, reducing power that is wasted on areas that might not be in use at that specific moment. This will thus improve the efficiency. An example of a DAS is shown in figure 4.

• The replacing of the coaxial cable with fiber is also a matter of data transmission speed. A fiber can carry an optical signal of the order Tb/s while the coaxial cable only can support a transmission speed of an order of Mb/s, resulting in a 106_-order

difference.

• With a centralized system, the C-RAN will be much easier to install and maintain as the BBUs for several antenna systems will be located at a central office compared to having the BBUs at each antenna. Thus, both time and money can be saved in the installation and maintenance process.

Figure 4: Using the concept of DAS, multiple antenna systems can be controlled from one base band unit [4].

The challenge with the C-RAN architecture however is, as mentioned in the introduction, that the mux and demuxers, i.e. the filters, will be placed next to the RRH close to the antenna, exposing them to possibly rougher climates, where mostly thermal changes may affect the performance of the rather sensitive filters. Thus, it is of interest to characterize these filters.

Moreover, the idea of the new architecture is to place the antennas a maximum of 20 km away from the CO. With such a short distance, the signals are much unlikely to suffer from dB-penalties from dispersion, polarization losses or optical signal to noise ratio (OSNR)-related penalties, which thus needn’t be neither considered in the study nor discussed for future work.

3.2

Transmission characteristics of filters

In the following section, a couple of relevant filter characteristics or parameters are defined and explained. Throughout the study and the report, the signals are described in frequencies, f rather than wavelengths, λ. Also, to simplify calculations, the power levels are measured

in a 10-based logarithmic scale with the unit dB. When analyzing optical systems, it is also of interest to study the power level related to 1 mW of power, where the unit then is denoted dBm. Adding to this, a system that sends information separated at different frequencies is called a Wavelength Division Multiplexing (WDM) system, compared to the Time Division Multiplexing (TDM) system.

3.2.1 Insertion loss and isolation

When a WDM-signal is transmitted through a filter or any other type of device, the signal will experience an attenuation or power loss, also denoted channel insertion loss, IL, which is defined as

IL = −10log(Pout Pin

) (3)

Where Pinand Poutis the input and output power of the device respectively. This is defined

for the biggest loss within the channel frequency range, defined as a nominal frequency fnom, in this case according to the (International Telecommunication Union-Standardization

Sector) ITU-T-grid, and an additional ±12.5 GHz range in which the channel is required to operate in. An illustration of this is shown in figure 5. This frequency range is also called a passband and is defined to be 25 GHz wide for a 100 GHz channel spacing.

Figure 5: Insertion loss for a DWDM signal transmitted through a filter. The insertion loss is always the lowest power within the channel frequency range or passband, fnom±12.5 GHz.

The blue curve illustrates the DWDM signal and the orange lines mark the channel frequency range. Here, the y-axis is normalised to the input power and the x-axis is normalised to fnom.

When analyzing a WDM system of several channels, especially DWDM, it is important to be able to distinguish channels from each other. One parameter that describes this is the isolation between channels which is the difference between the lowest power within the channel frequency range and the highest power from the adjacent channel frequency range, measured in dB. Figure 6 illustrates this. [21]

3.2.2 3 dB passband

The 3 dB passband width, W3dB, can be defined as the frequency range where the power

has dropped half of its power, or 3 dB, compared to the power value at the nominal center frequency, usually according the ITU-T grid. The width is also symmetrical around the

Figure 6: Adjacent isolation for a DWDM signal transmitted through a filter. The blue curve illustrates the DWDM signal, the orange lines mark the channel frequency range and the grey line marks the outer frequency limit for an adjacent channel that is 100 GHz apart in this case.

center frequency, i.e. the 3 dB passband is whithin the range of fnom± W3dB/2. Hence,

the power may have dropped 3 dB on one side of the passband but only 2 dB on the other side when comparing the same drift from the center wavelength. An illustration of this is shown in figure 7. [21]

Figure 7: The 3 dB passband of a channel shown according to the definition. [21]

3.2.3 Center wavelength drift

The center wavelength for a channel can be defined as the average of all wavelengths within a 3 dB drop from the peak value of a channel. [17] The difference between the center wave-length of the channel and the channel wavewave-length defined from the ITU-T grid is called the center wavelength offset. When the temperature of the filter changes, the center wavelength of the filter channels may drift leading to a change in the offset from the ITU-T grid. The drift is measured in GHz for this thesis but is usually also measured in nm.

3.2.4 Crosstalk

When multiplexing or demultiplexing a DWDM signal using a filter, the goal is to transmit as much power as possible from the wanted channel and attenuate all others. However, due to imperfections in the filter, the filter may leak power from the adjacent and non-adjacent channels which then interferes with the primary channel. This is called crosstalk and can

be defined as the ratio of the total power of the primary signal to the total power of the interfering signals, computed as

X = Rf0 −f0X(f )df Rf0 −f0S(f )df (4) where X(f ) is the interfering signal, S(f ) is the primary signal and f0is a frequency range

that covers the bandwidth of the signal.[22] Crosstalk is a big concern for DWDM systems as a higher crosstalk may increase the BER and thus limits the transmission distance. Figure 8 shows a simulation made in VPI of the transmission power of 16 channels through an AWG filter. As can be seen, the filter is far from perfect in isolating adjacent channels which all interfere with each other.

Figure 8: 16 AWG channels with a channel spacing of 100 GHz, simulated in VPI.

3.3

Receivers

3.3.1 Bit-Error Rate

When an optical signal is received at the receiver end, the signal is sampled at a decision time tD with a certain sampling frequency. One way to measure the performance of an

optical receiver is to look at the bit-error rate (BER) which is defined as the ratio of every wrong bit over every received bit, or as the probability to detect an incorrect bit. A common criterion for optical receivers is a BER below 10−9, i.e maximum one bit per billion may be incorrect. Since the probability for a 0 and a 1 to occur are equal, the BER can be defined using both the probabilities of detecting a 1 when a 0 was received and detecting a 0 when a 1 was received as [23]

BER = 1

2P (0/1) + P (1/0) (5) These probabilities will depend on a probability function p(I) of the sampled values I. I1

represents the average signal level of a 1-bit and I0represents the average signal level of a

0-bit. As the incoming signal will be affected by several kinds of noise, which can be described with Gaussian statistics, the signal will fluctuate around I1and I0, each with corresponding

variances σ1and σ0. Using these values, one can define the probabilities P (0/1) and P (1/0)

P (0/1) = 1 σ1 √ 2π Z ID −∞ exp  −(I − I1) 2 2σ2 1  dI = 1 2erfc  (I1− ID) σ1 √ 2  (6) P (1/0) = 1 σ0 √ 2π Z ∞ ID exp  −(I − I0) 2 2σ2 0  dI = 1 2erfc  (ID− I0) σ0 √ 2  (7) which are illustrated in figure 9. Here, erfc is the complementary error function which is defined as [27] erf c(x) = √2 π Z ∞ x exp(−y2)dy (8)

Figure 9: The fluctuations at the I1 and I0 levels can be described with two Gaussian

distributed functions with corresponding variances σ1and σ0from which P (0/1) and P (1/0)

can be calculated.

Once calculating the BER based on measurements, one must make sure to measure enough bits to ensure that the BER values are within some level of confidence. This can be done with the following formula for the confidence level CL

CL = 1 − e−N ×BER× E X k=0 (N × BERS)k k! (9)

where N is the number of transmitted bits, E is the number of incorrect bits and BERS is

the specified or required BER. Theoretically, CL will always be less than 100 % since one cannot measure for an infinite length of time, thus one must choose a reasonable confidence level. Many industries consider at least 95 % as a valid level, which has been decided to be used for this study as well. [24]

3.3.2 FEC

One way of improving the performance of an optical transmission system is to employ a forward error correction (FEC), which allows a system to tolerate a higher BER before the receiver reliability is reduced too much. [14] A FEC method that is very common is the Reed-Solomon (RS) FEC. It is based on adding redundant bits to the data words that are to be transmitted. Consider an amount of data with n codewords which are n symbols long. The data can be divided into block codes that are defined by k number of symbols of real data and an additional r numbers of redundant symbols (also called parity symbols) corresponding to the redundant data. By knowing the position and values of these symbols,

one can detect if they are wrong at the receiver end which then can be used to recover incorrect real data. Figure 10 describes how the redundant data rk− rn−1 can be encoded

to a stream of information d0− dk −1. [25]

Figure 10: A data stream being encoded with redundant bits.[25]

The disadvantage of the FEC method is that it requires an increased bit rate to transmit the same amount of information as without the FEC, which will lead to a need for larger bandwidth and perhaps wider channel separation which introduces a linear penalty. [26] But in total, a gain of several dB can be attained by the use of FEC and is thus beneficial for data transmission and detection and most of all to extend the power budget or reach in a network.

3.4

Filters

3.4.1 AWG

One of the most suitable filters for DWDM is the Arrayed Waveguide Grating (AWG). It is a passive and cheap filter with high reliability that can be manufactured in a large scale using a silica-based planar lightwave circuit (PLC) technology [15]. The principle of operation is based on separating the lightwave channels by phase shifting them. The incoming WDM signal, consisting of N channels with wavelengths λ1−λN, will first propagate through a free

propagation region (FPR) where the signal is spatially diverged and later coupled into an array of waveguides. Each waveguide will differ the length ∆L from its adjacent waveguide such that each waveguide length will differ a multiple of the center wavelength of the signal. The signal will then experience a phase shift in each waveguide which also has a dependence of the frequency [23]. This results in the channels being divided according to wavelength by leaving the output with a diffraction angle θn which will be different for every wavelength

λ1− λN. By using the concept of constructive and destructive interference at the image

plane, the different wavelengths can be separated to an array of N outputs carrying one channel each. The length difference ∆L between the center waveguide and waveguide n can be written as

∆L = nλc Ng

(10) where n is the order of channel, λc is the center wavelength of all channels and Ng is the

effective index of the waveguide mode. As the signal leaves the arrayed waveguides with a certain dispersion angle θn and propagates through a second FPR, it will focus on an

imaging plane where the channels are divided into different waveguides for transmission, i.e. output ports. This angle is very small and can be written and approximated as

θn= sin−1  (∆Φ − n2π) daβF P R  ≈ ∆Φ − n2π daβF P R (11) where ∆Φ = β∆L, β is the propagation constant of the waveguide mode, βF P R is the

as shown in figure 12 [16]. A benefit of the AWG is that it is reciprocal and may thus be used both for multiplexing and demultiplexing.

Figure 11: A schematic of an AWG [23].

Figure 12: A schematic of how each signal is focused at a single output port [16]. The aim with the filter is to achieve a transfer function with a low insertion loss, sharp skirts and small sidelobes, preferably close to a rectangular pass band function or a Super-Gaussian of high order to reduce crosstalk and thus the BER [28]. This is even more important for DWDM applications where the channel spacing usually is 100 GHz or lower which means that the channels are even more vulnerable to crosstalk. Assuming that the incoming signals have a Gaussian distribution exp(−y_2σ22) and that the total number of waveguides are 2P + 1,

where P is a positive integer, then the transfer function of the filter from the input waveguide to the central output waveguide can be expressed as

Hq(f ) = P

X

n=−P

Cnexp(j2πtnf − jκynyq of ) (12)

where yq o is the distance from the central output waveguide to the output waveguide n.

Since the AWG-filter is highly dependent on the phase shifts generated by the length dif-ference ∆L, it follows that the AWG is very sensitive to changes in temperature as it may implement small expansions or contractions in the material which may cause the channels to focus at the wrong output port, leading to higher losses, more crosstalk and in general a worse filter performance. To avoid this, one can either keep the temperature constant by installing it in a controlled environment, or one can use an athermal AWG that compensates for the material changes caused by thermal changes. This is generally done in two ways, either by applying a material with a negative _dTdn in the optical path, such as resin-filled grooves,[19] or by mechanically moving the input fiber with a metal board that will expand or contract with temperature so that it compensates for the focus shift [20]. An illustration of this is shown in figures 13 and 14.

Figure 13: A shift in focus on the output ports can be compensated by mechanically moving the output lens according to the change in temperature.[18]

Without a thermal compensation, a silica-based AWG can have a center wavelength drift of approximately 10 pm/◦C which can be severe for long transmissions at a very high or low temperature. But with a first order thermal compensation, the center wavelength drift can be reduced to only a couple of pm/◦C. A case with and without thermal compensation are shown in figure 14.

Figure 14: An overview of a conventional AWG where focus point output has a larger drift at higher and lower temperatures compared to an athermal AWG where a shift of the metal board compensates for the focus drift.

3.4.2 TFF

Figure 15: The principle of a Fabry-Perot filter consisting of two mirrors that separate lightwaves of different frequencies. [9].

The thin film filter (TFF) is a resonant multicavity filter that uses the same principle as a Fabry-Perot filter. The cavity consists of two highly reflective mirrors or films, parallel to each other, separated a distance l. This allows some wavelengths of light to propagate through the mirrors while some are reflected which is shown in figure 15. The light that will travel multiples of half their wavelength through the length of the cavity will always have the same phase when leaving the right mirror and are thus called resonant wavelengths [9].

The power transmitted through the filter can be expressed with the following function T (f ) = (1 − A 1−R) 2 1 + (2 √ R 1−Rsin(2πf τ ))2 (13)

where A is the absorption loss and R is the reflectivity for each mirror and τ is the propa-gation delay across one cavity length. Figure 16 shows the transfer function for a filter with A = 0 and R = 0.75,R = 0.90 and R = 0.99. One can see that a higher reflectivity leads to sharper filter functions with higher isolation.

Figure 16: A transfer function for a TFF with A = 0 and R = 0.75, R = 0.90 and R = 0.99 [9].

What also can be noted is that the transfer function is periodic in f and will peak for every f τ = k/2 for a positive integer k and may thus coincide with an adjacent channel. The spectral space between two peaks is called the free spectral range (FSR). To avoid crosstalk as much as possible, two different channels must be separated by at least one Full Width Half Maximum (FWHM) of the signals in addition to a spectral range distance of k × F SR for a positive integer k.

One of the benefits with a Fabry-Perot filter is that it can be tuned simply by changing the cavity length or the refractive index in the cavity. The filtered frequency is selected to satisfy

f0τ = k/2 (14)

for a positive integer k. And with

τ = ln

c (15)

where n is the refractive index, l is the distance between the mirrors and c is the speed of light. The filtering frequency f0 can be changed by either changing l or n.

When designing a TFF, several Fabry-Perot filters are put in cascade where each filter corresponds to one channel. The effect however is that for every filter that the signal passes through, the filter function will change, which can be seen in figure 18. After every filter, the transfer function becomes sharper at the skirts. The TFF is reciprocal like the AWG filter, meaning that it can be used both to multiplex and demultiplex channels. Figure 16 shows an overview of a TFF filter with eight Fabry-Perot filters in cascade, resulting in a multiplexing/demultiplexing of eight different channels with wavelengths λ1− λ8.

3.4.3 Interleaver

With an increasing demand for data capacity, a more effective usage of the bandwidths are needed to enable more channels to be packed within the DWDM-band. As many filter

Figure 17: An overview of a TFF filter with eight Fabry-Perot filters in cascade [9].

Figure 18: Transfer functions for one, two and three Fabry-Perot filters in cascade [9].

technologies are limited to a specified channel spacing, an optical interleaver can be used to cut down the channel spacings even more. [29] The purpose of an optical interleaver is to separate or combine incoming channels of a DWDM signal with a periodic spacing between channels. The most basic interleaver, denoted 1:2, divides the incoming signal into odd and even channels. An example could be a system with channels that are 50 GHz apart, which then are divided into two systems where the signals are 100 GHz apart. Other typical interleavers are 1:4 or 1:8 which separate every fourth or eighth channel respectively, but for a general case, there are also 1:N interleavers, separating every N:th channel [30]. Figure 19 shows an example of a 1:2 and a 1:4 interleaver.

Figure 19: A 1:2 and 1:4 interleaver. Here, λe= even λ, λo= odd λ and λa−d= four different

channel spacings [30].

inter-ferometer, where the output is a result from the interference created by the interferometer which thus will be periodic with frequency. The period of the interleaver is decided by the length difference of the two arms and the distance between the two mirrors where the signal is reflected. Figure 20 shows an example of an interleaver with two Gires-Tournois etalons (GTEs) as mirrors [29].

Figure 20: A schematic of a GTE interleaver [29]. The intensity of the output can be written as

I = 1 2  1 + cos 2π c ν2∆L + (ϕ2− ϕ1)
 (16) where ∆L is the difference between L1and L2, ν is the optical frequency and ϕ1and ϕ2are

the phase shifts between the two mirrors where the difference can be written

ϕ2− ϕ1= 2 tan−1  1 +√R2 1 −√R2 tan(φ)  − 2 tan−1 1 + √ R1 1 −√R2 tan(φ)  (17) where R1 and R2are the reflectivities of the front mirrors on the first and second arm and

φ is the phase shift due to the distance travelled within the GTEs.

In measurement tests, interleavers have been able to separate channels with spacings of 100, 50, 25 and 12.5 GHz for DWDM. The free spectral range (FSR) of the filter function decides the period of the interleaver [31] and can be seen in figure 21. In order to align the interleavers with the DWDM signals, to avoid a mismatch between signals and filters leading to walkoff losses, it is of high importance that both the signals and the interleaver follow the specifications set by the International Telecommunication Union (ITU). Apart from this possible alignment error, there can also be offset errors, or drifts, that may arise due to thermal changes [30], just as for the passive AWG or TFF.

4

Methods and measurements

The following section describes how the lab equipment was set up, how the data was recorded and handled and how the results were found from these. More information about the specific filters that were used are also presented here, such as how the interior of the filter looks like and how the filter channels are positioned. This section only considers the specification of the filters while the background theory and functionality is described in section 3.4 Filters.

4.1

Filters

4.1.1 TFF

When a signal enters the TFF from the line, the signal will first go through an 8-skip-0 filter which simply is just a bandpass filter that is selected to cover all the eight channels for this specific filter. In typical systems with 40 channels, the signals will go through five TFFs which are in cascade, each with an individual 8-skip-0 filter to filter out the relevant bandpass to fit the output ports of each filter. Here, the filter is designed to have four add/drop pairs, i.e four pairs of channels, one as a transceiver port and one as a receiver port, but since each port is reciprocal, it can be used as eight bidirectional channels. For this filter, the names of the ports are connected to the center frequency of that port, e.g. port number 926 has a center frequency of 192.6 THz. The ports are 100 GHz separated in frequency, which is typical for a DWDM filter.

Figure 22: A schematic of the TFF with an 8-skip-0 filter and the following four add/drop pairs.

4.1.2 AWG

Unlike the TFF, the AWG can mux and demux 40 channels at once from the line. This AWG is also athermal, meaning that when temperature changes, a metal board will compensate for the focus shift that appears and align the channel signal to the right port. Also the ports of the AWG are separated 100 GHz from each other.

4.1.3 Interleaver

This interleaver has one input to demultiplex channels of 50 GHz spacing into two outputs of 100 GHz spacings, one odd output and one even. The Interleaver is reciprocal and can thus also be used to multiplex two channels into one. In this thesis, the even output has been used since that matches channel spacing of the TFF and AWG.

Figure 23: A schematic of the 40 channel athermal AWG filter with a 100 GHz channel spacing, here with a monitor (tap coupler) attached to the line, but that won’t be used in this thesis. Channel 1 corresponds to the 191.9 THz channel, channel 40 corresponds to the 195.8 THz channel, TX means ”transceived” and RX means ”received”.

4.2

Lab setups

The following sections describe the procedure of the measurements, how the lab equipment was used and set up and how the measurements were performed. To start with, all mea-surements were put in two different categories: one for measurement of filter parameters such as passband, isolation, insertion loss, center wavelength drift and crosstalk, and one for measurement of the BER. These two tests were performed at different occasions for each filter. The following list describes the measurements in more detail.

Filter characterization:

• Measure insertion loss and isolation [dBm] with respect to temperature [◦_{C] using a}

lightwave multimeter (Agilent 8163A Lightwave Multimeter). The measurement will be done on a couple of chosen channels using an adjustable transceiver.

• Measure the 3 dB passband [GHz], with respect to temperature [◦_{C], from the filters’}

transfer functions using an optical spectrum analyzer (AQ6370D Yokagawa). The inserted signal will be from an amplified spontaneous emission (ASE) source.

• Measure center wavelength drift [GHz] with respect to temperature [◦_{C] using an}

optical spectrum analyzer (AQ6370D Yokagawa). The inserted signal will be from an amplified spontaneous emission (ASE) source.

• Measure crosstalk [dB] over temperature [◦_{C] using an optical spectrum analyzer}

(AQ6370D Yokagawa) and compare it with isolation measurements done with the lightwave multimeter.

BER measurements:

• Measure bit error rate (BER) with respect to temperature [◦_{C] and with respect to}

adjacent channels using a BER-tester (Anritsu MP1800A Signal Quality Analyzer). Before the measurements were initiated, the method had to be decided regarding what filter channels to collect data from, what temperatures and how the measurements should be performed in detail. Firstly, it was decided to analyze only three different channels for every filter setup, two at the edges of the filter channels and one in the middle. This resulted in choosing channels 919, 922 and 926 for the TFF and channels 1, 20 and 40 for

the AWG. Table 1 shows the specified frequencies and wavelengths for each channel. The temperatures at which to collect data were chosen to -40, -20, 0, 25, 50, 70 and 85◦C, since the standardized I-temp is defined as -40-85◦C. The filters have been located in a V¨otsch temperature chamber (VT3 7060) and the temperature has been measured both with the V¨otsch and a thermometer (Fluke 50D Thermometer) with probes located in the middle of the V¨otsch chamber and close to the filter module.

Filter Channel Frequency [THz] Wavelength [nm] TFF ch 919 1.9190 1562.23 ch 922 1.9220 1559.79 ch 926 1.9260 1556.55 AWG ch 1 1.9580 1531.12 ch 20 1.9390 1546.12 ch 40 1.9190 1562.23

Table 1: The specified frequency and wavelength for each channel that has been used in the measurements.

4.2.1 Filter characterization

The following section describes the procedures of each measurement for the filter character-ization in the same order as above. Here, the input power is a signal from the ASE-source which basically amplifies white noise to a power level of approximately 15-20 dBm and covers all frequencies relevant to the measurements done on the filters. This is shown in figure 24 where the ASE-source has been measured with the Yokogawa, covering the whole ASE-span.

Figure 24: A graph showing the power of the ASE-light measured with the Yokogawa. The filters used in this study all operate within the same frequency span as the passband of the ASE-light.

The following list describes the procedures that were done for each measurement in the filter characterization.

• Insertion loss. To achieve the insertion loss for every filter at every temperature, all the measured power data was used to plot graphs of the filter functions. Each channel, or filter function, was plotted against the frequency offset from the ITU-T definition (see table 1). Using this offset, the insertion loss could be found by taking the lowest power within the ±12.5 GHz offset.

To double check that the insertion loss had been calculated correctly, it was also measured with the Agilent 8163A Lightwave Multimeter. This was done both for the TFF and the AWG filter for all channels using an optically tunable transceiver laser

as an input signal. The signals were measured both with and without the filters for all channels. The difference between the input signal and the output signal of the filter is equal to the insertion loss of the filter. The results are provided in table 6. The measurement was done at a temperature of 25◦C.

• Isolation. To find the adjacent isolation for each channel, the highest power within the bandwidth of the adjacent channels according to the ITU-T grid were taken. This was done for both the right and left channel, were the isolation was noted as the worst of these two.

• Filter function. When using the Yokogawa to measure the filters’ transfer functions, the measurement procedure always started with collecting data from the ASE-source, sweeping across all relevant frequencies. This was done daily as a start for every mea-surement. Each channel was then collected separately to the Yokogawa and was swept across a relevant frequency band. Since the ASE-source is not perfectly distributed over the frequency-band within the TFF and AWG filters, and not at 0 dBm power, the correct transfer function had to be calculated by taking the difference between the ASE-source and the output signal from the chosen channel, which was done with the Yokogawa. This resulted in a plot of the power in logarithmic scale of the filter function against frequency in linear scale. An example of this measurement can be seen in figure 48. Apart from collecting data from the filter function, the Yokogawa’s filter functions were also used to calculate center wavelength, peak value, 3 dB pass-band width, 10 dB passpass-band width, ripple and also crosstalk to the left and right of the signal. This data was however only used to compare and confirm that my own calculations seemed correct.

• 3 dB passband. When calculating the 3 dB passband for each channel, the power at the center wavelength calculated by the Yokogawa was used. Then the power level was noted 3 dB below the center power level on both sides of the center wavelength. The minimum wavelength at where the power had dropped 3 dB, called f3dB was used to

define the 3 dB passband as the frequency range fcenter±fc e n t e r2−f3 d B where fcenter

is defined according to the ITU-T grid [21].

• Crosstalk. For the calculation of the crosstalk, the method consisted of finding the worst case adjacent and non-adjacent isolations of all channels that were measured. This was done by finding all maximum power levels within the adjacent and non-adjacent channels, i.e. within fchannel±12.5 GHz, where fchannelis defined according

to the ITU-T grid, converting them to a linear scale, add them up and then convert them back to logarithmic power. The result is the total crosstalk or the total isolation of all channels combined.

The measurements were done in the lab at Infinera with their own equipment. Table 2 shows a measurement matrix of all the measurements that were done in the filter characterization and figure 25 shows a sketch of the lab setup and what equipment was used where the red line shows the transmission line of the signal from start to end. Moreover, the interface of the Yokogawa is shown in figure 26 with an example of a measurement done on the AWG filter.

AWG TFF Interleaver +AWG Interleaver +TFF Insertion loss Isolation 3 dB passband

Center wavelength drift Crosstalk

Table 2: Depicting an overview matrix for all measurements that were done in the filter characterization.

Figure 25: The lab setup for the filter characterization. In the figure, the numbers correspond to: 1. ASE-source 2. V¨otsch chamber 3. V¨otsch thermometer 4. Interleaver 5. AWG or TFF 6. The Fluke thermometer 7. The Yokogawa Optical Spectrum Analyzer.

Figure 26: An overview of the Yokogawa interface for a measurement on the AWG filter. In the picture, the purple curve represents the light from the ASE-source, the yellow curve is the transfer function of the filter and the green curve is the difference between the transfer function and the source, i.e a normalization of the filter function relative the ASE-source. The pink curves in the middle of the AWG passband are the 10G signals transmitted at a frequency of 192.0 THz.

4.2.2 BER measurements

The following section describes the measurement procedure for the BER measurements. For the BER measurements, a system has been set up to simulate traffic between two nodes in the mobile fronthaul. The first node, the central office (CO), corresponds to the output ports of the transponder/muxponder, the first filter and the first interleaver, which are numbered 1-2, 3 and 4 in figure 27. These are kept in a temperature controlled office at a room temperature of approximately 25◦C. After the CO, the signal is attenuated with a variable optical attenuator (VOA) which is used to regulate the received power, and also corresponds to losses in the fiber here. The signal then reaches the remote node (RN) which corresponds to the filters at the antenna site. Here, the filter and the interleaver are put in the V¨otsch temperature chamber just as in the filter characterization. After this, the signal is attenuated again with a fiber of 5 dB attenuation to compensate for the limited attenuation provided by the VOA. The signal lastly goes back to the transponder/muxponder which has a forward error correction (FEC) function that detects and tries to correct the wrong bits. The FEC’s software has an inbuilt function that reads both the corrected bits and uncorrected bits which was used to calculate the BER before the FEC. The function accumulates all corrected and uncorrected bits once the function is initiated. Thus, to calculate the pre-FEC BER, each measurement had to be timed in order to get a rate of bit errors. Also, for every data point of BER measured, the received power was measured with the power meter. The confidence level CL in equation 9 was then calculated to double check that the measurements had at least a 95 % confidence level. When it comes to the BER measurements on the filters, the relative difference between BER-curves is of most interest, as this indicates a dB-penalty.

Figure 27: The lab setup for the BER measurements. In the figure, the numbers correspond to: 1. Main channel transceiver 2. Two adjacent transceivers 3. First AWG or TFF 4. First Interleaver 5. Variable Optical Attenuator (VOA) 6. Second Interleaver 7. Second AWG or TFF 8. The Fluke thermometer 9. V¨otsch chamber 10. V¨otsch’s thermometer 11. An attenuated fiber of 5 dB 12. FEC-reader 13. Transponder/Muxponder.

Regarding the BER measurements, the following points describe the different measurements that were decided to be performed.

• Measure BER with and without neighbouring channels. As crosstalk may appear differently when having an ASE-source as an input compared to ordinary transceivers with specified frequencies, it was interesting to study the effects of neigh-bouring channels, which also will be the case in a real traffic situation. From the

25◦C 85◦C CO RN With neigh-bours Without neigh-bours With neigh-bours Without neigh-bours Without AWG2 AWG3

Interleaver AWG2 TFF With AWG2 AWG3 Interleaver AWG2 TFF

Table 3: Depicting an overview matrix for all BER measurements that were performed. The filters have been put in two nodes, one in a central office (CO) and one in the remote node (RN) i.e close to the antenna for the realistic case.

results in the filter characterization, especially figures 41-42, it was decided together with the supervisors that only traffic on the adjacent channels would suffice. The adjacent channels are both numbered 2 in figure 27.

• Signal formats. The card in the transponder/muxponder scrambles a signal by itself if no signal is received at the receiver. When calculating BER, a scrambled signal is to prefer as random bits are harder to repair compared to defined patterns.

• New AWG filters. As the BER measurements include two filters at two nodes, instead of just measuring on one filter as in the filter characterization, it was decided to analyze two new AWG filters, which have the same article number, are manufactured in the same way and have the same functionality. Since it’s a node to node connection, it was relevant to connect an AWG-A with an AWG-B. It may be a disadvantage to not use the same filter as in the filter characterization, but as it was interesting to add another AWG, which doesn’t differ very much, it was considered to be a safe change. No measurements have been done on the new AWGs apart from the test measurements done by the manufacturers. Thus, a comparison between the filter characteristics for these three filters could be made. To avoid confusion, these new AWGs have been labeled ”AWG2” and ”AWG3”. When it comes to the TFF, the same filter has been used and only once in the BER measurements. Thus, the TFF is only labeled as ”TFF”. See table 3 for more details.

• With and without interleavers. Similar as in the filter characterization, the nodes are also to be connected both with and without interleavers. As for the case with the AWG, an additional interleaver had to be used. This interleaver is also the same type as the one used in the filter characterization and shouldn’t necessarily cause any apparent affect on the results. Measurement data to compare with can be seen in figure 59.

• Temperatures. For the BER measurements, it was decided to mainly focus on the temperatures that were worst in the filter characterization, which in general was for warmer temperatures. Thus, the measurements were done at 25 and 85◦C.

To add credibility to the study and to make sure that the test results are valid, the following extra measurements were made.

• Measure crosstalk from the adjacent channels. To easier understand how the BER-curves are affected by more traffic on the line, the power at the channel was measured for six different traffic situations using a power meter. These are represented in table 4 where 1 and 0 correspond to the laser being on and off respectively. The adjacent channels were chosen to be 100 GHz away from the main channel, i.e. one output port away at the TFF and AWG filters. All three channels for all filters were transmitted at 10 Gb/s at a temperature of 25◦C using the tunable transmitter. From

the results in figure 41 and 42 one can see that the adjacent channels contribute the most to the crosstalk compared to the non-adjacent channels. Thus, it was decided that only two adjacent channels, one at each side of the channel, would suffice for this measurement.

Adjacent channel left 0 1 0 1 0 1 Main channel 1 1 0 0 0 0 Adjacent channel right 0 1 0 0 1 1 Measured power [dB]

Table 4: An overview of six different traffic situations with or without adjacent channels and the main channel. Here 1 and 0 correspond to the laser being on and off respectively.

• Measuring BER with a BERT. To make sure that the FEC-reader software pro-gram could be used to calculate BER, the method was compared to another mea-surement that included a bit error rate tester (BERT). The BERT used for this was the Anritsu MP1800A Signal Quality Analyzer. This BERT generates a certain pseu-dorandom binary sequence (PRBS) that has a certain cycle of random bits while without the BERT, the signal is just scrambled randomly. The PRBS patterns that were considered were PRBS7 and PRBS31, i.e. cycles with 27_{− 1 and 2}31_{− 1 different}

bit-combinations per cycle. The setup for the BER-measurement with the BERT is very similar to the setup in figure 27 but without the adjacent channels and with the BERT instead of the transponder/muxponder. In order to achieve the correct wave-length on the transceiver, two additional transceivers and a HEX-light card had to be used. These transceivers have a specified maximum bit rate of 10.3 Gb/s however, but could be used anyway to support 11.1 Gb/s.

• Measuring the transceiver. To add further credibility to the measurements, the transceiver alone was measured for BER without any filter. The signal was sent at a 11.1 Gb/s rate with both a PRBS31 and PRBS7 pattern.

4.2.3 Temperature control

Since each of the four filter technologies had to be measured at -40, -20, 0, 25, 50, 70 and 85◦C, i.e at seven different temperatures, it implied that at least 7 × 4 = 28 measurements had to be done. To avoid as much sources of error as possible, it was important to make sure that each measurement would be as similar as possible in the procedure. Therefore, a first temperature test was made on the TFF filter, heating it from room temperature to 60

◦_{C and measuring the temperature both with the V¨otsch thermometer, but also with two}

probes from the Fluke thermometer, one located inside the filter module and one located in the middle of the V¨otsch chamber. By every third minute, the temperatures were sampled, which are summarized in table 5 below.

As can be seen from table 5 and figure 28 by comparing the temperatures, it’s easy to conclude that the heating of the filter is lagging behind the heating of the chamber but also the temperatures of the Fluke thermometer and the V¨otsch may differ by several degrees. After 12 minutes, there is a temperature difference of approximately 12 ◦C between the filter and the chamber temperature measured by the Fluke thermometer. As temperature increases, the material of the filter will expand, thus a temperature difference between the interior and the exterior of the filter may lead to tensions in the material which can be harmful. Since the filters are very sensitive to length contractions and extractions, it was decided to change the temperature using a ramp function with a lower time derivative compared to using a jump function that was used in this test. A ramp with a 2 ◦C/min derivative was used. Also, to save time, the cover of the TFF module was always removed to allow the filter to change temperature quicker. The AWG was however delivered without a cover. To make sure that the filters reached the right temperature, the time between starting

Time [min] Temperature inside the chamber[◦C] Temperature inside the filter[◦C] V¨otsch tem-perature [◦C] 0 22.8 22.9 22.1 3 27.7 24.8 34.5 6 35.6 27.0 44.7 9 42.2 30.5 54.3 12 47.5 35.0 58.7 15 50.6 39.3 59.6 18 52.7 43.0 60.2 21 54.3 46.1 60.1 24 55.3 48.7 60.1 27 56.3 50.9 60.0 30 56.9 52.9 60.0 33 57.4 54.1 60.0 36 57.7 55.2 60.0 39 57.9 56.0 60.0 42 58.2 56.8 60.0 45 58.5 57.5 60.0 48 58.6 57.8 60.0 51 58.7 58.1 60.0 54 58.8 58.3 60.0 57 59.0 58.6 60.0 60 59.0 58.8 60.0

Table 5: Temperatures inside the chamber, inside the filter and of the V¨otsch plotted against time.

Figure 28: A plot of the data provided in table 5.

the ramp and the measurements were always at least one hour long. Both the temperature from the Fluke and the V¨otsch were noted at each measurement. However, the temperatures from the Fluke and the V¨otsch always differed by a few degrees, probably due to a poor calibration of the Fluke, so the measurements were done only when both the temperatures from the Fluke and the V¨otsch had saturated.

4.3

Simulations

In order to validate and add credibility to the measurement results, a try to simulate the setups in the BER measurements has been performed by the use of VPItransmissionmaker. For these simulations, the transfer function of each filter setup has been measured with the Yokogawa and translated into a .txt-file that consists of two columns, one with the frequency and one with the filter attenuation at that specific frequency. A continuous wave (CW) distributed feedback laser (DFB) module was used as a laser source with a user-defined frequency and output power. The filter transfer function could then be drifted over frequency by using an inbuilt sweep function in VPI which could be interpreted as a temperature change causing a drift to appear. The signal is then received at an optical receiver that also estimates a BER. An overview of the modules that were used in the simulation program is shown in figure 29 and are further explained below.

Figure 29: An overview of the simulation setup.

The following list shows an explanation to all blocks or modules that are depicted in figure 29 according to the numbers.

• 1. A block corresponding to the laser or transmitter. The laser is further explained in figure 30.

• 2. The transfer function of the filters that have been measured by the Yokogawa from the filter characterization. The transfer function can be shifted over frequency in order to attain an offset that could correspond to the drift that appears when the filters are being heated up or cooled down. By using this offset, simulations of temperatures beyond -40-85◦C can be performed.

• 3. A variable attenuator that is used for stepping the attenuation in order to achieve the requested received power for the BER measurements.

• 4. A signal analyzer measuring the signal before it reaches the receiver.

• 5. An optical receiver, here chosen to be an avalanche photo diode (APD) with a thermal noise of 2.535 ×10−11 in order to achieve a simulation result as close as possible to the real case.

• 6. Power meter measuring the power after the filter and the variable attenuator. • 7. A numerical analyzer plotting BER over received power.

The following list shows an explanation to all blocks or modules that are depicted in figure 30 according to the numbers.

• 1. A laser module consisting of a distributed feedback laser (DFB) that creates a continuous wave (CW) optical signal with a desired frequency and power. For the simulations in this study, the laser has been used at a frequency of 192.0 THz, a power of 1 mW (0 dBm) and at 10 Gb/s.

Figure 30: An overview of the laser module that is represented as number 1 in figure 29.

• 2. A PRBS-generator creating a pattern of choice and the probabilities of sending a ”0” or a ”1”.

• 3. A coder that modulates the signal to Non Return to Zero (NRZ) samples.

• 4. A block that uses a Gaussian filter to smooth out the sampled signals with a user-defined rise-time.

• 5. A module that simulates a Mach-Zehnder modulator

• 6. A module that is used to add global labels to signals or blocks of the user’s choice. • 7. An optical attenuator used to attenuate the output signal.

5

Results and analysis

5.1

Filter characterization

The following sections present the results from the filter characterization. Each section of figures is followed by an analysis and explanation to the results. Shortly, the filter charac-terization concerns the characteristics of each filter setup regarding insertion loss, isolation, filter transfer functions, 3 dB passband, center wavelength drift and crosstalk that have been described above in section 3 Theory and 4 Methods and measurements.

5.1.1 Insertion loss and isolation

The following section addresses the measurements and calculations on insertion loss and isolation. The results provided in table 6 can be compared to the calculated results measured by the Yokogawa presented in figures 31 and 32.

Filter and channel num-ber Input power from transceiver [dBm] Measured power through filter [dBm] Calculated insertion loss [dB] TFF 919 0.20 -1.86 2.06 TFF 922 0.14 -1.71 1.85 TFF 926 0.08 -2.29 2.38 AWG 1 0.61 -3.67 4.28 AWG 20 0.56 -3.47 4.03 AWG 40 0.33 -4.49 4.82

Table 6: Table depicting the input power of the transceiver, received power to the power meter with the filters and the resulting insertion loss. The measurement was done at a temperature of 25◦C.

Figure 31: Insertion loss plotted over temperature for the TFF for all three channels.

Figure 32: Insertion loss plotted over temperature for the AWG for all three channels.

On pages 31-34, the adjacent isolation has been plotted for each filter technology. The three signal curves in every graph labeled (a) are taken from the lowest (i.e. worst) isolations in the graphs denoted (b)-(d). Also, the figure at the bottom of each page shows an example of where the isolation data has been found for three different temperatures from the adjacent channels. For the cases where the interleaver is connected, the worst adjacent isolation has been gathered from a channel passband 50 GHz away from the main channel instead of 100 GHz.

(a) Adjacent isolation for each channel. (b) Left and right adjacent isolation for channel 919.

(c) Left and right adjacent isolation for channel 922.

(d) Left and right adjacent isolation for channel 929.

Figure 33: Adjacent isolation plotted over temperature for the TFF.

Figure 34: TFF channel 919 at -40, 25 and 85◦C.

(a) Adjacent isolation for each channel. (b) Left and right adjacent isolation for channel 919.

(c) Left and right adjacent isolation for channel 922.

(d) Left and right adjacent isolation for channel 926.

Figure 35: Adjacent isolation plotted over temperature for the Interleaver+TFF.

Figure 36: Interleaver+TFF channel 1 at -40, 25 and 85◦C.

(a) Adjacent isolation for each channel. (b) Left and right adjacent isolation for channel 1.

(c) Left and right adjacent isolation for channel 20.

(d) Left and right adjacent isolation for channel 40.

Figure 37: Adjacent isolation plotted over temperature for the AWG.

Figure 38: AWG channel 1 at -40, 25 and 85◦C.