High Capacity Short-Reach Optical Communications1047182/... · 2016. 11. 16. · kilometers, and 2)...

High-Capacity Short-Reach

Optical Communications

RUI LIN

Doctoral Thesis in Information and Communication Technology School of Information and Communication Technology

KTH Royal Institute of Technology Stockholm, Sweden

December. 2016

KTH Royal Institute of Technology School of Information and Communication Technology

SE-164 40 Kista, SWEDEN TRITA-ICT 2016:33 ISBN: 978-91-7729-177-0

Akademisk avhandling som med tillstånd av Kungl Tekniska Högskolan framlägges till offentlig granskning för avläggande av doktorsexamen i Informations- och Kommunikationsteknik, måndag, den 12 december 2016, klockan 10.00, i Sal B, Electrum, Kungl Tekniska Högskolan Kistagågen 16, Kista.

© Rui Lin, December 2016

Tryck: Universitetsservice US-AB

ii

Abstract

The global traffic is experiencing an exponential growth due to the emerging bandwidth hungry applications and the increase of the number of users, posing a severe challenge to the communication networks in terms of capacity. As a future-proof technology to provide high capacity, fiber communication is widely implemented in different network segments. According to the transmission distance, the fiber networks can be categorized as the long-haul and the short-reach. This thesis focuses on the short-reach communication networks, including 1) fiber access network connecting the end users to the metro/core networks and typically covering tens of kilometers, and 2) optical datacenter network handling the traffic within the datacenter with transmission distance up to a few kilometers.

For fiber access networks, wavelength division multiplexing passive optical network (WDM-PON) is one of the promising candidates, where a dedicated wavelength channel is assigned to each user guaranteeing high data rate. The dense wavelength channels can enlarge the number of users supported in WDM-PON, but makes the signals vulnerable to the wavelength drift occurring in the lasers. In this regard, we propose two schemes based on optical frequency comb technique and carry out the experiments to verify that the proposed two schemes are able to generate stable multiple carriers for WDM-PONs. Meanwhile, radio-over-fiber techniques allow wireless signals to be optically transmitted for a long distance and are therefore widely considered for transmission of radio signals between central offices and the cells. Millimeter wave (MMW) over fiber techniques, on the other hand, offer high bandwidth and are hence promising for future high capacity mobile access. In this thesis, we propose and experimentally demonstrate a palm-shaped spectrum generation, where the high-power central carrier (like the middle finger of the palm) can be used at the cell for the upstream transmission, while multiple MMW bands like the other fingers of the palm are capable of transmitting different downstream data simultaneously.

Regarding optical datacenters networks, passive optical interconnects (POIs) have been proposed as an energy-efficient solution thanks to the fact that only passive optical components are used for interconnection of different servers. However, the high insertion loss of the passive components may result in a serious scalability problem. In this thesis, we develop a methodology that is able to take into account various physical- layer aspects, e.g., receiver types, modulat ion formats, to quantify the scalability of the POIs. Both theoretical analyses and experimenta l measurements have been performed to assess the scalability of various coupler-based POIs.

Key words: optical communication; optical frequency comb; radio-over-fiber; optical interconnect.

iii

Sammanfattning

Den globala datatrafiken växer exponentiellt, både på grund av nya bandbreddskrävande applikationer och ökningen av antalet användare. Detta innebär en utmaning för kommunikationsnätens kapacitet. Fiberoptisk kommunikation är en framtidssäker teknik för att möta detta kapacitetsbehov och används redan i stor utsträckning i olika delar av näten. Beroende på överföringsavstånd, kan fibernät kategoriseras som långdistansnät eller nät med kort räckvidd. Denna avhandling behandlar nät med kort räckvidd, innefattande dels 1) accessnät som förbinder slutanvändarna till stadsnätet/ huvudnätet och typiskt omfattar tiotals kilometer, dels 2) optiska datanätverk som hanterar den interna trafiken inom datacenter med överföringsavstånd upp till ett par kilometer.

För fiberaccessnät är en av de lovande teknikerna våglängdsmultiplexade passiva optiska nät (WDM-PON), där en dedicerad våglängdskanal tilldelas varje användare vilket garanterar hög datahastighet. Genom ett litet kanalavstånd så kan antalet användare i WDM-PON utökas men det gör samtidigt systemet känsligt för våglängdsdrift hos lasrarna. För att råda bot på detta, föreslår vi två system baserade på optisk frekvenskams-teknik. Vi validerar experimentellt att de kan generera stabila optiska bärvågor för WDM-PON. Radio-över –fiber-tekniken gör samtidigt det möjligt att sända radiosignaler över en lång sträcka och används därför i mobilsystem för överföring mellan centralstationen och radiocellerna. Millimetervågor (MMW) över fiber erbjuder ännu större modulationsbandbredd och är lovande för framtidens mobilradiosystem med hög kapacitet. I denna avhandling föreslår vi, och demonstrerar experimentellt, generation av ett frekvenskams-spektrum som är format som en handflata, där en central bärare med hög effekt (långfingret på handflatan) kan användas i radiocellerna för uppströms överföring, medan multipla MMW band (övriga fingrar) samtidigt kan överföra olika data nedströms. När det gäller nätverk för optiska datacenter, har passiva optiska interconnects (POI) föreslagits som en energieffektiv lösning, där endast passiva optiska komponenter används för ihopkoppling av servrarna. Höga inkopplingsförluster hos passiva optiska komponenter kan emellertid leda till allvarliga skalbarhetsproblem. I denna avhandling presenterar vi en nyutvecklad metod för att kvantifiera skalbarheten, vilken tar hänsyn till olika faktorer i det fysiska lagret som t.ex. mottagartyp och modulationsformat. Både teoretiska analyser och experimentella mätningar har utförts för att utvärdera skalbarheten hos olika kopplarbaserade POI.

Nyckelord: . fiberoptisk kommunikation; optisk frekvenskams; Radio-över –fiber; passiva optiska interconnects

iv

Acknowledgements

First and foremost, I would like to express my sincere gratitude to my main supervisor, Assoc. Prof. Jiajia Chen, for her invaluable support and for her contributions of time and ideas, for all the discussion and inspiration. I am thankful for the excellent example that Jiajia provides as a successful women in academy. The enthusiasm she has for research has affected and motivated me. I would like to express appreciation to my co-supervisor, Professor Lena Wosinska, for her conscious and unconscious guidance, which is priceless to my life. My sincere thanks goes to my supervisor in HUST, China, Prof. Ming Tang and Prof. Deming Liu, for the appreciated advices and generous support during the last years.

I would like to thank Prof. Robert Forchheimer for accepting the role of advance reviewer of this thesis and translating the abstract into Swedish. I am grateful to Dr. Gemma Vall-llosera for evaluating my PhD proposal and providing insightful comments. I would like to offer thanks to Assoc. Prof. Changyuan Yu for accepting the role of opponent and to Assoc. Prof. Juan Jose. Vegas Olmos, Prof. Magnus Karlsson, and Dr. Qin Wang for accepting the role of the member of the grading committee of my doctoral disputation.

I would give my special thanks to Dr. Xiaodan Pang and Dr. Oskars Ozolins from Acreo, Prof. Erik Agrell and Dr. Krzysztof Szczerba from Chalmers University of Technology, Prof. Xin Yin from Ghent University for the fruitful collaboration. It is a great experience to work with the excellent researchers. I also would like to offer my thanks to all the members in the VR project, Prof. Robert Forchheimer, Dr. Matteo Fiorani, Dr. Houman Rastegarfar, Dr. Dung Pham Van, Yuxin Cheng, Dr. Xuezhi Hong, and Dr. Ajmal Muhammad, for the interesting discussion in every one of our project meetings

I am grateful to all the members in our ONLab family, Dr. Pawel Wiatr, Dr. Marija Furdek, Dr. Meiqian Wang, Dr. Andrea Sgambelluri, Forough Yaghoubi, Muhammad Rehan Raza, Jun Li, Dr. Carlos Natalino Silva, and Dr. Aleksejs Udalcovs. I will miss the cheerful fika and delighted talks. I would also thank to my colleagues in China, Zhenhua, Ruoxu, Qiong for your great effort and help even though we have jet lag and long distance in between.

I would like to thank Sarah Winther, Susy Mathew, Collberg Sussane for helping with my administrative issues. This thesis cannot be completed without your help.

Many thanks to my friends in Sweden in particular, Xin, Yoyo, Jingna, Xiaofen, Xiaoke, Yanpeng, Tian, Min, Cheuk Wai, Roro, for the time we spend together..

Last but not least, I would like to express my heartfelt appreciation to my family, for giving me life and love, for their understanding and generous support during these years. My special thanks to Haiqing Zhao for his love and accompaniment.

Rui Lin

Stockholm, December 2016

v

Contents

Abstract ...................................................................................................................................... iii Sammanfattning ......................................................................................................................... iv

Acknowledgements .....................................................................................................................v

List of abbreviations ................................................................................................................. viii List of publications ......................................................................................................................x

Chapter 1 Introduction ....................................................................................................... 1

1.1. High capacity optical access network ......................................................................... 2

1.2. High capacity optical intra-datacenter network .......................................................... 4

1.3. Overview of the thesis contributions .......................................................................... 5

1.4. Thesis organization ..................................................................................................... 6

Chapter 2 Short-reach optical communication networks ............................................... 9

2.1. Fiber access network ................................................................................................... 9

2.1.1. Passive optical networks ..................................................................................... 9

2.1.2. Radio-over-fiber systems ...................................................................................11

2.2. Intra-datacenter networks.......................................................................................... 12

2.2.1. Energy consumption in datacenters .................................................................. 13

2.2.2. Characteristic of datacenter traffic .................................................................... 14

2.2.3. Energy efficient datacenter network architectures ............................................ 14

2.2.4. Advanced modulation formats applied for intra-datacenter optical links......... 15

Chapter 3 Optical frequency comb based DWDM-PON .............................................. 17

3.1. Mach-Zehnder modulator ......................................................................................... 17

3.2. Optical frequency comb generation .......................................................................... 19

3.3. DWDM-PON based on OFC .................................................................................... 22

Chapter 4 Palm-shaped spectrum based RoF system .................................................... 25

4.1. Palm-shaped spectrum generation. ........................................................................... 25

4.2. Dual-band MMW RoF system using a palm-shaped spectrum ................................ 26

Chapter 5 Scalability analysis of passive optical interconnects in DCN ...................... 29

5.1. Scalability assessment methodology......................................................................... 30

5.1.1. Link budget model ............................................................................................ 31

5.1.2. System power budget ........................................................................................ 32

5.2. Coupler-based POI architectures .............................................................................. 34

5.2.1. Description of the architectures ........................................................................ 34

5.2.2. Assessment results ............................................................................................ 36

5.3. Introducing EDFA in POIs ........................................................................................ 36

5.3.1. EDFA enabled POI architecture ........................................................................ 37

5.3.2. Assessment result .............................................................................................. 38

5.4. Study on the impact of modulation formats.............................................................. 39

5.4.1. PAM, DMT and EDB........................................................................................ 39

5.4.2. Experiment validation of DMT/M-PAM in POI ............................................... 41

5.4.3. Scalability assessment results ........................................................................... 42

Chapter 6 Conclusions and future works............................................................................. 45

6.1. Conclusions ............................................................................................................... 45

6.2. Future work ............................................................................................................... 46

Chapter 7 Summary of the original works ...................................................................... 49

References ............................................................................................................................... 53

vii

List of abbreviations AWG Arrayed waveguide grating

ASE Amplified spontaneous emission B2B Back-to-back BER Bit error rate

BPF Band pass filter BS Base station

CAZAC Constant amplitude zero auto-correlation CO Central office CS Central station

CW Continuous wave CWDM Coarse wavelength division multiplexing

DAC Digital-to-analogue converter DCN Datacenter network DFB-TWEAM Distribution feedback travelling wave electro-absorption modulator

DFE Decision feedback equalizer DMT Discrete multi-tone

DPMZM Dual parallel Mach Zedher modulator DSO Digital sampling oscilloscope DWDM Dense wavelength division multiplexing

EDFA Erdium doped fiber amplifier FBG fiber bragg grating

FEC Forward error correction HDTV High division television IM Intensity modulator

LO Local oscillator MAC Media acdess control

MEF multi-element fiber MEMS Micro-electro-mechanical system MMW Millimeter wave

M-PAM Multi-level pulse amplitude modulation MZM Mach-Zehnder modulators

viii

OFC Optical frequency comb OFDM Orthogonal frequency division multiplexing

OLT Optical line terminal ONI Optical network interface

ONU Optical network unit OSSR Optical sideband suppressed ratio P2P Point to point

PAM Pulse amplitude modulation PC Polarization controller

PD photo diode PIN Positive-intrinsic-negative PM Phase modulator

POI Passive optical interconnect PON Passive optical network

PPG Pulse-pattern generator PPG: Pulse-pattern generator RF Radio frequency

RIN Relative intensity noise RoF Radio over fiber

Rx Receiver SDM Spatial division multiplexing SNR Signal-to-noise ratio

TDM Time division multiplexing TDMA Time division multiple access

ToR Top-of-rack

Tx Transmitter

WDM Wavelength division multiplexing

VR Virtual reality

WSS Wavelength selective switch

ix

List of publications Publications included in this thesis: Paper I:

R. Lin, Z. Feng, M. Tang, S. Fu, P. Shum and D. Liu, “Spacing Switchable Flat Broadband Optical Comb Generation Based on Cascaded Electro-optical Modulator”, Asia Communications and Photonics Conference and Exhibition (ACP), 2013.

Paper II: R. Lin, M. Tang, R. Wang, Z. Feng, So. Fu, D. Liu, J. Chen and P. Shum, “An Ultra-dense Optical Comb Based DWDM-OFDM-PON System”, Progress in Electromagnetics Research Symposium (PIERS), 2014.

Paper III: R. Lin, Z. Feng, M. Tang, R. Wang, S. Fu, P. Shum, and D. Liu, “Palm-Shaped Optical Spectrum Generation for Fiber-Wireless Integrated Communication with Dual-Band Millimeter Wave Capability”, Asia Communications and Photonics Conference and Exhibition (ACP), 2014 (Best student paper).

Paper IV: R. Lin, Z. Feng, M. Tang, R. Wang, S. Fu, P. Shum, D. Liu, and J. Chen , “Palm-Shaped Spectrum Generation for Dual-band Millimeter Wave and Baseband Signals over Fiber”, Optics Communications, vol. 367, pp. 137-143, 2016.

Paper V: R. Lin, K. Szczerba, E. Agrell, L. Wosinska, M. Tang, and J. Chen, “Scalability Analysis of Coupler Based Optical Interconnects”, Photonics Journal, submitted.

Paper VI: R. Lin, K. Szczerba, E. Agrell, L. Wosinska, M. Tang, and J. Chen, “To Overcome the Scalability Limitation of Passive Optical Interconnects in Datacentres”, Asia Communications and Photonics Conference and Exhibition (ACP) 2016, accepted.

Paper VII: R. Lin, X. Pang, O. Ozolins, Z. Feng, A. Djupsjöbacka, U. Westergren, R. Schatz, G. Jacobsen, M. Tang, S. Fu, D. Liu, and J. Chen, “Performance Evaluation of PAM and DMT for Short-range Optical Transmission with High Speed InGaAsP DFB-TWEAM”, Optical Fiber Communication Conference (OFC), 2016.

Paper VIII: R. Lin, X. Pang, O. Ozolins, Z. Feng, A. Djupsjöbacka, U. Westergren, R. Schatz, G. Jacobsen, M. Tang, S. Fu, D. Liu, and J. Chen, “Experimental Validation of Scalability Improvement for Passive Optical Interconnect by Implementing Digital Equalization”, European conference and exhibition on optical communication (ECOC), 2016.

Paper IX: X. Yin, M. Verplaetse, R. Lin, J. V. Kerrebrouck, O. Ozolins, T. D. Keulenaer, X. Pang, R. Pierco, R. Vaernewyce, A. Vyncke, R. Schats, U. Westergren, G. Jacobsen, S. Popov, J. Chen, G. Torf and J. Bauwelinck, “First Demonstration of Real-Time 100 Gbit/s 3-Level Duobinary Transmission for Optical Interconnects”, European conference and exhibition on optical communication (ECOC), 2016, post deadline paper.

x

Publications not included in the thesis: Paper I: R. Wang, R. Lin, M, Tang, H. Zhang, Z, Feng, S. Fu, D. Liu, and P. Shum, ”Electrically

Programmable All-Fiber Structured Second Order Optical Temporal Differentiator”, Photonics Journal, IEEE, vol. 7, no. 3, pp 1-10, 2015.

Paper II: Z. Feng, B. Li, M. Tang, L. Gan, R. Wang, R. Lin, Z. Xu, S. Fu, L. Deng, W. Tong, S. Long, L. Zhang, R. Zhang, S, Liu, and P. Shum, “Multicore-Fiber-Enabled WSDM Optical Access Network With Centralized Carrier Delivery and RSOA-Based Adaptive Modulation”, Photonics Journal, IEEE, vol. 7, no. 4, pp. 1-9, 2015.

Paper III: Z. Feng, M. Tang, Q. Wu, R. Lin, R. Wang, S. Fu, L. Deng, D. Liu, and P. Shum, "A Simplified Adaptive Modulation Scheme for RSOA Based DDO-OFDM System using CAZAC Precoding." OFC, W3A6, 2016.

Paper IV: J. Wu, M. Tang, L. Xu, S. Zhu, J. Cheng, Z. Feng, L. Zhang, X. Wang, R. Lin, L. Deng, S. Fu, P. Shum, and D. Liu, “A Robust and Efficient Frequency Offset Correction Algorithm with Experimental Verification for Coherent Optical OFDM System”, Journal of Lightwave Technology, vol. 33, no. 18, pp. 3801-3807. 2015.

Paper V: R. Wang, M. Tang, L. Zhang, H. Zhang, Z. Feng, R. Lin, S. Fu, D. Liu, and P. Shum, “Demonstration of Programmable In-Band OSNR Monitoring Using LCFBG With Commercial Thermal Printer Head”, Photonics Journal, IEEE, vol. 7, no. 4, pp 1-9, 2015

Paper VI: M. Verplaetse, R. Lin, J. Van Kerrebrouck, O. Ozolins, T. De Keulenaer, X. Pang, R. Pierco, R. Vaernewyck, A. Vyncke, R. Schatz, U. Westergren, G. Jacobsen, S. Popov, J. Chen, G. Torfs, J. Bauwelinck and X. Yin “First Demonstration of Real-Time 100 Gb/s 3-Level Duobinary Transmission for Optical Interconnects”, Journal of Lightwave Technology (invited).

xi

Chapter 1

Introduction

Nowadays, everyone and everything is connected with the rest of the world as a consequence of the fast-developing information society. It is quite common to see someone taking a selfie and Instagramming it to win likes from friends. To realize it, the traffic carrying the photo, location information and request for a certain type of services, etc., may pass through access networks, metro networks and core networks, and then arrive at one or several servers in a rack within a datacenter. After being processed in the datacenter, an acknowledgement is sent back through the communication network, and finally the picture appears on the screen at the user’s hand.

Today, the high capacity demand in every segment of the network is, on one hand, driven by the increase of end user count and the emerging bandwidth-consuming Internet applications, such as virtual reality (VR), high definition television (HDTV), video-based social network, etc., and on the other hand, posing significant challenges to the communication networks. The communica t ion network hierarchy is shown in Figure 1, which can be divided into two major categories according to the transmission distance, namely long-haul and short-reach networks. Metro and core networks with transmission distance of hundreds or thousands of kilometers have relatively long reach and belong to the former category. Short-haul networks can be subcategorized into access networks and intra-datacenter networks (DCNs). Access networks bridge the end users to the metro/core networks with several tens of kilometers and shorter optical fiber in between while DCNs interconnect the servers and racks within the datacenter typically with a very short transmiss ion distance (up to several kilometers). In the long-haul communication networks, optical communication is widely deployed and considered a future-proof technology, providing high-

1

Chapter 1 Introduction

capacity. A transmission speed exceeding 1 Pb/s carried by a single fiber has already been demonstrated in the laboratory [1]–[3], while wavelength division multiplexing (WDM) system supporting 20 Tb/s per fiber has already been deployed in the long-haul communication networks [4]. In the recent years, optical communication has also gradually been applied for short-reach communication, which is the main focus of this thesis.

1.1. High capacity optical access network

Access network has been seen as the last mile of the information highway. Optical access is superior to copper-based techniques, such as coaxial cable and copper twisted pair, in terms of both capacity and transmission distance. Among different optical access solutions, passive optical networks (PONs) use only passive components in the outside plant. It attracts a great interest in both academia and industry because of the low cost and the low energy consumption. Besides, passive components have also low probability of failure compared to active ones and hence high reliability. In a PON system, an optical line terminal (OLT) is located at the central office, connected to one or multiple optical network units (ONUs) at the user ends. In time division multiplexing (TDM) PON, the power splitters are used at the remote node. The downstream traffic is broadcast to all the end users while the upstream uses time division multiple access (TDMA) techniques to avoid that several ONUs send traffic simultaneously. To support a large amount of end users, a high splitting ratio of the optical splitter at the remote node is required, which may lead to a power budget problem. Incorporating WDM [5] techniques with PON allows a dedicated wavelength to carry the communication between the OLT and each ONU, leading to much higher capacity on a per-ONU basis compared with the TDM-PON [6]. Therefore, WDM-PON is

Figure 1. The typical communication network hierarchy.

2

1.1. High capacity optical access network

considered one of the candidates to tackle the bandwidth bottleneck of the access network [7], [8]. Coarse WDM (CWDM) and dense WDM (DWDM) are distinguished by the spacing between neighboring wavelengths. DWDM-PON typically refers to the WDM-PON system with less than 100 GHz channel (about 0.8 nm at 1550 nm) [9], [10]. Thanks to the tightly packed wavelength carriers, DWDM-PON is able to fit more than 40 channels in a fiber and can therefore accommodate a large number of clients with a desirable access capacity [11]. However, the narrow wavelength channel spacing means less tolerance to any wavelength drift caused by the optical sources. DWDM-PON is much more susceptible to crosstalk than CWDM-PON where an array of laser diodes typically are deployed as the multi-carrier optical source and each wavelength is individually generated, monitored and controlled. If DWDM-PON adopts the same approach for the carrier generation, all wavelengths need to be precisely controlled to prevent any drift. Otherwise there may be significant crosstalk between adjacent channels. To avoid the impact of frequency grid fluctuation which potentially downgrades the channel transmission quality, a steady multi-carrier source is needed with stable frequency spacing between the wavelengths. Besides, considering the potential network upgrade, which may further increase the number of wavelengths in DWDM-PON, a flexible channel spacing scheme for the multi-carrier generator is favorable.

Apart from the fixed broadband access, the access of mobile users also possesses big challenges to the existing communication networks in terms of high bandwidth. It is forecasted that the mobile data traffic will grow three times faster than the fixed IP traffic [12]. The emerging wireless standards keep expanding the wireless transmission speed [13]. Most of them are working at the microwave range from 2 to 5 GHz (see Figure 2). According to a rule of thumb approximate ly 10% of the carrier frequency is counted as the available bandwidth. Hence, wireless signals transmitted by the carriers lower than 10 GHz hardly reach a data rate beyond 1Gb/s. Driven by the rapidly growing bandwidth demand, the sub-millimeter wave and millimeter wave (MMW) are exploited [14]–[16]. Carrying data on multiple MMW bands further increases the capacity. However, the propagation of MMW is limited by the high loss in the atmosphere. To solve this problem, radio-over-fiber (RoF) technique is widely considered to expand the transmiss ion

Figure 2. The spectrum usage of multiple wireless services. GSM: global system for mobile system; UMTS: universal mobile telecommunication system; WIMAX: worldwide interoperability for microwave access; WLAN: wireless local area network;

UWB: ultra-wideband; LMDS: local multipoint distribution service; MVDS: multipoint video distribution system.

3


distance of the MMWs thanks to the low loss in fiber transmission. The RoF technique transmits the radio frequency (RF) signals from the central office (CO) to the cells by using the fiber infrastructure. In a RoF system, optical-to-electrical conversion is only needed in the cells. The configuration at the cells is thus simplified. However, the generation of multiple-band MMW is still a problem. There are many researches focusing on the multiple-band MMW generation in RoF system [17], [18]. In order to realize multiple-band MMW generation, many of the schemes have complex configuration, either in the cells [17] or in the CO [19], [20]. Besides, most of the existing solutions e.g., [17], [18] carry the same data on different MMW bands, which is not spectrum efficient for high capacity mobile access.

1.2. High capacity optical intra-datacenter network

Over the last few years, the increase of cloud computing, social networking and streaming video calls for more powerful datacenters. The applications requiring intensive interactions within the datacenters need high bandwidth in communication among the servers. On the other hand, running datacenter is energy consuming. The global datacenters consume 3% of the overall energy use, which equals to the yearly amount of energy consumed by Italy or Spain [21]. Due to problems with thermal dissipation, the datacenters cannot afford an energy consumption increase

Figure 3. Diagram of the three-tier datacenter architecture.

4

1.3. Overview of the thesis contributions

proportional to the capacity growth. Consequently, datacenters have to support a dramatica lly growing capacity with a limited increase of energy consumption. Typically, DCN includes several tiers [22], i.e., edge tier, aggregation and core tier, respectively, as shown in Figure 3. Servers in a rack are connected by a top-of-rack (ToR) switch and further communicate with servers in the other racks through aggregation switches and core switches. Optical transport technology, which has been widely adopted in telecommunication networks for its high capacity and low power consumption, is considered as a prominent candidate in the intra-datacenter networks as well. In this thesis, we refer to intra-datacenter networks as datacenter networks (DCNs), which handle the traffic staying inside the datacenters. The networking of datacenters, which takes approximately 20% of the overall datacenter energy consumption [23], is now dominated by electrical switches, becoming the bottleneck of the capacity upgrading and energy efficiency improvement in DCNs. Some works introduced passive optical interconnects (POIs) based on arrayed waveguide gratings (AWGs) [24] and optical couplers [25] into DCNs, achieving remarkable energy saving. However, AWG-based POIs are not flexible in terms of spectrum allocation, and the optical couplers with a large splitting ratio cause high insertion loss, resulting in a scalability problem. Therefore, a proper design of POI architecture is of key importance for the energy efficient DCNs. Moreover, a comprehensive scalability analysis needs to be carried out to understand the maximum size and capacity of the POIs. Apart from the architecture design, physical-layer implementation, including power loss, type of receivers, etc., lead to different levels of transmission performance, and therefore affect the network scalability. For example, as two main streams of advanced modulation techniques in DCN deployment, discrete multi-tone (DMT) and multi-level pulse amplitude modulation (M-PAM) have different impact on transmission performance, which can affect the scalability of the optical interconnects. Besides, a new alternative, i.e., electrical duo-binary modulation (EDB)also gains attention in the short-reach optical communication [26], as it enables the usage of components and packaging technologies with lower bandwidth, as well as the reduction of link cost and power consumption. Moreover, its narrowed spectral bandwidth also improves the chromatic dispersion tolerance and reduces the equalization requirements of the high-speed serial rate link.

1.3. Overview of the thesis contributions

In this thesis, we address several challenges in short-haul optical communications and networking in terms of the design and assessment of optical access networks and DCNs which are summarized as follows.

For optical access network, we propose two optical frequency comb (OFC) generation schemes as the multi-carrier source of the DWDM-PON system to replace the conventional laser array based scheme where individual control and management of each wavelength is required. The feasibility of using such OFC generation schemes in DWDM-PON system is experimenta l ly validated. Besides, the channel spacing of the proposed OFC generation scheme can be flexib ly

5


reconfigured, supporting the potential network upgrade by narrowing the channel spacing, rather than replacing the overall optical sources.

In order to support the future high capacity mobile access, we propose a concept of the palm-shaped spectrum and combine it with RoF techniques to facilitate the dual-band MMW communication. The palm-shaped spectrum is capable of simultaneously generating dual-band MMW signals. Making full use of the MMW spectrum by carrying different data on two bands, the communication capacity of the proposed RoF system is much improved compared with existing schemes, such as [17]–[18]. Besides, the proposed palm-shaped spectrum generation scheme has also an advantage in terms of upstream transmission performance compared with the OFC scheme where all the tones have similar power level.

Regarding the DCNs, a methodology is developed to assess the scalability of the POIs considering the characteristics of the physical layer, e.g., insertion loss, receiver type, receiver bandwidth, modulation format, symbol level spacing and decision thresholds. As case studies, the scalability performance of different POI-based solutions are theoretically analyzed in terms of the maximum size of the interconnect and the highest data rate that can be supported. As an extension of the work, an erbium-doped fiber amplifier (EDFA) is introduced to the POI to increase the power budget and hence the scalability. The improvement of the scalability and the impact of the extra cost and energy consumption caused by the EDFA is evaluated. Furthermore, the factor of employing different modulation formats in the POI is experimentally investigated and thus the scalability performance of POI employing advanced modulation formats are then analyzed

1.4. Thesis organization

The remainder of the thesis is organized as follows.

Chapter 2 presents the background for the short-haul optical communications, which include the fiber access networks, RoF systems, and energy efficient DCNs.

Chapter 3 presents the work included in Paper I and Paper II where two OFC generation schemes are proposed as the multi-carrier source for the DWDM-PONs. Both schemes are evaluated and compared. Furthermore, we implemented the one with the lower driving voltage in the DWDM-PON system, to experimentally verify its feasibility.

Chapter 4 presents the work in Paper III and Paper IV. A concept of palm-shaped spectrum is proposed. Based on it, the RoF system that carries dual-band MMW signals is described and evaluated.

In Chapter 5, a methodology is introduced to evaluate the scalability of optical interconnects in DCNs, which is included in Paper V. The assessment framework considers the fundamenta l physical- layer features, e.g., insertion loss, receiver type and bandwidth, modulation format, the symbol spacing and the impact of thresholds in decision. Based on the scalability analysis methodology, several POI architectures are evaluated as case studies. The maximum interconnect size and the highest data rate that can be supported by the POIs are theoretically calculated in

6

1.4. Thesis organization

Paper V. We evaluate the scalability improvement in Paper VI by introducing a single EDFA in the POI architecture. Several competitive modulation techniques in short-reach communications, including DMT, PAM and EDB are investigated. The related work is included in Paper VII, Paper VIII and Paper IX, respectively. The impact of deploying DMT and PAM on the scalability of the POIs is investigated based on the experimental results of the transmission performance evaluation. This work is included in both Paper VII and VIII.

Chapter 6 concludes the work and discusses some possible future topics related to the work presented in the thesis.

Chapter 7 presents a brief summary of the author’s contribution to each publication included in the thesis.

7

8

Chapter 2

Short-reach optical communication networks

This chapter presents some background related to short-reach optical communication networks. The short-reach optical communication networks considered in this thesis include two categories: 1) fiber access networks, which connect the end users to the edge of aggregation/core networks, typically having transmission distance of tens of kilometers, and 2) intra-datacenter networks (DCNs), which handle the communications within the datacenter with a transmission distance up to a few kilometers.

2.1. Fiber access network

Access networks provide the last mile connection to the Internet for the end users. There are several types of access technologies categorized by the use of different transmission medium, such as copper, wireless (air) and optical fiber. Fiber access network utilizes optical fiber to provide the connectivity and surpasses the others in terms of high capacity and long reach. Typically, a fiber access network consists of an optical line terminal (OLT) located in the CO connected to one or a group of ONUs at the end points with fibers.

2.1.1. Passive optical networks

Passive optical networks (PONs) use only passive components, e.g., wavelength or power splitters and combiners, in the outside plant to interconnect different fiber sections. Such network architectures reduce the cost and energy consumption while improving the reliability performance compared to active optical networks, where active equipment is required in the field. By implementing different multiplexing techniques, various types of PONs can be realized, such as

9

Chapter 2. Short-reach optical communication networks

TDM-PON and WDM-PON. 1 Gb/s TDM-PONs, such as, Gigabit capable PON (GPON) [27] and Ethernet PON (EPON) [28], are widely deployed. 10 Gb/s TDM-PONs, such as XG-PON1 [29] and 10G-EPON [30], are in the early stage of deployment. In TDM-PON, the downstream packets from the OLT are broadcast to all the ONUs by optical splitters and each ONU is assigned a certain time slot. The signals from the OLT suffer from a high insertion loss of the power splitters, especially when the OLT serves a large number of ONUs. It sets a limit to the power budget in the link and thus the reach. Besides, the capacity allocated by a single wavelength is shared among all the ONUs, which makes it difficult to fulfill the future bandwidth requirement of the end users.

WDM PON, as the name reflects, multiplexes several wavelength channels, each of which is assigned to one ONU It is one of the best ways to expand the PON capacity and thus it is gaining wide interest. The first deployment of WDM PON was carried out in Korea 2007 [31]. A typical WDM-PON architecture is shown in Figure 4. A typical WDM-PON architecture.. Wavelength de-multiplexers such as AWGs replace the optical power splitters in TDM-PON. Rather than sharing a common wavelength among 32, 64 or even more ONUs in TDM-PON, WDM-PON allows each subscriber having one or more dedicated wavelengths. Such a feature also enables the flexibility in accommodating different bandwidth requirements of the users. Besides, the power budget problem can be solved in WDM-PON, since an AWG, having much less insertion loss than a power splitter, can be used at the remote node. Hence, a WDM-PON can have a much larger coverage compared to a TDM-PON.

A straightforward way to support more clients and further improve access capacity in WDM-PON, is to decrease the wavelength spacing and thus increase the number of available channels. DWDM-PON is a solution where wavelengths are close to each other, with less than 100 GHz difference [9], [10]. However, the tight channel difference makes the system sensitive to crosstalk, especially when high bandwidth signals are transmitted. The drift of the laser source on any wavelength may significantly downgrade the transmission performance.

OFC composed of multiple frequency tones generated from a single wavelength with equal amplitude and constant spacing is a promising candidate for the multi-carrier source in DWDM-PON system. The conventional OFC generation scheme can be based on mode-locked lasers [32],

Figure 4. A typical WDM-PON architecture.

10

2.1. Fiber access network

supercontinuum optical source [33] and optical fiber nonlinearity effects [34]. However, the control of a mode-locked laser is difficult. The supercontinuum source based methods need complex design. The nonlinear effect based approach requires a long high nonlinear fiber, which is expensive. Besides, high input optical power is needed in the OFC generation based on nonlinearity effects. With the development of LiNbO3 crystal waveguide based electro-optic modulator, OFC generation involving Mach-Zehnder modulators (MZMs) becomes attractive [35], [36]. The channel spacing between any of the neighboring tones in these approaches is locked to the driving radio frequency (RF) which is stable. However, the approach proposed in [35] requires tailored driving RF and the one proposed in [36] needs extra high dispersion media to generate OFC. Both of them are expensive.

2.1.2. Radio-over-fiber systems

It is envisioned to have an interconnected world where everything and everyone are associated to each other from anywhere. However, wireless signal significantly degrades with the increase of the signal propagation distance. It is therefore widely deployed for the backhaul, exchanging the radio frequency (RF) signals between the central office (CO) and the cell sites, as shown in Figure 5.

The difference between a RoF link and a conventional digital baseband optical link[37] for the transmission of RF signals is shown in Figure 6. In the conventional case, the baseband signal is directly converted to optical spectrum before the fiber transmission by electrical-to-optical (E/O) converter. At the base station (BS), signal processing, including, but not limited to, demodulat ion and modulation of the baseband signal is implemented before amplifying and broadcasting it through the air. In the RoF system, the wireless signal is generated by a RF up-conversion module and then converted to optical signal by E/O conversion module. Using RoF technique, the configuration at cell sites is simplified due to the fact that baseband signal processing is not needed at the cell site anymore.

Figure 5. The basic diagram of a RoF communication system.

11


In order to meet the increasing bandwidth requirement of wireless communication, the use of the MMW band is attractive. The MMW band covers the frequency range of 30 GHz to 300 GHz (corresponding to the waveband of one to ten millimeters), which has a potential to provide high capacity for wireless access. 60 GHz band, for example, has already been standardized for short range wireless applications [38], [39], and even higher frequency bands that are over 100 GHz also attract a lot of attention [40]. However, the MMW band suffers higher propagation attenuation through the atmosphere than the in-use operating bands of mobile service providers, which limits the typical transmission distance of MMW to below 100 meters. This makes it attractive to use RoF technology to distribute the MMW signal to the antennas. [41].

There are many studies concentrating on MMW based RoF schemes. In [17] the authors employ a single Mach-Zehnder modulator (MZM) in the CS and heterodyne mixing technique to realize multi-band MMW signal generation by using an optical local oscillator (LO) at the cell, resulting in a complex system configuration. Paper [19] proposes multi-band signal generation with dual-drive modulator and two separate clock signals, requiring high frequency synthesizer. Optical frequency comb (OFC) with flat spectrum and equal frequency spacing is an alternative approach to generate multi-band MMW [18]. In this scheme only one comb line is allocated to carry downstream data while the other lines are used for beating to generate multiple bands of MMWs. However, it should be noted that in most of the existing schemes, e.g., [17]–[19], different MMW bands are modulated with the same data, resulting in inefficient spectrum utilization.

2.2. Intra-datacenter networks

Datacenters are becoming increasingly important in today’s world. Many of online activit ies including social media, email, web searching, online consumer services, which make our society more productive and efficient, are supported by the datacenters. The data-intensive applications creates, as mentioned, a big challenge when it comes to the energy consumption. Optical

Figure 6. Diagram of RF signal transmission on (a) digital baseband optical link and (b) RoF.

12


communication, which is characterized by high capacity, low loss and high energy efficiency, is playing a key role in delivering a green networking solution for datacenters. This trend leads to a high priority of focusing on the energy consumption problem for the intra-datacenter networks, which are referred as datacenter networks (DCNs) in this thesis.

2.2.1. Energy consumption in datacenters

The prospect of the future datacenter systems from the aspect of computation capacity, bandwidth requirement and power consumption is listed in Table 1. The peak performance and bandwidth demand of datacenters are expected to increase rapidly, however, the allowable growth rate of power consumption is limited by two [42], [43] in every four years. To address the power consumption problem in the datacenters, the first thing to figure out is how the power consumption is distributed. According to [41], see Figure 7, in a mega datacenter the network consumes approximately 30% of the total IT energy, while servers and storage obsess the rest 70% [44]. Among the subcategories of networking, access switches at the top-of-rack (ToR) occupy half of

Figure 7. Energy consumption distribution in a datacenter [44]

TABLE 1 Performance, BW requirements and power consumption bound for future computing system [42], [43]

Year Peak Performance Bandwidth requirements

Power Consumption bound

2012 10 PF 1 PB/s 5 MW 2016 100 PF 20 PB/s 10 MW 2020 1000 PF 400 PB/s 20 MW

*PF: Peta floating point operations per second

13


the overall switching power. Therefore, to reduce the power consumed at ToR significantly is important in the design of energy efficient DCNs.

2.2.2. Characteristic of datacenter traffic

A clear understanding of the main features of the traffic generated by the servers is crucial to the design of an efficient DCN architecture. The traffic characteristics vary for different applications while sharing some common features as explained in below:

• Traffic flow locality: The traffic flow locality describes the portion of the datacenter traffic that stays locally. If the traffic is directed to another server in the same rack, it is referred as intra-rack traffic, otherwise it is inter-rack traffic. The ratio of these two types of traffic varies among different types of datacenter. Specifically, in the cloud computing datacenters up to 80% of the traffic is intra-rack traffic, which is handled by the ToR switches. In such datacenters, reducing the energy consumed at the edge tier is crucial to address the thermal dissipation problem of DCNs due to the large amount of intra-rack communications.

• Traffic size and duration: Most of the traffic flows between two servers are considerably small in size and last under a few hundreds of milliseconds. In order to ensure the low latency of the traffic small reconfiguration overhead is required for the intra-datacenter communication.

• Bursty traffic: The intra-rack traffic is very bursty while the traffic in aggregation and core tier tends to be less bursty. Such a feature leads to diverse requirements for the design of different tiers in DCNs.

• Concurrent traffic flows: In most of the datacenters the average count of concurrent flow per server is around 10 [23]. It would be favorable to take the multi-cast feature into account in the design of the DCN architecture.

2.2.3. Energy efficient datacenter network architectures

As it was mentioned previously, current commodity Ethernet switches within the datacenters may not sustain the expected traffic increase with acceptable energy consumption [45], which in turn brings a need for high capacity and energy-efficient solutions. On the other hand, optical transmission technology, which has been widely adopted in telecommunication networks for its high throughput, short latency and low power consumption, can be a prominent solution in DCNs as well [44]. Switches at different tiers handle the traffic among the servers, racks and clusters. There have been many optical interconnects based on optical switching for DCNs investigated in the literature. Hybrid optical/electronic solutions, e.g., HELIOS [46], and C-through [47] have been proposed to increase the capacity and reduce the power consumption. In these approaches, optical switching is used to deal with big flows but electrical switches are still kept to handle the traffic at the packet level. Studies on all-optical interconnect architectures at the aggregation or core tier have also been carried out [48]–[50]. In [48], optical packet switching is adopted to offer small switching granularity but requires O/E and E/O conversions for buffering. Micro-electro-mechanical system (MEMS) based optical interconnects proposed in [49] need long connection

14


configuration time and are not efficient for bursty traffic patterns. In [50], a hybrid spatial division multiplexing (SDM) and TDM network is introduced to support large-capacity DCNs. However, the high cost of multi-element fiber (MEF) prevents them to be used for interconnecting racks. Moreover, in order to realize the communication between two racks in different datacenters across the metro network, a converter between SDM and WDM is required for the approach, which is still not commercially available.

One way to improve the datacenter energy efficiency is to reduce the usage of electrical components. Many existing DCN architectures are based on optical switching [48]–[50]. In most of these designs, optical switches replace the electrical ones in the aggregation and core tier. However, the energy saved by introducing optical switching in higher levels in datacenter is limited [25]. On the other hand, up to 90 % of overall energy for switching is consumed by the ToR switches due to the large number of racks in the datacenter [51]. Therefore, reducing the ToR switching energy consumption should be given the priority to improve the datacenter energy efficiency. There are some works focusing on passive optical interconnect (POI) at ToR [24], [52]. Compared to the switch-based architectures, the using of passive components, e.g., optical couplers to interconnect the ports has inherent advantages from the aspects of cost, energy consumption and reliability. Besides, due to the fact that no hardware reconfiguration is needed, the POI architectures has the potential to have a good performance in terms of latency. The scalability of coupler-based POI is analyzed in [52] only considering the insertion loss, which is essential but not sufficient. Different physical layer features, including but not limited to, such as receiver type, operation bandwidth, modulation formats, modulation and decision schemes, should be taken into consideration in the scalability assessment.

2.2.4. Advanced modulation formats applied for intra-datacenter optical links

Because of the large amount of transponders, DCN is very sensitive to cost and power consumption of the transceivers. Intensity modulation and direct detection (IM/DD) techniques are popular in short-reach communications in DCNs mainly because of its simplicity and low-cost deployment compared to the coherent solutions. In IM/DD approaches, the bit rate per wavelength is the product of two fundamental parameters: symbols per second (Baud rate) and bits per symbol (modulation order). Binary intensity modulation, i.e., on-off-keying (OOK) is the simplest IM/DD technique. However, it requires electro-optical components with high bandwidth, especially when a high data rate is required. Using advanced modulation formats, i.e., increasing the modulat ion order is a way to provide high capacity. As a result, M-PAM [53]–[56] and DMT [57], [58] over the low-cost and low-energy integrated transceivers become the two mainstreams in Datacom.

M-PAM refers to multi- level intensity modulation [59], for instance, 4-PAM and 8-PAM refers to the modulation with 4 and 8 amplitude levels, respectively. Compared to binary modulation, e.g. OOK, the spectral efficiency of M-PAM is improved by a factor of log2𝑀𝑀 when transmitting at the same data rate.

15


DMT is a multi-carrier technique, which is a special implementation of orthogonal frequency division multiplexing (OFDM). DMT slices the spectrum, as shown in Figure 8, where each of the subcarriers occupies only a fraction of the channel bandwidth. The bit and power allocation for each subcarrier depends on the channel response at the corresponding subcarrier frequency. In other words, in DMT, the subcarriers with high signal-to-noise ratio (SNR) can be allocated with high order modulation formats, while the subcarriers having low SNR are assigned low order modulation formats, leading to the high utilization of the spectrum.

There are some works [60]–[64] on the comparison of M-PAM and DMT in terms of energy consumption, performance, etc., which are summarized in Table 2. In general, M-PAM has lower implementation complexity and lower power consumption while DMT has obvious advantages in terms of spectrum utilization and robustness in case of non-flat channel response.

Table 2 The comparison between M-PAM and DMT [60]–[64]

M-PAM DMT

Implementation complexity + O

Power consumption ++ +

Spectrum utilization + ++

Robustness O ++

++: very good; +: good; O: mediu

(a) (b) (c)

Figure 8. (a) the subcarriers of a DMT signal; (b) channel response of the transmission link; (c) power and bit allocation for the subcarriers according to the channel response.[63].

16

Chapter 3

Optical frequency comb based DWDM-PON

As one of the promising candidates to meet the high bandwidth demand of the access network, WDM-PON has been widely investigated for many years. A laser array consisting of mult ip le individual laser diodes is conventionally considered as the multi-carrier source in WDM-PON. However, the wavelength drift and the complex control of the individual lasers adversely affect the system performance, especially in DWDM-PON where the channels are closely spaced and consequently more susceptible to crosstalk.

In this chapter, we focus on DWDM-PON and aim at providing stable multi-carrier source for DWDM-PON. This chapter starts with an introduction of Mach-Zehnder modulator (MZM), based on which two optical frequency comb (OFC) generation schemes are proposed in Paper I and Paper II, respectively. Both schemes are able to provide multiple carriers with variable frequency spacing. Such a feature makes the DWDM-PON system flexible where channel spacing can be adjusted according to the request. Particularly, it can be a cost-efficient solution for system upgrade, in which bandwidth capacity can be improved by narrowing the spacing among the carriers to increase the supported number of end points. In Paper II, the proof of concept experiment is carried out for validation of the OFC based DWDM-PON system.

3.1. Mach-Zehnder modulator

An MZM has two waveguide interferometer arms [65], as shown in Fig. 9. The input optical power is split into two halves traveling along the waveguide. The refractive index of the electro-optic crystal based waveguide can be adjusted by applying voltage on the arm(s). When the waves

17

Chapter 3. Optical frequency comb based DWDM-PON

from the two arms are recombined, the phase difference between the arms leads to an interference effect that can be used for amplitude modulation.

Biased by DC current 𝑉𝑉𝐷𝐷𝐷𝐷 and a sinusoidal modulation signal 𝑉𝑉𝑚𝑚 cos(𝜔𝜔𝑅𝑅𝑅𝑅𝑡𝑡) where ω𝑅𝑅𝑅𝑅 is the angular frequency of the sinusoidal signal, the applied driving signal is the superposition of the two terms,

𝑉𝑉(𝑡𝑡) = 𝑉𝑉𝐷𝐷𝐷𝐷 + 𝑉𝑉𝑚𝑚cos (ω𝑅𝑅𝑅𝑅t). (3.1)

The output optical field 𝐸𝐸𝑜𝑜 can be expressed as:

𝐸𝐸𝑜𝑜(𝑡𝑡) = 𝐸𝐸𝐼𝐼 [𝑐𝑐𝑐𝑐𝑐𝑐πV (t)2𝑉𝑉𝜋𝜋

]cos(ω0t), (3.2)

where 𝐸𝐸𝐼𝐼 is the amplitude of the input optical field, ω0 is the angular frequency of the input optical field, and 𝑉𝑉𝜋𝜋 is the half wave voltage and equal to the required voltage to change the output light intensity from its minimum value to the maximum value.

Using (2.1) and (2.2) we can get

𝐸𝐸𝑜𝑜(𝑡𝑡) = 𝐸𝐸𝐼𝐼 �𝑐𝑐𝑐𝑐𝑐𝑐π𝑉𝑉𝐷𝐷𝐷𝐷 + 𝑉𝑉𝑚𝑚 cos(ω𝑅𝑅𝑅𝑅t)

2𝑉𝑉𝜋𝜋� cos(ω0t)

= 𝐸𝐸𝐼𝐼 �𝑐𝑐𝑐𝑐𝑐𝑐𝑉𝑉𝐷𝐷𝐷𝐷2𝑉𝑉𝜋𝜋

cos � 𝑉𝑉𝑚𝑚2𝑉𝑉𝜋𝜋

cos(ω𝑅𝑅𝑅𝑅t)� − 𝑐𝑐𝑠𝑠𝑠𝑠 𝑉𝑉𝐷𝐷𝐷𝐷2𝑉𝑉𝜋𝜋

sin � 𝑉𝑉𝑚𝑚2𝑉𝑉𝜋𝜋

cos(ω𝑅𝑅𝑅𝑅t)�� cos (ω0t). (3.3)

Using the Bessel series expansion of the trigonometric functions,

sin(𝑥𝑥𝑐𝑐𝑐𝑐𝑐𝑐𝑥𝑥) = −2∑ (−1)𝑛𝑛𝐽𝐽2𝑛𝑛−1(𝑥𝑥)sin [(2𝑠𝑠 − 1)𝑥𝑥]∞𝑛𝑛=1 ,

cos(𝑥𝑥𝑐𝑐𝑐𝑐𝑐𝑐𝑥𝑥) = 𝐽𝐽0(𝑥𝑥) + 2∑ (−1)𝑛𝑛𝐽𝐽2𝑛𝑛(𝑥𝑥)cos (2n𝑥𝑥)∞𝑛𝑛=1 , (3.4)

where 𝐽𝐽𝑛𝑛 is the Bessel function of the first kind of order n.

Setting 𝑚𝑚 = 𝑉𝑉𝑚𝑚2𝑉𝑉𝜋𝜋

, the output optical field becomes

Figure 9. Internal configuration of Mach-Zehnder Modulator devices

18

3.2. Optical frequency comb generation

𝐸𝐸𝑜𝑜(𝑡𝑡) = 𝐸𝐸𝐼𝐼𝑐𝑐𝑐𝑐𝑐𝑐𝑉𝑉𝐷𝐷𝐷𝐷2𝑉𝑉𝜋𝜋

�𝐽𝐽0(𝑚𝑚)𝑐𝑐𝑐𝑐𝑐𝑐(ω0t)

+ �[𝐽𝐽2𝑛𝑛(𝑚𝑚) cos(ω0t + 2nω𝑅𝑅𝑅𝑅t − n𝜋𝜋) + 𝐽𝐽2𝑛𝑛(𝑚𝑚) cos(ω0t − 2nω𝑅𝑅𝑅𝑅t + n𝜋𝜋)]∞

𝑛𝑛=1

�

+𝐸𝐸𝐼𝐼𝑐𝑐𝑠𝑠𝑠𝑠𝑉𝑉𝐷𝐷𝐷𝐷2𝑉𝑉𝜋𝜋

��[𝐽𝐽2𝑛𝑛−1(𝑚𝑚) cos(ω0t + (2n− 1)ω𝑅𝑅𝑅𝑅t − n𝜋𝜋)∞

𝑛𝑛=1

+ 𝐽𝐽2𝑛𝑛−1(𝑚𝑚) cos(ω0t − (2n− 1)ω𝑅𝑅𝑅𝑅t + n𝜋𝜋)]�

(3.5)

Take the quadrature bias point as an example, i.e., 𝑉𝑉𝐷𝐷𝐷𝐷 = 𝑉𝑉𝜋𝜋/2, (2.5) becomes

𝐸𝐸𝑜𝑜(𝑡𝑡) = 𝐸𝐸𝐼𝐼 ��[𝐽𝐽2𝑛𝑛−1(𝑚𝑚) cos(ω0t + (2n− 1)ω𝑅𝑅𝑅𝑅 t − n𝜋𝜋)∞

𝑛𝑛=1

+ 𝐽𝐽2𝑛𝑛−1(𝑚𝑚) cos(ω0t− (2n− 1)ω𝑅𝑅𝑅𝑅t + n𝜋𝜋)]� , (3.6)

which means the fundamental and odd harmonics are suppressed while the even order sidebands are excited. The amplitude of the harmonics is proportional to the value of the corresponding Bessel functions associated with the modulation index m.

By adjusting the bias voltage and m, different output spectrum can be obtained. Note that, overdriving the MZM, i.e., 𝑉𝑉𝑚𝑚 > 𝑉𝑉𝜋𝜋/2 , can result in the harmonics of different orders.


Optical frequency comb (OFC) is referred to as an optical spectrum consisting of mult ip le equidistant spectrum lines [66]. Introducing OFC in DWDM-PON as the multi-carrier source significantly stabilizes the light source thanks to the locked channel spacing among each wavelength tone. It also helps to simplify the control and the management of the multi-carr ier source and hence reduce the system cost comparing to the conventional way in which individua l lasers form a laser array to generate multiple carriers and need to be controlled separately.

19


There are several performance indicators to measure the quality of an OFC generation scheme. Firstly, high optical sideband suppression ratio (OSSR) of the generated OFC is needed to satisfy the high capacity data transmission. Secondly, the OFC needs to be as flat as possible. The smaller power difference among carriers facilitates the better signal modulation and demodulation. Last but not least, the OFC generation scheme needs to be adjustable in terms of central wavelength and the channel spacing. The flexible configuration of the multi-carrier light source can easily support the system upgrade and reconfiguration at low cost.

In Paper I and II, we experimentally demonstrate two flexible OFC generation schemes by deploying LiNbNO3 based MZMs. The generation scheme proposed in Paper I consists of a continuous wave (CW) laser, a phase modulator (PM), an intensity modulator (IM) and an optical bandpass filter, as shown in Figure 10 (a). Two RF sources oscillating at different frequencies are required to drive the PM and the IM at each stage respectively. The driving signal frequency at each stage determines the channel spacing of the output OFC while the high modulation power excites the high order harmonics. The phase modulator working at the first stage is overdriven by 25 GHz RF signal with significant high power to generate OFC with channel spacing of 25 GHz

Figure 10. (a).The block diagram of the optical frequency comb generation scheme based on cascaded phase modulator and intensity modulator, (b) optical frequency comb lines with channel spacing of 8.33 GHz

20


followed by the IM. To demonstrate the switching ability of the OFC generation scheme, we apply either 12.5 GHz or 8.33 GHz sinusoidal signals to the IM by, in order to obtain dense spectrum lines with different frequency spacing. The channel spacing can be adjusted by changing the RF at each stage. With appropriate bias voltage, the harmonic sidebands amplitude is flattened. In Paper I, 50 OFC lines with 12.5 GHz or 8.33 GHz frequency with less than 3 dB spectral power variation are obtained, among which the 50 OFC lines with channel spacing of 8.33 GHz are shown in Figure 10 (b).

However, the driving voltage required in this method is 26.2 dBm in the first stage and 19.6 dBm in the second stage. Such high driving voltages are very likely to exceed the maximum voltage of the modulators. To avoid overdriving the modulators, we came up with the second OFC generation scheme, which is presented in Paper II.

In Paper II, a modified two-stage generation scheme of flat OFC is proposed and experimentally demonstrated. The generation scheme is based on a dual-parallel Mach-Zehnder modulator (DPMZM) followed by an IM. The system configuration is illustrated in Figure 12. The DPMZM is working in a push-and-pull scheme. With the two odd order sidebands from the upper-arm and the even order sidebands from the lower-arm, an OFC with 5 tones is generated. The higher order sidebands are not excited because of the small driving voltage applied in the scheme. After the first-stage generation, an IM is deployed for producing more spectral lines with shrinking frequency spacing. The output OFC channel frequency spacing can be adjusted by tuning the frequency of the driving RF-signal. We simulated the proposed system using VPItransmissionMaker™ Optical Systems (v.9.0) and carried out experiments to validate it. As shown in Figure (a), 5 flat comb lines with spacing of 25 GHz are generated after the first-stage DPMZM and 15 flat comb lines with equal power are obtained after the second-stage DPMZM, see Figure (b). The experimental results agree with the theoretical analysis and simulation. With

CWBias1Bias2

f

f

f

f

Bias3

1/3 Frequency divider

3-way PS

RF Amplifier3

Amplifier1

Amplifier2

Figure 11. The generation scheme of the OFC based on cascaded DPMZM and IM.

21


the unwanted sidebands suppressed, 1.4 dB power variation and over 15 dB in-band OSSR are obtained among all the 15 comb lines at the output of the second-stage DPMZM, as shown in Figure (c) and (d). The OFC generation can be scaled up by employing multi-wavelength arrays instead of the single wavelength CW source. A 4-laser array is capable of offering 60 comb lines with equal frequency spacing and peak-to-peak power. In this OFC generation solution, the employment of monolithic device, i.e., the DPMZM, improves the stability of the generation system in contrast to the one using separate modulators. Compared to another scheme based on DPMZM [67], we can save two driving sources and thus reduce the system cost.

3.3. DWDM-PON based on OFC

The OFC generation schemes proposed are perfect to be the light sources of a DWDM system due to the flexible channel spacing, simple configuration and low cost. On the other hand, orthogonal frequency division multiplexing (OFDM) is attractive for the access network because of its high spectral efficiency and resistance to the channel impairment [68], [69]. Besides, OFDM is widely applied in the 4G mobile network and is expected to be used in the future 5G as well. The investigation of DWDM-PON system carrying OFDM signals is important for the seamless

Figure 12. (a). The simulated spectrum of 5 comb lines with 25 GHz channel spacing at the first stage; (b). the

simulated spectrum of 15 comb lines with 8.33 GHz channel spacing at second stage; (c) the experiment result of 5 comb lines with 25 GHz channel spacing at the first stage; (d) the experiment spectrum of 15 comb lines with 8.33

GHz channel spacing at second stage.

22

3.3. DWDM-PON based on OFC

integration of the wired-and-wireless communication system. In Paper II, we propose a DWDM-PON architecture based on the OFC for OFDM signal transmission. An OFC generator is employed at the OLT working as the multi-carrier source (see Figure 13). The comb lines are divided into individual DWDM channels by the DEMUX. Each carrier is modulated by an IM with the downstream 10 Gb/s quadrature phase shift keying OFDM signals for the corresponding ONU. All the modulated carriers are then multiplexed by the MUX and sent to the feeder fiber. At the ONU side, the received optical power is divided into two parts. One part is used for direct detection of the downstream OFDM signal while the other half of the optical carrier is re-modulated by 2.5 Gb/s OOK signal and sent back to the OLT for the upstream transmission. The reuse of the carrier eliminates the need of accommodating lasers in the ONU, which helps to reduce the system cost. In the experimental validation, the downstream OFDM signal can be received with the BER less than 3.8E-3 and error-free OOK transmission over 20km SMF is also realized for the upstream.

Figure 13. The OFC based DWDM-PON architecture.

Comb

OFDM

RX

IM

DD

Feeder fiber ONU

MUX

MUX/DEMUX

OOK

Upstream

Downstream

EDFA

OLT

23

24

Chapter 4

Palm-shaped spectrum based RoF system

The exponentially growing mobile traffic volume drives the exploring of new technology for providing large bandwidth in mobile networks. MMW is capable of providing several GHz of available bandwidth and is hence a promising candidate to satisfy the high capacity demand. Multiple bands of MMW can further expand the wireless communication capacity. Generation and transmission of multiple bands of MMW are widely investigated, e.g., [17]–[19]. However, most of the existing approaches do not support to carry various data flow on different MMW bands, resulting in inefficient spectrum utilization. The solutions that are able to make full use of the MMW spectrum require rather complex system configurations. In Paper III, we propose and experimentally validate a generation scheme for multi-band MMW signals. Paper IV further extends Paper III with the assessment of the overall RoF system where the two MMW bands are efficiently used by carrying separate data. Both the downstream and upstream transmiss ion performance are evaluated when different MMW bands are adopted in the RoF system.

4.1. Palm-shaped spectrum generation.

In Paper III, we proposed a concept of palm-shaped spectrum which is capable of simultaneous ly generating dual-band MMWs. The central carrier looks like the middle finger with higher amplitude than the other four fingers that have equal power. The generation scheme for the RoF system is illustrated as Figure 14. The palm-shaped spectrum generation scheme consists of a CW laser and a DPMZM. Both arms of the DPMZM are properly biased while the PM is adjusted so that both arms are being modulated in-phase. The upper-arm works at the carrier suppressed mode and the lower-arm is biased near the maximum point of the transmission curve. By applying a RF-

25

Chapter 4. Palm-shaped spectrum based RoF system

signal with frequency fc on both arms, the two first order sidebands with a suppressed central carrier are generated by the upper-arm while in the lower-arm, the first order sidebands are suppressed and the second order sidebands as well as the central optical carrier are induced. The driving voltages of the DPMZM are configured to enable the modulation index of the upper-arm m1, and lower-arm m2 satisfying the following relationship:

𝑱𝑱𝟐𝟐(𝒎𝒎𝟐𝟐) = 𝑱𝑱𝟏𝟏(𝒎𝒎𝟏𝟏) (4.1)

and

𝑱𝑱𝟎𝟎(𝒎𝒎𝟐𝟐) > 𝑱𝑱𝟐𝟐(𝒎𝒎𝟐𝟐) (4.2)

where 𝑱𝑱𝒏𝒏(.) is Bessel function of the first kind of n-th order. The combination of the outputs of both arms composes the palm-shaped spectrum.

The generation scheme of the palm-shaped spectrum and the OFC (see in Figure 12 (a)) are quite similar. However, the generation of OFC poses a stringent requirement on the modulat ion indexes, i.e., 𝐽𝐽0(𝑚𝑚2) = 𝐽𝐽2(𝑚𝑚2) = 𝐽𝐽1(𝑚𝑚1) , resulting in complex operation and configuration in practice.

4.2. Dual-band MMW RoF system using a palm-shaped spectrum

The palm-shaped spectrum generation scheme can produce dual-band MMW signals for the downstream, and reuse the middle finger channel for the upstream transmission. The system schematic diagram of the RoF system based on palm-shaped spectrum is illustrated in Figure 14. For the downstream data, i.e., Data1 and Data2 are mixed with the driving RF fc thus modulated on the two MMW frequencies respectively. At the cell, two cascaded FBGs are employed to separate

~Data1

fcfc

RxCentral Office Cell

FBG1

FBG2MZM

MZM

MZM

fc MMW1Rx 1

MMW2Rx 2

Antenna

2 fc4 fc

Upstream signal

PM

EDFA

Data2

PC

Figure 14. Palm-shaped spectrum generation scheme for RoF system. FBG: fiber Bragg grating, PM: phase

modulator, EDFA: Erbium-doped fiber amplifier, PC：polarization controller

26

4.2. Dual-band MMW RoF system using a palm-shaped spectrum

the first and the second order sidebands successively, i.e., the four spectrum lines for beating to generate signals in two MMW bands. High speed PDs are adopted to convert the MMW signals from optical to electrical field. In our proposed scheme, the configuration in the CO is simple because only one DPMZM is required to generate and modulate signals in two MMW bands. Comparing to the existing dual-band solutions based on DPMZM [20], we simplify the system configuration and reduce the insertion loss in the CO by cutting down the number of the needed

Figure 15. BER performance and eye diagram measured at different points with fiber transmission: (a)

downstream signal of 40 GHz MMW (Data2), (b) 80 GHz MMW (Data1) and (c) upstream in case of fc= 20 GHz, and (d) downstream signal of 60 GHz MMW (Data2), (e) 120 GHz MMW (Data1) and (f) upstream in

case of fc = 30GHz.

27

Chapter 4. Palm-shaped spectrum based RoF system

optical devices. Moreover, a high spectral efficiency is realized in our system since both MMW bands are assigned for different data delivery. The spectral efficiency provided in our system is doubled compared to other methods that use different MMW carrier bands to transmit the same data [17], [18]. Regarding the upstream data, the central optical carrier is modulated and sent back to the CO. Considering OFC as a benchmark, the experimental results reveal that the palm-shaped spectrum has an advantage in supporting upstream transmission, thanks to its higher OSSR of the central carrier. Besides, the high power central carrier in the palm-shaped spectrum does not affect the MMW generation. It is because the sidebands in both palm and OFC spectrums possess a similar level of optical power, which determines the quality of the MMW signals.

We carried out simulations with VPItransmissionMaker™ Optical Systems (v.9.2) to validate the feasibility of the dual-band MMW RoF system based on the generated palm spectrum. The simulation setup is the same as the diagram shows in Figure 14. The DPMZM is configured with the driving voltage of 0.087 and 0.47 times of the half-wave voltage of the modulator, which is much lower than the driving voltage required in [23]. Different data (e.g., Data1 and Data2 in Figure 14) are mixed with the driving RF and modulated on the upper- and lower-arm of the DPMZM. 2.5 Gb/s OOK signal is adopted for both bands. At the cell, FBG1 with 80% reflection ratio and 3 dB bandwidth of 0.2 nm is used to extract the middle finger of the palm. The reflected signal then passes FBG2 with 80% reflection ratio and 3 dB bandwidth of 0.64 nm. Two first order sidebands pass through the FBG2 where the second order of the sidebands are reflected. High speed PDs are used to detect the generated dual-band MMW signals. In the simulation, a driving frequency fc of 20 GHz (or 30 GHz) is applied. As a consequence, the achieved dual-band MMWs at the cell are 40 GHz and 80 GHz (or 60 GHz and 120 GHz). The transmission distances in the two cases are 25 km (19 km), which are the maximum reach determined by the fiber dispersion. The downstream transmission performance can be seen in Figure 15 (a)–(d). The higher frequency of the MMW is, the larger power penalty is observed.

Regarding the upstream transmission, 2.5 Gb/s OOK signal is modulated on the middle finger carrier of the palm. Thanks to its high OSSR, negligible degradation is observed on the central carrier (see Figure 15 (e) and (f)).

28

Chapter 5

Scalability analysis of passive optical interconnects in DCN

Another important application area for short-reach optical communications is datacenter networks (DCN). Compared to commodity switches based DCNs, POIs have the inherent advantage of low energy consumption thanks to the use of passive components for interconnect. In addition, the switching overhead in POIs can be much reduced in contrast to the one of optical switching based solutions which includes long reconfiguration time. Compared with the AWG based POIs [24], optical coupler-based POIs [25], [51], [70] have more flexibility in wavelength/spec trum allocation. However, the high splitting loss of the optical coupler may result in a scalability problem. There are several research works focusing on the scalability assessment for the switch-based DCN architectures, such as assessing the scalability by counting the ports of the switch [49], [50], or associating with the control and management efficiency [71], [72]. In contrast to the switch-based architectures, coupler-based POIs suffer from high insertion loss and thus have strict requirements on the system power budget. Physical- layer impairment becomes the key factor to limit the scalability of the POIs. As a result, the aforementioned evaluation methods, which are used for switch-based DCN architectures [48], [49], [71], [72], are not appropriate for POIs. The authors of [52] have assessed scalability of a coupler-based POI architecture, where the insertio n loss of the couplers is the only considered factor, neglecting other physical- layer characterist ics that may affect the accuracy of the scalability assessment.

In this chapter, we present a physical- layer evaluation methodology for scalability analysis of the POIs in DCNs. Based on the developed framework, the scalability of POIs in DCN possessing

29

Chapter 5. Scalability analysis of passive optical interconnects in DCN

different physical- layer features is theoretically and experimentally evaluated. Several coupler-based POI architectures are presented as case studies which are included in Paper V. In order to improve the scalability of the POIs, introducing EDFA to increase the system power budget is an option. In Paper VI, the impact of the extra EDFA is analyzed in terms of the system cost, energy consumption, as well as the DCN architecture scalability. Besides, deploying different modulat ion techniques can also affect the scalability of the considered POI architectures. There are two mainstreams of advanced modulation for short-reach optical interconnect in both academia, and industry, i.e., M-PAM and DMT. To investigate the potential impact of advanced modulat ion formats, we first experimentally assess the transmission performance of the short-reach optical link deploying M-PAM and DMT in Paper VII and then carry out a scalability analysis in the context of the POI in Paper VIII. Duo-binary modulation, as another new alternative for high speed short-reach optical interconnect is also investigated in terms of transmission performance, which is included in Paper IX.

5.1. Scalability assessment methodology

The scalability evaluation framework is developed considering the fundamental impairments in the signal transmission, e.g., the power loss and the receiver noises. In this work, we focus on the deployment of SMF integrated with 1550 nm band, which is widely accepted for the future high capacity short-reach applications [73]. Given the short transmission distance in the DCNs,

Figure 16. Flow chart of the scalability assessment methodology.

Receiver sensitivity

Input parameter:• Receiver type: PIN/APD；• Receiver bandwidth；• Modulation format；• Modulation/Decision scheme• Allowed BER level；

System power budget (PB)

Input parameter:• Size of the interconnect N；• Insertion loss of all the

components included in the link；

Link Loss

Link budget(LB)

Power margin

PB ≥ LB

Feasible

Transmitter output power

Yes

Not feasible

No

30


especially for small-scale ones, e.g., intra-rack communications which have typically less than 10 m propagation distance, it is reasonable to neglect the impairment caused by chromatic dispersion and fiber nonlinearity. The procedure of the developed scalability assessment methodology is illustrated in Figure 16. The process can be divided into two subsections. One is to calculate the link budget which consists of the minimum power loss in the link and the margin reserved for other impairment compensation. The other part is to figure out the system power budget, which is equal to the difference between the launch power of the transmitter and the receiver sensitivity. Here the receiver sensitivity means the minimum power that is required by the receiver to reach a given bit error ratio (BER) level. There are many factors that affect the receiver sensitivity, such as the type of the receivers, modulation formats of the signal, symbol spacing and the decision thresholds, etc., which are covered in the proposed scalability evaluation methodology.

Given the size of a certain POI architecture, the feasibility of such interconnect can be demonstrated under the condition where the link budget is not exceeding the system power budget. The work presented in this section is included in Paper V.

5.1.1. Link budget model

The equivalent block diagram for the coupler-based optical interconnects is shown in Figure 17. As mentioned before, it is reasonable to neglect the fiber-related impairment in the link because of the short transmission distance, and assume that the signal deterioration is dominated by the decrease of the SNR due to power loss. The link budget, which is the minimum power needed to cover the link loss and other impairment, can be expressed as

where ILi, stands for the insertion loss of each component that the optical signal passes. In the coupler-based POI architectures, the splitting loss sets a lower bound of the coupler insertion loss [74]. Given an optical coupler with n ports, the splitting power loss due to passing the coupler once is 10log10n dB. Apart from the insertion loss, a certain power margin, Pmargin, should also be reserved in the link budget calculation for managing loss variations.

Link Budget = 𝐼𝐼𝐼𝐼total + Pmargin = ∑ 𝐼𝐼𝐼𝐼𝑠𝑠𝑠𝑠𝑠𝑠=0 + 𝑃𝑃margin , (5.1)

Figure 17. Equivalent block diagram for the optical interconnect link. Tx: transmitter, BPF: band pass filter, PD:

photo diode, Rx: receiver.

31


5.1.2. System power budget

As illustrated in Figure 16, the power budget can be obtained based on the values of the transmit ter launch power and the receiver sensitivity. For a commercial transmitter, the output power is typically a given value. The receiver sensitivity is defined as the minimum input power to the receiver that is able to satisfy the signal transmission under a given BER level for the used modulation format. The receiver noise can significantly affect the transmission performance which needs to be analyzed for accurate assessment of the POI scalability. The two types of photodiodes (i.e., positive- intrinsic-negative (PIN) diode and avalanched photodiode, APD) at the receivers are considered in the analysis.

When the optical power𝑃𝑃opt is received, the photocurrent converted by the PIN diode is

where 𝑅𝑅d is the responsivity of the photodiode and 𝑃𝑃opt is the received optical power. The noises at the receiver consist of three terms [75], i.e., thermal noise, which is only related to the thermal motion of the electrons, shot noise, which is due to the particle nature of light, and relative intens ity noise (RIN), mainly due to spontaneous emission from the transmitter laser. The total noise current variance in the receiver is

where the three terms on the right side of Eq. (5.3) represent thermal noise, shot noise and RIN, respectively. For a PIN diode, the noise can be written as

where 𝑘𝑘B,𝑇𝑇, 𝐹𝐹n, ∆𝑓𝑓, and 𝑅𝑅Lare the Bolzmann constant, the temperature in Kelvin, the noise figure of the receiver amplifier, the bandwidth of the photodiode, and the resistance, respectively. The shot noise induced current variance is proportional to the combination of incident current 𝐼𝐼opt and dark current 𝐼𝐼d, which is usually far smaller than 𝐼𝐼opt and can be neglected. q is elementary charge. The RIN depends on the square of the photocurrent, and 𝜉𝜉 is the average relative intensity noise spectral density.

Optical receivers that employ APDs generally provide better sensitivity than the ones with PIN diodes. This improvement is due to the internal gain in an APD that increases the photocurrent by a multiplication factor 𝑀𝑀APD. The photocurrent converted by an APD when receiving 𝑃𝑃opt is

The thermal noise in an APD diode can be expressed the same as the one of a PIN diode. The shot noise term is in a different form, can be expressed as:

𝐼𝐼opt = 𝑅𝑅d ∙ 𝑃𝑃opt, (5.2)

𝜎𝜎2 = 𝜎𝜎T2 + 𝜎𝜎s2 + 𝜎𝜎RIN2 , (5.3)

𝜎𝜎2 = 4𝑘𝑘B𝑇𝑇𝑅𝑅n∆𝑓𝑓𝑅𝑅L

+ 2𝑞𝑞�𝐼𝐼opt + 𝐼𝐼d�∆𝑓𝑓 + 𝜉𝜉 ∙ 𝐼𝐼opt2∆𝑓𝑓, (5.4)

𝐼𝐼opt = 𝑀𝑀APD ∙ 𝑅𝑅d ∙ 𝑃𝑃opt. (5.5)

32


With Eqs. (5.2)－(5.6), the noise current variance in the receiver can be obtained with the received optical power.

Using M-PAM with Gray labeling in the analysis, the BER can be approximated as a function [76]

where 𝑃𝑃𝑖𝑖𝑖𝑖 is the probability of transmitting PAM symbol 𝑠𝑠 but receiving symbol 𝑗𝑗 . It can be expressed as

In Eq (5.8), 𝐼𝐼𝑖𝑖 is the photocurrent of symbol i, scaled so that its average satisfies

�1𝑀𝑀� ∑ 𝐼𝐼𝑖𝑖 = 𝐼𝐼opt .𝑖𝑖 𝐼𝐼th,𝑖𝑖 denotes the decision threshold between symbols j and j−1, 𝜎𝜎𝑖𝑖 is the noise

current variance at symbol 𝑠𝑠 and can be calculated using Eqs. (5.1)－(5.5). 𝐼𝐼th,0 should be interpreted as −∞ and 𝐼𝐼th,𝑀𝑀 as +∞. Then, the optimum threshold level 𝐼𝐼th ,𝑖𝑖 that minimizes the error rate can be expressed as:

When 𝜎𝜎𝑖𝑖 = 𝜎𝜎𝑖𝑖−1, i.e., the photocurrent independent thermal noise dominates the receiver noises, equal distant symbols in modulation with the decision thresholds half-way between the two symbols enables the least errors. However, this modulation and decision scheme is suboptimal [77] when the dominant noise depends on the received optical power, e.g., shot noise which is linear ly proportional to the received optical power. When shot noise is the main factor in the overall receiver noises, the best transmission performance can be obtained with quadratic symbol spacing in modulation and the corresponding optimum thresholds achieved by Eq. (5.9). This finding shows that the symbol spacing and the thresholds should be adjusted accordingly to different receiver noise.

Given a BER requirement, the receiver sensitivity can be obtained by Eqs. (5.2)－(5.9). The impact of receiver type, transceiver bandwidth, modulation format, symbol level spacing and thresholds are included in the sensitivity evaluation. The system power budget can thus be obtained as the difference between the launch power of the transmitter and the receiver sensitivity.

𝜎𝜎s2 = 2𝑞𝑞 𝑀𝑀APD2 �𝑘𝑘A𝑀𝑀APD + (1− 𝑘𝑘A) �2− 1

𝑀𝑀APD��𝑅𝑅d𝑃𝑃opt∆𝑓𝑓. (5.6)

𝐵𝐵𝐸𝐸𝑅𝑅 ≈ 𝑆𝑆𝑆𝑆𝑅𝑅log2 𝑀𝑀

= 1log2 𝑀𝑀

1𝑀𝑀∑ ∑ 𝑃𝑃𝑖𝑖𝑖𝑖𝑀𝑀−1

𝑖𝑖=0,𝑖𝑖≠𝑖𝑖𝑀𝑀−1𝑖𝑖=0 , (5.7)

𝑃𝑃𝑖𝑖𝑖𝑖 = 12

erfc�𝐼𝐼𝑡𝑡ℎ ,𝑗𝑗−𝐼𝐼𝑖𝑖𝜎𝜎𝑖𝑖√2

� − 12

erfc�𝐼𝐼𝑡𝑡ℎ ,𝑗𝑗+1−𝐼𝐼𝑖𝑖𝜎𝜎𝑖𝑖√2

�. (5.8)

𝐼𝐼𝑡𝑡ℎ,𝑖𝑖 = 𝐼𝐼 𝑖𝑖𝜎𝜎𝑖𝑖−1−𝐼𝐼𝑖𝑖−1𝜎𝜎𝑖𝑖𝜎𝜎𝑖𝑖−1+ 𝜎𝜎𝑖𝑖

. (5.9)

33

5.2. Coupler-based POI architectures

In Paper V, we present three coupler-based POI architectures that can be applied in small size datacenters or the edge tier of DCNs. The scalability of the POIs is investigated by using the proposed methodology.

5.2.1. Description of the architectures

All the proposed POI architectures presented are based on SMF and C-band, which allows for DWDM adoption offering high capacity and interoperability throughout both the intra- and inter-datacenter network. An implementation of the architectures at the edge tier of datacenter (i.e., at ToR) is taken as an example to describe the way they work. In all the schemes, each server is connected to the ToR through an optical network interface (ONI). A tunable laser and a tunable filter are used at the transmitter and the receiver side, respectively. It has been demonstrated that tuning the transceivers can be realized in nanoseconds [78], [79]. Fast tunable transceivers are

N x 2Coupler

WSS

1

2

N

WTT

RX OTF

Server 1 ONI

Server 2 ONI

Server N ONI ISO

· · ·

· · ·

Server 1

(N+1)x(N+1)Coupler

N+1 N+1

WSS

1 1

N N

WTT

RX OTF

Server 2 ONI

Server N ONI

2 2

· · ·

· · ·

(a) Scheme I (b) Scheme II

Server 1

(K+1)×(K+1) Coupler

ISO

T×2 coupler

K+1

K+1WSS

1 1

K

Server 2 ONI

Server N ONI

Server(N− T)

Server(N −T+1) ONI

Server T ONI

WTT

RX OTF

K

· · ·

· · ·

· · ·

· · ·

ONI

ONI

(c) Scheme III

Figure 18. Architectures of coupler-based POI. (ONI: optical network interface; WTT: wavelength tunable transmitter; RX: receiver; OTF: optical tunable filter; WSS: wavelength selective switch; ISO: isolator.)

34

5.2. Coupler-based POI architectures

technically feasible and the price is expected to be significantly reduced once the market demand increases.

For the traffic within the rack, the data sent from the server passes the coupler at ToR and broadcast to all the servers in the same rack. The traffic with the destination outside of the rack, on the other hand, is selected and directed by the wavelength select switch (WSS). The media access control (MAC) protocol proposed in [52] can be applied in all the three architectures in order to avoid confliction that the same wavelength is assigned to the more than one traffic request. Among all three architectures shown in Figure 18, Scheme I is based on an N×2 coupler and Scheme III utilizes two-stage cascaded couplers, both only requiring a single-fiber port in each ONI. This single-port feature helps to reduce the number of ports and amount of fibers in the ToR interconnect, and consequently improves the spatial efficiency in datacenters. Scheme II is a double-fiber scheme, where each server is assigned to a dual-port ONI. The advantage of doing so is to reduce the insertion loss for the intra-rack communication. However, it may increase the cabling complexity.

The datacenter can benefit from introducing POI at ToR in terms of cost, energy consumption and reliability. However, the significant insertion loss the proposed POI architectures can be a bottleneck to the total capacity. The scalability of the POI can be assessed by determining two indicators, interconnect size, i.e., the maximum number of available ports that can be connected, and the highest data rate that can be supported with the required transmission quality. As Figure 16 shows, the feasibility of a certain POI size can only be demonstrated when the system power budget supported by the transceiver is larger than the link budget. In the proposed scalability evaluation methodology, the power loss of all the components in the link and all the noise components in the receiver should be considered.

Figure 19. The relationship between the size of interconnect N and the maximum data rate in error-free transmission

(BER=10-12) with a receiver employing (a) a PIN diode and (b) an APD.

35


5.2.2. Assessment results

As case studies, we investigate the coupler-based POIs presented in Paper V. The results reveal that when the received optical power is in the range of −10 to 5 dBm, the dominant noise in a PIN diode is thermal noise while shot noise is the main factor in an APD. Consequently, with various types of the receivers, the optimum symbol spacing in modulation and decision thresholds are different. Equal distant symbol in M-PAM leads to the optimum transmission performance when a PIN diode is deployed at the receiver. For a receiver employing an APD, equal interspaced symbol in M-PAM is suboptimal. Instead, the quadratical distant symbol and the corresponding decision thresholds enable better BER performance.

Based on the methodology presented in Chapter 5.1, the system power budget ensuring error-free M-PAM signal transmission (i.e., BER≤1E-12) can be achieved. The highest data rate that can be allowed on a per-server basis, and the maximum number of servers that can be held in a rack are obtained. As shown in Figure. 19 and Figure. 20, Scheme I based on the N×2 coupler is only able to be used for very small interconnects (less than 32 ports available). Both Schemes II and III are capable of hosting interconnects of more than 500 ports. With standard forward error correction (FEC) and the low cost PIN receiver employment, the theoretical interconnect capacity can be beyond 5 Tb/s, which is sufficient for a medium size datacenter. Note that the scalability can be further enhanced by using more advanced transceivers, e.g., transceivers with bandwidth larger than 10 GHz.

5.3. Introducing EDFA in POIs

In the previously presented POI options, Scheme I in which the servers are connected by an N×2 coupler, suffers from the highest insertion loss and hence has the worst scalability. One way to improve the scalability performance is to increase the system power budget by introducing

Figure 20. The relationship between the size of interconnect N and the maximum data rate under the FEC threshold of

BER=10-3 with a receiver employing (a) PIN diode and (b) APD.

36

5.3. Introducing EDFA in POIs

amplifier(s). Since the presented DCN architecture operates in 1550 nm region, an EDFA, which is widely used in optical communications within C band, is introduced in Scheme I to enhance the scalability. The application of such a POI at ToR is taken as an example for the scalability analysis.

5.3.1. EDFA enabled POI architecture

As shown in Figure 21 (a), a single EDFA is introduced to the N×2 coupler-based POI architecture on the loop-back path to improve the system power budget of the intra-rack communication. The equivalent link model of the intra-rack communication is illustrated in Figure 21 (b). Since EDFA is able to amplify the wideband signals in the whole C band, only one EDFA is needed for the whole rack. The intra-rack signals are selected by the WSS and then amplified by the EDFA on the loop-back path (see the red arrow in Figure 21). The EDFA causes extra cost and energy consumption to the POI architectures. For a rack hosting 60 servers, the cost increases by only 0.6% and power dissipation grows by only 1% based on the data collected from the literature [25], [80]. Besides, the intra-rack traffic is very bursty, consequently the wavelengths that carry the traffic need to be fast switched on and off, which potentially causes transient variations of EDFA

Figure 21. Thermal noise, shot noise, relative intensity noise, ASE noise and total noise for the given

system with a 10 GHz bandwidth PIN diode at the receiver.

Figure 22. (a) The coupler-based optical interconnect at ToR with an EDFA; (b) The equivalent

block diagram for the link within the coupler-based optical interconnects.

37


gain [81], [82]. It may significantly degrade the transmission performance and damage the receivers. Fortunately, such transient gain can be suppressed by applying proper techniques [83], [84].

5.3.2. Assessment result

To quantify the scalability of the POI with an EDFA, the link loss and the system power budget are calculated. EDFA boosts the power budget while causing amplified spontaneous emission (ASE) noise, which may affect the receiver sensitivity. With this in mind, the noises at the PIN receiver are analysed considering the factor of ASE noise. As illustrated in Figure 22, the ASE noise can be ignored compared to the dominant noises, i.e., thermal noise and shot noise.

Figure 23. BER performance of 10 GBaud M-PAM with a PIN diode at the receiver.

Figure 24. The relationship between the size of interconnect N and the maximum data rate (a) in

error-free transmission (BER=10-12) and (b) under a FEC limit threshold (BER= 10-3).

38

5.4. Study on the impact of modulation formats

Accordingly, having the EDFA hardly affects the modulation and decision scheme. The sensitivity of a 10 GHz PIN diode receiver can be obtained by applying the proposed methodology with a given BER requirement. The BER performance of 10 GBaud M-PAM signal transmission as a function of the received optical power is shown in Figure 23.

The scalability of the POI architecture with and without EDFA are shown in Figure 24. The introduction of a single EDFA enables the POI hosting more than 100 ports for error-free transmission. By applying FEC the capacity of the interconnect can be further enhanced. The peak switching capacity increases from 160 Gb/s to over 2 TGb/s.


In mega datacenters the bandwidth demand per server is reaching 10 Gb/s and beyond [85]. The most popular choices of modulation formats for IM/DD to meet the capacity demand are mult i-level pulse amplitude modulation (M-PAM) and discrete multi-tone (DMT) [59]. Although, some works have been carried out on comparison of PAM and DMT for high data rate transmission using low cost transceivers[60]–[62], such studies have not yet been put in the context of POI architectures, which have high insertion loss even within very short reach. Besides, a new modulation format, i.e., electrical duo-binary (EDB), is also gaining a lot of interests, thanks to its high tolerance to chromatic dispersion (comparing to OOK) and reduced equalization requirement for the high -speed serial rate link.

5.4.1. PAM, DMT and EDB

In Paper VII we experimentally evaluate the short-reach optical links deploying two advanced modulation formats. As shown in Figure 25. Experimental setup of the PAM and DMT transmissions (PPG: pulse-pattern generator, DAC: digital-to-analog converter, AWG: arbitrary waveform generator, DSO: digital sampling oscilloscope). Insets: eye diagram of electrical 28 Gbaud 4-PAM generated from the DAC; electrical spectrum of generated and received DMT signals with pre-emphasis., a high-speed compact InGaAsP basaed

Figure 25. Experimental setup of the PAM and DMT transmissions (PPG: pulse-pattern generator, DAC: digital-

to-analog converter, AWG: arbitrary waveform generator, DSO: digital sampling oscilloscope). Insets: eye diagram of electrical 28 Gbaud 4-PAM generated from the DAC; electrical spectrum of generated and received

DMT signals with pre-emphasis.

DFB-TWEAM

AWG

DSO

2bitDAC

Resa

mpli

ngSy

nchr

oniza

tion

S/P

FFT

Chan

el es

timat

ion In

verse

CAZA

CBE

R Co

unt

DMT

Resa

mpli

ng

Low

pass

filte

ring

DFE

BER

Coun

t

PAMOffline

S/P

Map

ping

CAZA

C IFF

TCy

clic P

refix

P/S

DMT

PPG

PAM

Pre-

emph

asis

Frequency, GHz

-25 -20 -15 -10 -5 0 5 10 15 20 25

Powe

r, dB

-150

-140

-130

-120

-110

-100

-90

-80

-70

-60

-50

1548,4 1548,6 1548,8 1549,0-70

-60

-50

-40

-30

-20

-10

0

powe

r (dB

)

Wavelength (nm)

Frequency, GHz

-25 -20 -15 -10 -5 0 5 10 15 20 25

Powe

r, dB

-150

-140

-130

-120

-110

-100

-90

-80

39


monolithically integrated distribution feedback travelling wave electro-absorption modulator (DFB-TWEAM) module with 3 dB bandwidth of ~70 GHz [86] is deployed to modulate and transmit data. After 4 km SMF transmission, the signal is directly detected by a PIN diode receiver with 3 dB bandwidth of 90 GHz. 56 Gb/s 4-PAM and 25 Gb/s DMT signals transmission are implemented and their performance are compared. Digital signal processing is applied in order to improve the transmission performance. The distortion induced by 4 km SMF transmission in 4-PAM case is compensated by a three-tap decision-feedback equalizer (DFE).

As showing in Figure 26, reaching a BER lower than the FEC limit, the receiver sensitivity of 4-PAM signal is improved by 3 dB by introducing DFE, while the 56 Gbaud OOK is available only if DFE is employed. In DMT case, the adaptive bit and power loading and the constant amplitude zero auto-correlation (CAZAC) sequence equalization which is presented in detail in [87], is implemented to improve the receiver sensitivity. It can be seen that an improvement of ~1 dB can be obtained at BER = 1.0⋅10−3. The BER performance of both advanced modulation formats indicate that, digital equalization remains essential for high speed short-haul transmission even without bandwidth limitation from the components. Compared with the high tap numbers employed in many of the previous studies [60], [62], the complexity of the considered digita l equalizer is largely reduced.

Apart from the aforementioned DMT and M-PAM, EDB becomes another competitive solution, which is able to utilize the low pass filtering effect of the channel as the equalizer and consequently reduces the bandwidth requirement of both electrical and optical devices [88]. In Paper IX, a real-

Figure 26. BER performance versus received optical power for (a) PAM and (b) DMT.

DFB-TWEAM PIN

SSMF

VOA4:1

MUX6-tapFFEPRBS

Electrical Tx chip DEMUX

XORDB

frontendError

Detector

Electrical Rx chip

Figure 27. Experimental setup of the real-time 100 Gb/s 3-level duo-binary optical transmission.

40


time 100 Gb/s EDB optical link is demonstrated with the in-house developed transceiver chips. The experiment setup is shown in Figure 27. At the transmitter side, four electrical 25 Gb/s NRZ signals are generated and multiplexed into a serial 100Gb/s stream. This 100Gb/s signal is pre-equalized by the transmitter chip and amplified to drive the DFB-TWEAM. After SSMF transmission, a variable optical attenuator is used before a high speed PD to convert the signals. Subsequently the duo-binary frontend demodulates the 3-level eyes with separate thresholds levels which are adjusted to obtain the lowest BER. The demodulated signal is then de-serialized by the on-chip DEMUX and one of the four outputs is then used for the real-time error detection. The BER performance of the 100 Gb/s EDB over different fiber lengths can be seen in Figure 28. Negligible power penalty is observed for 500 m and 1 km applications with high received optical power. For the 2 km SSMF transmission, the BER threshold of 3.8E-3 can still be reached with HD-FEC applied.

5.4.2. Experiment validation of DMT/M-PAM in POI

In Paper VIII we first experimentally compare the transmission performance and then analyze the scalability of employing M-PAM and DMT in the context of POI at ToR. The experimental setup is shown in Figure 29. The same transmitter as the one mentioned in the work of Paper VII is used. The laser is centred at 1548.5 nm. The driving current of the laser is set to 94 mA and the EAM is biased at 1.83 V to work in the linear regime. The signals are generated by the arbitrary waveform generator with 3 dB bandwidth < 13 GHz. No additional SMF is introduced between the transmitter and receiver in order to imitate the ultra-short fibre transmission at ToR. The back-to-back (B2B) signal is detected by an opto-electronic receiver (ORTEL Receiver module 2860C) of 10 GHz bandwidth. The signal is captured by an 80GSa/s digital sampling oscilloscope (DSO)

Figure 28. BER curves of 100Gbit/s electrical duo-binary transmission at different fibre lengths.

41


and processed off-line. The bandwidth of the link is determined by the component with smallest bandwidth, which is the photo-diode in the receiver in this case. In other words, the operation bandwidth link is 10 GHz. The same digital equalization methods as presented in Paper VII are applied. The b2b transmission performance of 10 Gb/s and 20 Gb/s PAM and DMT signals are investigated in Paper VIII, as shown in Figure 30.

5.4.3. Scalability assessment results

Based on the BER performance of both PAM and DMT with and without digital equaliza t ion

Figure 29. Experiment setup of transmission performance evaluation of M-PAM and DMT at ToR.

Figure 30. The scalability analysis of the POI when different types of FEC is applied without and with

digital equalization employing PAM and DMT at 10 Gb/s and 20 Gb/s

42


shown in Figure 30, we further investigate the scalability of employing both modulation formats in the context of POI. Three types of FEC are considered, i.e., SD-FEC, E-FEC, and G.709 FEC, which have levels of coding redundancy [89] in descending order. The corresponding required BER threshold before FEC processing are 3.8⋅10−3, 1.0⋅10−3, and 8.0⋅10−5, respectively. The comparison between the cases with and without digital equalization while applying the three types of FEC, are listed in Figure 31. We can observe that without digital equalization the POI architecture is capable of hosting more than 500 ports when 10 Gb/s OOK with SD-FEC is used. The number of ports that can be interconnected by the POI scheme is up to 63 achieving 20 Gb/s per port when 4-PAM is employed. On the other hand, the adoption of DMT at both data rates only allows interconnection of 31 ONIs in the case that SD-FEC is allowed without any digita l equalization applied. The use of the equalization doubles the size of the POI using DMT. Applying E-FEC, which has less coding redundancy than SD-FEC, the digital equalization increases the scalability of the POI by a factor of 2 when 20 Gb/s 4-PAM and DMT is employed. When G.709 FEC is used, the POI can only be feasible when applying equalization for both 4-PAM and DMT.

43

44

Chapter 6

Conclusions and future works

This chapter summarizes the main contribution of this thesis and concludes some possible

directions for the future work.

6.1. Conclusions

In this thesis, we study high capacity and short-reach optical communication networks. In particular, we focus on fiber access network with the signal propagation distance of tens of kilometers, and optical datacenter networks with the transmission distance of a few kilometers.

For the fiber access network, we address the integration of DWDM technique and PON to meet the high capacity demand. Introducing OFC in DWDM PON as the multi-carrier source, significantly stabilizes the light source due to the locked channel spacing between the various wavelength carriers. We have investigated two OFC generation schemes, each providing equal carrier power and allowing the frequency spacing to be reconfigurable without needing a complex system configuration. A proof-of-concept OFC based DWDM-PON has also been demonstrated, where OFDM has been employed to flexibly allocate downstream sub-carriers to the end users. Moreover, for the upstream transmission, the carrier is reused and thus the cost of the overall system can be reduced.

MMW band is able to provide enormous bandwidth, becoming a promising candidate to meet the growing capacity requirement of the wireless access network. However, the serious attenuation of wireless propagation extremely limits the coverage of MMW. Thanks to the low signal

45

Chapter 6. Conclusions and future works

attenuation, RoF technique is widely considered to provide high-capacity connections between the central office and antennas. With this in mind, we have proposed a dual-band MMW generation scheme, referred to as the palm-shaped spectrum generation scheme, which can double the operation bandwidth. The optical carrier (i.e., the middle finger of the palm) can be used for upstream transmission at the cell, while dual-band MMWs carrying with different data for downstream are generated by direct beating between symmetric sidebands of the palm-shaped spectrum. A proof-of-concept experiment has been carried out to confirm that the proposed spectrum generation scheme can provide significantly higher OSSR and hence better transmiss ion performance in the upstream direction compared with the optical frequency comb where all the wavelengths have the same power. In addition, the feasibility of employing the proposed palm-shaped spectrum generation in the RoF system has been verified by simulation.

Regarding the optical datacenter networks, we have focused on the energy efficient optical interconnect solutions, i.e., POIs. A scalability evaluation methodology is developed considering the fundamental impairments in the ultra-short-reach transmission in datacenter networks, i.e., power loss and receiver noise. Besides, the other physical- layer parameters, such as the receiver bandwidth, modulation format, symbol spacing and thresholds have also been considered in the proposed scalability evaluation framework. Based on the proposed methodology, the scalability of the coupler-based POI architectures has been evaluated. The results have shown that the coupler based POI is capable of interconnecting more than 500 ports, which fits a small or medium size datacenter. We have also proposed to enhance the scalability by introducing only a single EDFA in a coupler-based POI architecture. The switching capacity can be improved by a factor of 16. Furthermore, the relationship between the system scalability and different advanced modulat ion formats has been also studied and transmission performance of both PAM and DMT signals in a short-reach optical link has been experimentally investigated. The digital signal processing improves the transmission performance and thus the scalability of the interconnect. It is shown that PAM surpasses DMT in terms of transmission performance when the data rates are 10 Gb/s and 20 Gb/s. Furthermore, the scalability of POI employing PAM and DMT are investigated with the implementation of different FEC techniques.

6.2. Future work

High capacity optical short-reach transmission has been a hot topic in the recent years. Compared to the long-haul transmission, impairment compensation is simplified in short-reach communications. On the other hand, cost-efficiency and energy-efficiency become more important, leading to many new research questions to be addressed. Some of them are listed below:

Which techniques are proper to offer 100/400 Gb/s and beyond per channel to satisfy the continuously increased capacity demand?

Are there any other modulation formats that are able to outperform DMT and M-PAM to meet the low cost requirements in short-haul communications?

46

6.2. Future work

As highly integrated and low power devices are preferred in both fiber access networks and DCNs, can they at the same time satisfy the requirement of high capacity as well as providing sufficient signal quality?

How does the physical layer configuration affect the network layer performance? For instance, to achieve the acceptable transmission quality, a minimum spacing between two channels needs to be reserved to resist the crosstalk. It dramatically influences the spectrum utiliza t ion and potentially degrades other network characteristics, such as latency and throughput, especially under high load. Such a cross-layer design problem in short-reach communica t ion needs to be exploited.

47

48

Chapter 7 Summary of the original works

In this chapter, the publication included in the thesis is presented as well as a summary of the author’s contribution.

Paper I: R. Lin, Z. Feng, M. Tang, S. Fu, P. Shum and D. Liu, “Spacing Switchable Flat Broadband Optical Comb Generation Based on Cascaded Electro-optical Modulator”, Asia Communications and Photonics Conference and Exhibition (ACP), 2013

This paper proposed a simple optical frequency comb generation scheme by using cascaded LiNbO3 based phase modulator and intensity modulator. Variable spacing of the comb is demonstrated by realizing the comb with 12.5 GHz and 8.33 GHz frequency difference.

Contributions: Original idea, design and implementation of the experimental setup, and preparation of the manuscript.

Paper II:

R. Lin, M. Tang, R. Wang, Z. Feng, So. Fu, D. Liu, J. Chen and P. Shum, "An Ultra-dense Optical Comb Based DWDM-OFDM-PON System", Progress in Electromagnetics Research Symposium (PIERS), 2014.

In this paper, we designed a reconfigurable DWDM PON system based on an optical frequency comb generation scheme. The OFC generation scheme at the OLT is able to provide ultra-dense wavelengths with switchable frequency spacing. 2.5 Gb/s QPSK-OFDM is used to encode the

49

downstream data while the carrier is re-used and modulated with OOK signal to carry the upstream data. The proposed PON system is experimentally demonstrated.


Paper III:

R. Lin, Z. Feng, M. Tang, R. Wang, S. Fu, P. Shum, and D. Liu, "Palm-Shaped Optical Spectrum Generation for Fiber-Wireless Integrated Communication with Dual-Band Millimeter Wave Capability", Asia Communications and Photonics Conference and Exhibition (ACP), 2014 (Best student paper)

A concept of palm-shaped spectrum for the integration of fixed and mobile communication is proposed with the capability of simultaneously generating dual-band MMW by using a single DPMZM. With proper control of the bias voltage of both arms of the DPMZM, a high power finger and four flat sidebands are generated. The MMW is generated in up-converted way. Two MMW can be produced by beating of the corresponding sidebands. The proof-of-concept experiment is implemented. The palm-shaped spectrum shows better performance in terms of the transmission performance of the middle finger wavelength compared to optical frequency comb.


Paper IV:

R. Lin, Z. Feng, M. Tang, R. Wang, S. Fu, P. Shum, D. Liu, and J. Chen , “Palm-Shaped Spectrum Generation for Dual-band Millimeter Wave and Baseband Signals over Fiber”, Optics Communications, vol. 367, pp. 137-143, 2016.

As an extension of Paper V, the simulation of the over-all system performance is implemented. In the simulation, the downstream data on two MMW bands are carried by the sidebands while the middle finger of the palm-shaped spectrum carries the upstream data. Different MMWs configuration is implemented, i.e., 40 GHz and 80 GHz MMW with 25 km fiber transmission, and, 60 GHz and 120 GHz MMW with 19 km fiber transmission. Both downstream and upstream data are encoded by OOK signal for the feasibility demonstration.

Contributions: Original idea, design and implementation of the simulation and preparation of the manuscript.

Paper V:

R. Lin, K. Szczerba, E. Agrell, L. Wosinska, M. Tang, and J. Chen, “Physical- layer Evaluation Methodology for Scalability Analysis of Optical Interconnects”, Photonics Journal, submitted.

In this paper, we developed a methodology assessing the scalability of optical interconnect in datacenter networks considering the fundamental impairments in the short-reach optical links,

50

i.e., the power loss and the receiver noise. The power loss plus by the power margin is the least link budget, while the difference between the output power of the transmitter and the receiver sensitivity determines the maximum power can be provided by the transceivers. By analyzing the receiver noises, the optimum modulation and decision schemes for M-PAM with PIN and APD at the receivers are exploited. Besides, three POIs are presented and evaluated in terms of scalability as case studies.

Contributions: Original idea, development of the methodology, theoretical calculation and preparation of the manuscript.

Paper VI:

R. Lin, K. Szczerba, E. Agrell, L. Wosinska, M. Tang, and J. Chen, “To Overcome the Scalability Limitation of Passive Optical Interconnects in Datacentres”, Asia Communications and Photonics Conference and Exhibition (ACP), 2016, accepted.

Based on the scalability analysis results in Paper V, a single EDFA is introduced in a coupler-based POI to boost the scalability. The impact of the extra EDFA is analyzed from the aspects of system cost, energy consumption, noise, as well as the scalability. The results shows that a single EDFA will not infect the dominate noise at the receiver side thus the modulation and decision scheme.

Contribution: Original idea, design of the architecture, theoretical calculations, preparation of the manuscript.

Paper VII R. Lin, X. Pang, O. Ozolins, Z. Feng, A. Djupsjöbacka, U. Westergren, R. Schatz, G. Jacobsen, M. Tang, S. Fu, D. Liu, and J. Chen “Performance Evaluation of PAM and DMT for Short-range Optical Transmission with High Speed InGaAsP DFB-TWEAM”, Optical Fiber Communication Conference (OFC), Th2A.58, 2016.

In this paper, the performance of both PAM and DMT are experimentally evaluated. 56 Gb/s 4-PAM signal and 25 Gb/s DMT signals over 4 km single mode fiber transmission is implemented. Digital signal processing is employed to improve the transmission performance.

Contribution: Original idea, design of the experiment setup, implementation of the experiments, and preparation of the manuscript.

Paper VIII R. Lin, X. Pang, O. Ozolins, Z. Feng, A. Djupsjöbacka, U. Westergren, R. Schatz, G. Jacobsen, M. Tang, S. Fu, D. Liu, and J. Chen “Experimenta l Validation of Scalability Improvement for Passive Optical Interconnect by Implementing Digital Equalization”, European conference and exhibition on optical communication (ECOC), 2016.

51

In this paper, 10 Gb/s and 20 Gb/s PAM and DMT deployment under the context of the coupler-based POI is experimentally evaluated and compared. We provide a trade-off between the use of FEC with different system overhead and the scalability of the POI.

Contribution: Original idea, design of the experimental setup, implementation of the experiment, preparation of the manuscript.

Paper IX X. Yin, M. Verplaetse, R. Lin, J. V. Kerrebrouck, O. Ozolins, T. D. Keulenaer, X. Pang, R. Pierco, R. Vaernewyce, A. Vyncke, R. Schats, U. Westergren, G. Jacobsen, S. Popov, J. Chen, G. Torf and J. Bauwelinck, “First Demonstrat ion of Real-Time 100 Gbit/s 3-Level Duobinary Transmission for Optical Interconnects”, European conference and exhibition on optical communication (ECOC), 2016, post deadline paper.

In this paper, the first real-time 100Gb/s 3-level duobinary transmission up to 2km SSMF enabled by the in-house developed transmitter and receiver chips.

Contribution: Initialization the collaboration with Ghent University, implementation of the experiment, preparation of part of the manuscript.

52

References [1] K. Igarashi, D. Souma, Y. Wakayama, K. Takeshima, Y. Kawaguchi, T. Tsuritani, I. Morita, and M. Suzuki,

“114 space-division-multiplexed transmission over 9.8-km weakly-coupled-6-mode uncoupled-19-core fibers,” in Optical Fiber Communication Conference, 2015.

[2] J. Sakaguchi, W. Klaus, J.-M. D. Mendinueta, B. Puttnam, R. S. Luis, Y. Awaji, N. Wada, T. Hayashi, T. Nakanishi, T. Watanabe, Y. Kokubun, and T. Takahata, T. Kobayashi, “Realizing a 36-core, 3-mode fiber with 108 spatial channels,” in Optical Fiber Communication Conference, 2015.

[3] B. J. Puttnam, R. S. Luís, W. Klaus, J. Sakaguchi, J.-M. D. Mendinueta, Y. Awaji, N. Wada, Y. Tamura, T. Hayashi, M. Hirano, and J. Marciante, “2.15 Pb/s transmission using a 22 core homogeneous single-mode multi-core fiber and wideband optical comb,” in European conference and exhibition on optical communication (Ecoc), PDP, 2015.

[4] P. J. Winzer, “Spatial multiplexing in fiber optics: The 10X scaling of metro/core capacities,” Bell Labs Tech. J., vol. 19, pp. 22–30, 2014.

[5] A. Banerjee, Y. Park, F. Clarke, H. Song, S. Yang, G. Kramer, K. Kim, and B. Mukherjee, “Wavelength-division-multiplexed passive optical network (WDM-PON) technologies for broadband access: a review [Invited],” J. Opt. Netw., vol. 4, no. 11, pp. 737–756, 2005.

[6] L. Gutierrez, P. Garfias, M. De Andrade, and S. Sallent, “Next Generation Optical Access Networks : from TDM to WDM,” 2006.

[7] G. K. Chang, A. Chowdhury, Z. J. Z. Jia, H. C. Chien, M. F. Huang, J. Yu, and G. Ellinas, “Key Technologies of WDM-PON for Future Converged Optical Broadband Access Networks [Invited],” J. Opt. Commun. Netw., vol. 1, no. 4, pp. 35–50, 2009.

[8] M. P. Thakur, S. Mikroulis, C. C. Renaud, J. E. Mitchell, and A. Stöhr, “DWDM-PON / mm-Wave Wireless Converged Next Generation Access Topology using Coherent Heterodyne Detection,” International Conference on Transparent Optical Networks (ICTON), 2014.

[9] International Telecommunication Union - ITU-T, Recommendation ITU-T G.694.1, Spectral grids for WDM applications: DWDM frequency grid.

[10] International Telecommunication Union - ITU-T, Recommendation ITU-T G.671: Transmission characteristics of optical components and subsystems.

[11] D. Nesset and S. Member, “NG-PON2 technology and standards,” J. Ligthwave Technol., vol. 33, no. 5, pp. 1136–1143, 2015.

[12] E. Summary, "Cisco Visual Networking Index: Forecast and Methodology, 2014-2019 White Paper", online available: http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns1175/Cloud_Index_White_Paper.html#wp9000816.

[13] IEEE Std 802.11, Information technology—Telecommunications and information exchange between systems — Local and metropolitan area networks — Specific requirements.

[14] C. Lim, A. Nirmalathas, M. Bakaul, P. Gamage, K. L. Lee, Y. Yang, D. Novak, and R. Waterhouse, “Fiber-wireless networks and subsystem technologies,” J. Light. Technol., vol. 28, no. 4, pp. 390–405, 2010.

[15] A. Hirata, M. Harada, and T. Nagatsuma, “120-GHz Wireless link using photonic techniques for generation, modulation, and emission of millimeter-wave signals,” J. Light. Technol., vol. 21, no. 10, pp. 2145–2153 , 2003.

[16] L. Zhang, X. Hu, P. Cao, T. Wang, and Y. Su, “A bidirectional radio over fiber system with multiband-signal generation using one single-drive MZM.,” Opt. Express, vol. 19, pp. 5196–5201, 2011.

[17] L. Zhang, C. Ye, X. Hu, Z. Li, S.-H. Fan, Y.-T. Hsueh, Q. Chang, Y. Su, and G.-K. Chang, “Generation of multiband signals in a bidirectional wireless over fiber system with high scalability using heterodyne mixing technique,” IEEE Photonics Technol. Lett., vol. 24, no. 18, pp. 1621–1624, Sep. 2012.

53

[18] C. Zhang, T. Ning, J. Li, L. Pei, C. Zhang, and S. Ma, “A full-duplex wired / wireless integrated ROF system based on tunable optical frequency comb generator,” Opt. Commun., vol. 344, pp. 65–70, 2014.

[19] Y. T. Hsueh, Z. Jia, H. C. Chien, J. Yu, and G. K. Chang, “A novel bidirectional 60-GHz radio-over-fiber scheme with multiband signal generation using a single intensity modulator,” IEEE Photonics Technol. Lett., vol. 21, no. 18, pp. 1338–1340, 2009.

[20] Q. Chang, S. Member, H. Fu, Y. Su, and S. Member, “Simultaneous Generation and Transmission of Downstream Multiband Signals and Upstream Data in a Bidirectional Radio-Over-Fiber System,” vol. 20, no. 3, pp. 2007–2009, 2008.

[21] P. Delforge, “America ’ s Data Centers Are Wasting Huge Amounts of Energy,” Nat. Resour. Def. Counc., vol. IB:14-08-A, no. August, pp. 1–5, 2014.

[22] W. Machines, The Datacenter as a Computer. 2009.

[23] C. Kachris, K. Bergman, and I. Tomkos, optical interconnects for data center networks, vol. 1. 2013.

[24] Y. Gong, X. Hong, Y. Lu, S. He, and J. Chen, “Passive optical interconnects at top of the rack: offering high energy efficiency for datacenters,” Opt. Express, vol. 23, no. 6, pp. 7957-7970, 2015.

[25] M. Fiorani, S. Aleksic, M. Casoni, L. Wosinska, and J. Chen, “Energy-Efficient Elastic Optical Interconnect Architecture for Data Centers,” IEEE Commun. Lett., vol. 18, no. 9, pp. 1531–1534, 2014.

[26] X. Yin, F. Blache, B. Moeneclaey, J. Van Kerrebrouck, R. Brenot, and G. Coudyzer, “40-Gb / s TDM-PON downstream with low-cost EML transmitter and 3-level detection APD receiver,” in Optical Fiber Communication Conference, 2016.

[27] ITU-T, “G.985 Corrigendum 1: 100 Mbit/s point-to-point Ethernet based optical access system,” vol. 985, no. 2003, 2005.

[28] IEEE, Std 802.3ah, Ethernet in the First Mile.

[29] International Telecommunication Union - ITU-T, G.987 Series, 10 Gigabit-Capable Passive Optical Network (XG-PON).

[30] IEEE Std 802.3av, Physical Layer Specifications and Management Parameters for 10 Gb/s Passive Optical Networks.

[31] C. H. Lee, S. M. Lee, K. M. Choi, J. H. Moon, S. G. Mun,, K. T. Jeong, B. Kiml, “WDM-PON experiences in Korea [Invited].,” J. Opt. Networking, vol. 6, no. 5, pp. 451–464, 2007.

[32] S. Gee, F. Quinlan, S. Ozharar, P. J. Delfyett, J. J. Plant, and P. W. Juodawlkis, “Optical frequency comb generation from modelocked diode lasers - techniques applications,” in Digest of the,LEOS Summer Topical Meetings, 2005.

[33] H. Takara, T. Ohara, K. Mori, K. Sato, E. Yamada, Y. Inoue, T. Shibata, M. Abe, T. Morioka, and K.-I. Sato, “More than 1000 channel optical frequency chain generation from single supercontinuum source with 12.5 GHz channel spacing,” Electron. Lett., vol. 36, no. 25, p. 2089, 2000.

[34] P. Del’Haye, A. Schliesser, T. Wilken, R. Holzwarth, T. J. Kippenberg, and Ieee, “Kerr Nonlinearity induced Optical Frequency Comb Generation in Microcavities,” Conf. Lasers Electro-Optics/Quantum Electron. Laser Sci. Conf., 2007.

[35] R. Wu, V. R. Supradeepa, C. M. Long, D. E. Leaird, and A. M. Weiner, “Generation of very flat optical frequency combs from continuous-wave lasers using cascaded intensity and phase modulators driven by tailored radio frequency waveforms.,” Opt. Lett., vol. 35, no. 19, pp. 3234–3236, 2010.

[36] T. Yamamoto, T. Komukai, K. Suzuki, and A. Takada, “Spectrally flattened phase-locked multi-carrier light generator with phase modulators and chirped fibre Bragg grating,” Trans. Korean Inst. Electr. Eng., vol. 57, no. 6, pp. 982–984, 2008.

[37] N. J. Gomes, P. P. Monteiro, and A. Gameiro, Next Generation Wireless Communications Using Radio over Fiber. 2012.

54

[38] S. K. Yong and C. C. Chong, “An overview of multigigabit wireless through millimeter wave technology: Potentials and technical challenges,” Eurasip J. Wirel. Commun. Netw., vol. 2007, 2007.

[39] C. Hansen, “Industry perspectives WiGig: multi-Gigabit wireless communications in the 60 Gh Z B,” Wireless Commun. vol. 18, no.6,. December, pp. 6–7, 2011.

[40] Federal communications commission office of engineering and technology policy and rules division, “FCC online table of frequency allocations,” 2007.

[41] N. Guo, R. C. Qiu, S. S. Mo, and K. Takahashi, “60-GHz Millimeter-Wave Radio: Principle, Technology, and New Results,” EURASIP J. Wirel. Commun. Netw., vol. 2007, pp. 1–8, 2007.

[42] P. K. Pepeljugoski, J. a. Kash, F. Doany, D. M. Kuchta, L. Schares, C. Schow, M. Taubenblatt, B. J. Offrein, and A. Benner, “Low Power and High Density Optical Interconnects for Future Supercomputers,” Optical Fiber Communication Conference (OFC), 2010.

[43] M. A. Taubenblatt, “Optical Interconnects for High-Performance Computing,” J. Light. Technol., vol. 30, no. 4, pp. 448–457, 2012.

[44] Christoforos Kachris, K. Bergman, and Ioannis Tomkos, Optical Interconnects for Future Data Center Networks. 2013.

[45] C. Kachris and I. Tomkos, “A Survey on Optical Interconnects for Data Centers," Communications Surveys & Tutorials, vol. 14, no. 4, pp. 1021–1036, 2012.

[46] N. Farrington, G. Porter, S. Radhakrishnan, H. H. Bazzaz, V. Subramanya, Y. Fainman, G. Papen, and A. Vahdat, “Helios: a hybrid electrical/optical switch architecture for modular data centers,” ACM SIGCOMM Comput. Commun. Rev., vol. 40, no. 4, p. 339, 2010.

[47] G. Wang, D. G. Andersen, M. Kaminsky, K. Papagiannaki, T. S. E. Ng, M. Kozuch, and M. Ryan, “c-Through : Part-time Optics in Data Centers,” ACM SIGCOMM Comput. Commun. Rev., vol. 40, no. 4, pp. 327–338, 2010.

[48] Y. Yin, R. Proietti, X. Ye, C. J. Nitta, V. Akella, and S. J. B. Yoo, “LIONS: An AWGR-based low-latency optical switch for high-performance computing and data centers,”J. Sel. Top. Quantum Electron., vol. 19, no. 2, 2013.

[49] K. Chen, A. Singla, A. Singh, K. Ramachandran, L. Xu, Y. Zhang, X. Wen, and Y. Chen, “OSA: An optical switching architecture for data center networks with unprecedented flexibility,” Trans. Netw., vol. 22, no. 2, pp. 498–511, 2014.

[50] S. Yan, E. Hugues-salas, V. J. F. Rancaňo, Y. Shu, G. M. Saridis, R. Rofoee, Y. Yan, A. Peters, S. Jain, T. May-smith, D. J. Richardson, and G. Zervas, “Archon: A function programmable optical interconnect architecture for transparent intra and inter data center SDM/TDM/WDM networking,” J. Light. Technol., vol. 33, no. 8, pp. 1586–1595, 2014.

[51] J. Chen, Y. Gong, M. Fiorani, and S. Aleksic, “Optical interconnects at top of the rack for energy-efficient datacenters,” Commun. Mag., vol. 53, no. 8, pp. 140–148, 2015.

[52] W. Ni, C. Huang, Y. L. Liu, W. Li, K. W. Leong, and J. Wu, “POXN: A new passive optical cross-connection network for low-cost power-efficient datacenters,” J. Light. Technol., vol. 32, no. 8, pp. 1482–1500, 2014.

[53] V. Olmos, J. José, T. Monroy, N. Eiselt, J. Wei, H. Griesser, A. Dochhan, M. Eiselt, and J. Elbers, “First Real-Time 400G PAM-4 Demonstration for Inter-Data Center Transmission over 100 km of SSMF at 1550 nm,” Optical Fiber Communication Conference (OFC), 2016.

[54] J. Lee, S. Shahramian, N. Kaneda, Y. Baeyens, J. Sinsky, L. Buhl, J. Weiner, U. V. Koc, A. Konczykowska, J. Y. Dupuy, F. Jorge, R. Aroca, T. Pfau, and Y. K. Chen, “Demonstration of 112-Gbit/s optical transmission using 56GBaud PAM-4 driver and clock-and-data recovery ICs,” Eur. Conf. Opt. Commun. ECOC, vol. 2015–Novem, no. 1, pp. 4–6, 2015.

[55] F. Karinou, R. Rodes, K. Prince, I. Roudas, and I. T. Monroy, “IM / DD vs . 4-PAM Using a 1550-nm VCSEL

55

over Short- Range SMF / MMF Links for Optical Interconnects,” pp. 9–11, 2013.

[56] F. Karinou, L. Deng, R. R. Lopez, K. Prince, J. B. Jensen, and I. T. Monroy, “Performance comparison of 850-nm and 1550-nm VCSELs exploiting OOK, OFDM, and 4-PAM over SMF/MMF links for low-cost optical interconnects,” Opt. Fiber Technol., vol. 19, no. 3, pp. 206–212, 2013.

[57] J. Lee, P. Dong, N. Kaneda, and Y.-K. Chen, “Discrete Multi-Tone Transmission for Short-Reach Optical Connections,” Optical Fiber Communication Conference (OFC), 2016.

[58] A. Dochhan, H. Grieser, M. Eiselt, and J.-P. J.-P. Elbers, “Flexible bandwidth 448 Gb/s DMT transmission for next generation data center inter-connects,” European Conference on Optical Communication (ECOC), 2014.

[59] C. Cole, I. Lyubomirsky, A. Ghiasi, and V. Telang, “Higher-order modulation for client optics,” IEEE Commun. Mag., vol. 51, no. 3, pp. 50–57, 2013.

[60] K. Zhong, X. Zhou, T. Gui, L. Tao, Y. Gao, W. Chen, J. Man, L. Zeng, A. Pak, T. Lau, and C. Lu, “Experimental study of PAM-4 , CAP-16, and DMT for 100 Gb/s Short Reach Optical Transmission Systems,” Opt. Express, vol. 23, no. 2, pp. 2346–2349, 2015.

[61] Y. Kai, M. Nishihara, T. Tanaka, T. Takahara, Lei Li, Zhenning Tao, Bo Liu, J. C. Rasmussen, and T. Drenski, “C. Chen, X. Tang, and Z. Zhang, ‘Transmission of 56-Gb/s PAM-4 over 26-km Single Mode Fiber Using Maximum Likelihood Sequence Estimation,’ European Conference on Optical Communication (ECOC), 2013.

[62] A. Yekani, M. Chagnon, C. S. Park, M. Poulin, D. V Plant, and L. A. Rusch, “Experimental Comparison of PAM vs . DMT using an O-Band Silicon Photonic Modulator at Different Propagation Distances,” in Ecoc, 2015, no. 1, pp. 8–10.

[63] I. Geneva, “Discrete Multi-tone Technology for 100G Ethernet ( 100GbE ),” Ethernet Technol. summit, 2012.

[64] S. Bhoja, “400GE PAM Modulation for Single-Mode Fiber Data Centers,” Ethernet Technol. summit, 2014.

[65] B. LN and S. Itzhak, “An optical fiber dispersion measurement technique and system,” 2015.

[66] T. Udem, J. Reichert, R. Holzwarth, and T. Hänsch, “Absolute Optical Frequency Measurement of the Cesium $D_1$ Line with a Mode-Locked Laser,” Phys. Rev. Lett., vol. 82, no. 18, pp. 3568–3571, 1999.

[67] X. Zhou, X. Zheng, H. Wen, H. Zhang, and B. Zhou, “Generation of broadband optical frequency comb with rectangular envelope using cascaded intensity and dual-parallel modulators,” Opt. Commun., vol. 313, pp. 356–359, 2014.

[68] N. Cvijetic, “OFDM for next-generation optical access networks,” J. Light. Technol., vol. 30, no. 4, pp. 384–398, 2012.

[69] K. Qiu, X. Vi, H. Zhang, M. Deng, and C. Zhang, “OFDM-PON Optical Fiber Access Technologies,” In Asia Communications and Photonics Conference and Exhibition (ACP), 2011.

[70] Y. Cheng, M. Fiorani, L. Wosinska, and J. Chen, “Reliable and Cost Efficient Passive Optical Interconnects for Data Centers,” IEEE Commun. Lett., vol. 19, no. 11, pp. 1913–1916, 2015.

[71] M. P. Anastasopoulos, A. Tzanakaki, and D. Simeonidou, “Scalable Monitoring and Optimization Techniques for Megascale Data Centers,” J. Light. Technol., vol. 34, no. 8, pp. 1980–1989, 2016.

[72] X. Meng, V. Pappas, and L. Zhang, “Improving the Scalability of Data Center Networks with Traffic-aware Virtual Machine Placement,” in International Conference on Computer Communications (INFOCOM), 2010.

[73] A. Bechtolsheim, A. Networks, M. Paniccia, I. Fellow, and I. Corporation, “100G CLR4 Industry Alliance,” 2014.

[74] G. P. Agrawal, Fiber-Optic Communications Systems, Third Edition, 2002.

[75] G. P. Agrawal, Lightwave technology telecommunication systems, 2012.

[76] K. Szczerba, P. Westbergh, J. Karout, J. S. Gustavsson, Å. Haglund, M. Karlsson, P. a. Andrekson, E. Agrell,

56

and A. Larsson, “4-PAM for High-Speed Short-Range Optical Communications,” J. Opt. Commun. Netw., vol. 4, no. 11, pp. 885–894, 2012.

[77] J. G. Proakis and M. Salehi, Digital Communications, 5th ed. McGraw-Hill Education, 2007.

[78] C. Chan, K. Sherman, M. Zirngibl, “A fast 100-channel wavelength-tunable transmitter for optical packet switching,” IEEE Photon. Technol. Lett., vol. 13, no. 7, pp. 729–731, 2001.

[79] J. E. Simsarian, P. Bernasconi, J. Gripp, M. C. Larson, D. T. Neilson, “Fast Tunable Lasers for Optical Routers and Networks,” in Conference on Lasers and Electro-Optics and 2006 Quantum Electronics and Laser Science Conference (CLEO), 2006

[80] M. C. Inc, MRV fd edfa datasheet, ,online available: http://www.mrv.com/sites/default/files/datasheets/us_pdfs/mrv-fd-edfa_2.pdf.

[81] E. Desurvire, “Analysis of Transient Gain Saturation and Recovery in Erbium-Doped Fiber Amplifiers,” IEEE Photonics Technology Letters, vol. 1, no. 8. pp. 196–199, 1989.

[82] M. I. Hayee and A. E. Willner, “Transmission penalties due to EDFA gain transients in add-drop multiplexed WDM networks,” Photonics Technol. Lett., vol. 11, no. 7, pp. 889–891, 1999.

[83] J. R. F. de O. and A. C. B. H. S. Carvalho, I. J. G. Cassimiro, F. H. C. S. Filho, “AGC EDFA transient suppression algorithm assisted by cognitive neural network,” in Telecommunications Symposium (ITS), 2014.

[84] G. Goeger and B. Lankl, “Techniques for suppression of Raman and EDFA gain transients in dynamically switched transparent photonic networks,” in European Conference on Optical Communication (ECOC), 2002.

[85] Broadcom and Intel, “40G Ethernet Market Potential,” 2007.

[86] M. Chaciński, U. Westergren, R. Schatz, B. Stoltz, S. Hammerfeldt, and L. Thylén, “Monolithically integrated 100 GHz DFB-TWEAM,” J. Light. Technol., vol. 27, no. 16, pp. 3410–3415, 2009.

[87] Z. Feng,X. M. Tang, R. Lin, R. Wang, Q. Wu, L. Zhang, L. Xu, X. Wang, C. Zhou, J. Wu, S. Zhou, L. Deng, S. Fu, D. Liu, P.P. Shum, “SNR equalized optical direct-detected OFDM transmission with CAZAC equalization,” in In: Conference on Lasers and Electro-Optics Europe - Technical Digest, 2015.

[88] M. Barazande-pour, G. Koziuk, and John Khoury, “NRZ , PAM4 and Duobinary modulation schemes for 10G Serial Ethernet,” online available: http://www.ieee802.org/3/ap/public/may04/barazande-pour_01_0504.pdf

[89] Juniper, “Junos ® OS Ethernet Interfaces Feature Guide for Routing Devices,” online available: http://www.juniper.net/techpubs/en_US/junos14.1/information-products/pathway-pages/config-guide-network-interfaces/network-interfaces-ethernet.pdf.

57

Date post:	07-Sep-2020
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times

High Capacity Short-Reach Optical Communications1047182/... · 2016. 11. 16. · kilometers, and 2)...

Documents