Optically Amplified WDM Networks
This page intentionally left blank
Optically Amplified WDM Networks
John Zyskind Atul Srivastava
AMSTERDAM BOSTON HEIDELBERG LONDON NEW YORK OXFORD PARIS SAN DIEGO SAN FRANCISCO SINGAPORE SYDNEY TOKYO
Academic Press is an imprint of Elsevier
Academic Press is an imprint of Elsevier 30 Corporate Drive, Suite 400, Burlington, MA 01803, USA 525 B Street, Suite 1900, San Diego, CA 92101-4495, USA 84 Theobald’s Road, London WC1X 8RR, UK First edition 2011 Copyright Ó 2011 Elsevier Inc. All rights reserved No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without the prior written permission of the publisher Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333; email:
[email protected]. Alternatively you can submit your request online by visiting the Elsevier web site at http://elsevier.com/ locate/permissions, and selecting Obtaining permission to use Elsevier material Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the Library of Congress British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 978-0-12-374965-9 For information on all Academic Press publications visit our website at books.elsevier.com Printed and bound in USA 11 12 10 9 8 7 6 5 4 3 2 1
For the loving memory of my mother Maya, and to my wife Sonali, daughter Srishti and sister Sushma, with love - Atul Srivastava
Dedicated with love to the memory of my father Professor Harold Zyskind - John Zyskind
This page intentionally left blank
Contents Foreword ...................................................................................................................ix Author Biographies...................................................................................................xi CHAPTER 1
Optical Amplifiers for Next Generation WDM Networks: A Perspective and Overview .......................................................1
CHAPTER 2
ROADM-Based Networks...........................................................23
CHAPTER 3
Challenges and Opportunities in Future High-Capacity Optical Transmission Systems ..................................................47
CHAPTER 4
EDFAs, Raman Amplifiers and Hybrid Raman/EDFAs ....................83
CHAPTER 5
Dynamic and Static Gain Changes of Optical Amplifiers at ROADM Nodes ...................................................................117
CHAPTER 6
Mastering Power TransientsdA Prerequisite for Future Optical Networks....................................................155
CHAPTER 7
Spectral Power Fluctuations in DWDM Networks Caused by Spectral-Hole Burning and Stimulated Raman Scattering ..... 201
CHAPTER 8
Amplifier Issues for Physical Layer Network Control ................221
CHAPTER 9
Advanced Amplifier Schemes in Long-Haul Undersea Systems .................................................................253
CHAPTER 10 Challenges for Long-haul and Ultra-long-haul Dynamic Networks.................................................................277 CHAPTER 11 Transport Solutions for Optically Amplified Networks ............... 297 CHAPTER 12 Optical Amplifier for Maintenance Friendly Fiber Networks....... 341 CHAPTER 13 Low Cost Optical Amplifiers....................................................363 CHAPTER 14 Semiconductor Optical Amplifiers for Metro and Access Networks ............................................................387 CHAPTER 15 Market Trends for Optical Amplifiers.......................................417 Index ......................................................................................................................445
vii
This page intentionally left blank
Foreword Rod Alferness Bell Labs, Alcatel-Lucent The power of light, harnessed with an array of components to generate, modulate, manipulate, and detect it, and supported by low-loss optical fiber for transmission that ushered in a new era of information transmission systems in the 1970s, is an incredible gift to mankind. One could have hardly expected more, butdalmost on queuedthe invention and development of the optical amplifier in the late ’80s and early ’90s completed the technology suite, unleashing the full potential and power of optics for communication networks. The resulting cost-effective, robust, highcapacity optical networks, together with packet-based data networks that ride over them, enabled the world-wide web that has dramatically revolutionized our daily lives. The global growth of WDM (wavelength division multiplexing) optical networks over the last 10 years has been remarkable. While most optical networks are generally not directly visible to the typical consumer, the very visible internet would be impossible without them. Spanning continents, crossing oceans, reaching across metropolitan areas and now also providing direct fiber to home connections, commercially deployed optical transmission systems with per fiber capacity as high as several Terabit/sec provide the enabling high-capacity connectivity that underpins the world-wide web. No longer simple point-to-point links, today’s optical networks are flexible, switchable wavelength routed networks, both ring and mesh, that provide wavelength granular networked pipes inside the physical fiber with alloptical on and off ramps in much the way time slots are used in time-division-based transport networks. None of this would be possible without the optical fiber amplifier. The optical amplifier is truly a gift of nature that is as close to ideal as one could expect. It is spectrally matched to fiber’s low-loss window and provides highly efficient, broadband, low noise gain. Critical for its enabling of WDM, it has a temporal response that allows essentially unlimited signal data rates while allowing multiple wavelengths to be amplified without cross-talk between independent communication signals carried by neighboring wavelengths. The potential of WDM to tap the bandwidth of fiber, without requiring superhigh bit rates and the necessary enabling high-speed electronics, had been well known for some time. But, WDM was not a cost-effective solution for high-capacity systems as long as each wavelength channel had to be separated and regenerated individually by a discrete electronic regenerator. However, the optical amplifier, with its ability to amplify multiple wavelengths simultaneously, first and foremost, made DWDM (dense wavelength division multiplexing) the cost-effective approach to building very-high-capacity optical transmission systems. That capability aloned first demonstrated in commercial products in the mid-1990sdwas revolutionary.
ix
x
Foreword
What at the time was far less obvious to most, even if a few could foresee it, was that in enabling DWDM transmission, the optical amplifier was also preparing the way to a fundamentally new network architecture using wavelengths as the networked parameterdthe common unit of “currency” for enabling and managing a network. These WDM networks, while they also depended on an array of other new technologies, including, most importantly, optical switching elements to build the wavelength add/drop multiplexers and wavelength cross-connect networking elements, depended on optical amplifiers not only to enable WDM transmission but also to compensate for the losses in these switching elements. WDM networks offered the potential to provision, manage, and protect capacity based on wavelength “chunks” via fully flexible, switched wavelength networks. It is these WDM networks that are the focus of this book. While to many this vision appeared far-out, it was actually a very natural consequence of adopting WDM for transmission systems. Nevertheless, a tremendous world-wide research effort was required to provide the knowledge base needed to answer key questions, invent and develop new technologies, and refine and demonstrate the value proposition of WDM networks to convince service providers around the globe to deploy these networks for both long-haul and metro networks. The editors of Optically Amplified WDM Networks, John Zyskind and Atul Srivastava, who played key roles in taking optical networks from a vision to reality, have assembled a group of world-known researchers and engineers to address the critical areas of the field. This comprehensive book covers the broad areas important to WDM networks. From the dynamics of optical amplifiers critical to the inherent power transients in reconfigured networks, to basic (and not basic) amplifier design, to the considerations and design of wavelength add/drop multiplexers, to a perspective of future market trendsdall are well covered. Not limited to fiber amplifiersderbium-doped and Ramandthey also address the potential role of semiconductor amplifiers with its somewhat less ideal temporal characteristics but possible cost advantages, especially when integrated on a single photonic integrated circuit with other optical functions. That role seems particularly interesting for future metro and access applications. This book provides a wealth of information, insight, and reference information presented in many cases by the people who did the original work in the field. As such the book should prove very helpful to researchers and practicing engineers in or entering the field, including students. It is also a useful resource for researchers addressing the next frontier for optical networkingdhigh-speed optical packet switched networksdwhich is expected to benefit from many of the same technologies and is at a stage today that WDM networks were about 15 years ago.
Author Biographies CHAPTER 1. (ATUL SRIVASTAVA AND JOHN ZYSKIND) Atul Srivastava has over 30-years of research and development experience, and is credited with many advances in the field of optics, semiconductor opto-electronics, and high-capacity optical fiber networks. At Bell Laboratories he was responsible for several key inventions in optical amplifiers including the ultra-wideband EDFA, fast gain control in amplifiers and the first demonstration of the 100-channel long distance terabit capacity WDM transmission He facilitated founding of a start-up company, Onetta in 2000 and as the Vice President of Technology at Onetta, led research and development of optical amplifiers and WDM sub-systems. He is currently president of a new technology consulting startup OneTerabit. He is credited with over 100 publications and15 patents and is a recipient of the Bell Laboratories President’s Gold Award, the Trophee du Telephone. He is also a Fellow of Optical Society of America.
[email protected] John Zyskind received his bachelors degree from the University of Chicago and his Ph.D. from the California Institute of Technology where he was a Fannie and John Hertz Fellow. In 1982 he joined Bell Laboratories where he did pioneering research on optical amplifiers for DWDM optical networks, led optical amplifier research for the MONET optical networking program, was named Distinguished Member of Technical Staff and received the Bell Labs President’s Gold Award. Dr. Zyskind has directed the development of commercial Terabit/sec, ultralong haul optical network products at Sycamore Networks and of optically amplified systems for hut skipping applications at Optovia Corporation and JDSU. He is currently Director of System Engineering at Oclaro’s Transport Systems Solutions Division. Dr. Zyskind has published over 200 refereed papers and conference presentations, has delivered 35 invited talks, holds 26 patents and has published two book chapters.. He has taught 18 short courses at OFC and CLEO. Dr. Zyskind is a Fellow of the Optical Society of America.
[email protected] CHAPTER 2. ROADM BASED NETWORKS (BRANDON COLLINGS AND PETER ROORDA) Brandon Collings has over 15 years of optical networking research, design and development experience at Bell Laboratories, Internet Photonics, Ciena and JDSU.
xi
xii
Author biographies
He is currently the Chief Technology Officer for JDSU where he assists in the development of optical architectures and enabling components and sub-systems for next generation agile optical networks.
[email protected]. Peter Roorda has over 15 years experience in the design and development of agile optical networks and their key subsystems. In various technical roles at Nortel Networks and Innovance Networks, Peter has developed product concepts and architectures for emerging WDM, ROADM, optical amplifier and high speed transmission technologies. Currently at JDSU, Peter is product line manager for ROADM products.
[email protected] CHAPTER 3. CHALLENGES AND OPPORTUNITIES IN FUTURE HIGH-CAPACITY OPTICAL TRANSMISSION SYSTEMS (XIANG LIU) Xiang Liu received his Ph.D. degree in applied physics from Cornell University in 2000. He joined Bell Labs as a member of technical staff in 2000, and has been working on high-speed optical communication technologies since then. Dr. Liu has authored/coauthored over 190 journal and conference papers, and holds over 35 US patents. He is a senior member of the IEEE and the OSA.
[email protected] CHAPTER 4. OPTICAL AMPLIFIERS: CHALLENGES AND OPPORTUNITIES (JOHN ZYSKIND AND MAXIM BOLSHTYANSKY) Maxim Bolshtyansky received the M.S. in physics from Chelyabinsk Technical University, Russia in 1993, and the second M.S. and Ph.D. in optical physics from CREOL at University of Central Florida, Orlando, US in 1999. Since that time, he was working in various engineering and research roles at Lucent Technology, Onetta Inc, and is presently employed by JDSU. His research interests include detailed investigation and modeling of the gain media such as Raman and EDF and control algorithms for telecom applications.
[email protected] Author biographies
CHAPTER 5. DYNAMIC AND STATIC GAIN CHANGES OF OPTICAL AMPLIFIERS AT ROADM NODES (ETSUKO ISHIKAWA, SETSUHISA TANABE, MASATO NISHIHARA, AND YOUICHI AKASAKA) Etsuko Ishikawa Is currently director of Research at Fujitsu Ltd. Her achievements include development of S-band optical amplifier using silica-based erbium doped fiber; research in spectral hole burning mechanism in erbium doped fiber amplifier.
[email protected] Setsuhisa Tanabe received the BS, MS, and PhD degrees in material chemistry from Kyoto University, Japan, in 1986, 1988, and 1993. He became an Assistant Professor of Kyoto University, where he was promoted to a Full Professor in 2008 at Graduate School of Human and Environmental Studies. He is the author of more than 100 original papers, 22 books, and 23 invited review papers. He is also the holder of 24 patents on rare-earth doped optical amplifiers and glass ceramic phosphors for solid-state lighting.
[email protected] Masato Nishihara received the B.E. degree in Electrical Engineering in 1998 and the M.E. degree in Electronics Engineering in 2000 from the University of Tokyo, Tokyo, Japan. He joined Fujitsu Laboratories Ltd., Kawasaki, Japan in 2000 and engaged in the research and development of the optical and electrical devices for the long-haul optical fiber transmission system.
[email protected] Youichi Akasaka of Fujitsu Laboratories of America has been working in the telecommunications industry since 1993, focusing on photonics innovation. He covers diverse areas of optical communications from components to system/ network. He received the B.S. degree from Kyoto University and M.S. and Ph.D. degrees from University of Tokyo. He received the IEICE Young Engineer Award for his pioneering work on optical fiber design in 1995.
[email protected] CHAPTER 6. MASTERING POWER TRANSIENTS A PREREQUISITE FOR FUTURE OPTICAL NETWORKS (PETER KRUMMRICH) P. M. Krummrich received his Dr.-Ing. degree in Electrical Engineering from Technische Universitaet Braunschweig, Germany, in 1995, where he worked on Praseodymium-doped fiber amplifiers. In 1995 he joined Siemens AG where his research interest focused on distributed Erbium-doped fiber amplifiers, Raman
xiii
xiv
Author biographies
amplification, power transients, advanced modulation formats, adaptive equalizers, and PMD compensation. Since 2007 he is holding the chair for high frequency technology as full professor at Technische Universitaet Dortmund.
[email protected] CHAPTER 7. SPECTRAL POWER FLUCTUATIONS IN DWDM NETWORKS CAUSED BY SPECTRAL-HOLE BURNING AND STIMULATED RAMAN SCATTERING (JOERG-PETER ELBERS AND CORNELIUS FUERST) Dr. Jo¨rg-Peter Elbers received the diploma and the Dr.-Ing. degree in electrical engineering from Dortmund University, Germany, in 1996 and 2000, respectively. From 1999-2001 he was with Siemens AG e Optical Networks, last as Director of Network Architecture in the Advanced Technology Department. In 2001 he joined Marconi Communications (now Ericsson) as Director of Technology in the Optical Product Unit. Since September 2007 he is with ADVA AG Optical Networking, where he is currently Vice President Advanced Technology in the CTO office. Dr. Elbers authored and co-authored more than 70 scientific publications and 15 patents. He is member of the IEEE LEOS (Laser and Electro-Optics Society) as well as the German VDI (Association of German Engineers) and VDE (German Association for Electrical, Electronic & Information Technologies). Dr. Elbers serves in technical programme committee of the European Conference on Optical Communication (ECOC). He is also member of the VDE expert committee for optical communications engineering.
[email protected] Cornelius Fuerst received the diploma and Ph.D. in physics from the Technical University of Munich, Germany, in 1995 and 1998, respectively, where he did research on femtosecond pulse lasers and ultrafast quantum effects of semiconductors. In 1998 he joined the fiber communication industry working for Siemens Optical Networks, Marconi Communications, Ericsson and ADVA Optical Networking (since 2009). Cornelius Fu¨rst filed more than 40 publications and 10 patent applications in the field of optical networking.
[email protected] CHAPTER 8. AMPLIFIER ISSUES FOR PHYSICAL LAYER NETWORK CONTROL (DANIEL C. KILPER AND CHRISTOPHER A. WHITE) Dr. Daniel Kilper is currently a member of the Bell Laboratories Optical Transmission Systems and Networks Research Department at Alcatel-Lucent. He received
Author biographies
BS degrees in Electrical Engineering and Physics from the Virginia Polytechnic Institute and State University in 1990 and the PhD and MS degrees in physics from the University of Michigan, Ann Arbor in 1992 and 1996. He was a research scientist at the Optical Technology Center at Montana State University before serving as an assistant professor in physics at the University of North Carolina at Charlotte until 2000. He is a senior member of IEEE and an associate editor for the OSA/ IEEE Journal of Optical Communications and Networking. He currently serves as interim chair of the GreenTouch Consortium technical committee. While at Bell Laboratories he has conducted research on optical performance monitoring, network energy trends, and on transmission, architectures and control systems for transparent and re-configurable optical networks. He has authored or co-authored more than 80 journal publications and conference presentations, three book chapters and six patents.
[email protected] Christopher A. White is a distinguished member of technical staff in the Bell Labs’ Chief Scientist’s Office. He holds a Ph.D. in theoretical quantum chemistry from the University of California, Berkeley. His research interests include the simulation and control of complex physical systems ranging from optical networks, to the next generation of smart power grid, and to the propagation of ideas in organizations.
[email protected] CHAPTER 9. ADVANCED AMPLIFIER SCHEMES IN LONG-HAUL UNDERSEA SYSTEMS (ALAN LUCERO) Dr. Alan J. Lucero received his M.S. and Ph.D. in physics in 1989 and 1993 from the University of Connecticut. After completing a two-year postdoctoral fellowship at Bell Laboratories and two years at AT&T Advanced Technologies Systems, he assisted in the establishment of the Photonics Research and Test Center for Corning, Inc.. He joined Tyco Telecommunciations in 2000, where his current concentrations include 10, 40, and 100 Gb/s transport, advanced optical amplification schemes, coherent transmission, novel dispersion maps, and Q-fluctuation statistics. Dr. Lucero is a member of Phi Beta Kappa
[email protected] CHAPTER 10. CHALLENGES FOR LONG HAUL AND ULTRA-LONG HAUL DYNAMIC NETWORKS (MARTIN BIRK AND KATHY TSE) Martin Birk received his M.S. and Ph.D. degrees from University of Ulm, Germany, in 1994 and 1999, respectively. Since 1999, he has been with AT&T Labs
xv
xvi
Author biographies
in New Jersey, working on high-speed fiber optic transmission at data rates of 40Gbit/s and above.
[email protected] Kathy Tse leads a team at AT&T working on Optical Systems performance and requirements. She has worked in the fiber optics are at AT&T since 1985. She received her MS and PhD from Brown University and BSc from Cornell.
[email protected] CHAPTER 11. TRANSPORT SOLUTIONS FOR OPTICALLY AMPLIFIED NETWORK (WERNER WEIERSHAUSEN AND MALTE SCHNEIDERS) Werner Weiershausen received his Dipl.-Ing. degree in electrical and RF engineering from Technical University of Braunschweig in 1992, including AlGaAs based VCSEL research at University of Ulm. From 1992 to 1997 he has been with the Research Center of Deutsche Bundespost/T-Nova and 1997 with the University of Technology, Darmstadt, as a scientist in the fields of InP based semiconductor technology, integrated optics and fiber components. In 1998 he joined the Photonic Systems and Optical Networks Group at T-Systems, first working on theoretical and experimental research in the fields of high-speed optical WDM transmission and measurement methodology, later being project leader for different projects on R&D and technical consulting for optical networks. 2008 he changed to the Technical Engineering Center of Deutsche Telekom, working on the strategic evolution of the next-generation optical packet platform. Werner Weiershausen has been active in several national and European R&D projects (ACTS, IST, COST, BMBF) and different standardization bodies (ITU-T, IEC, DKE). Since 2003 he has been serving as SPIE Editor, Symposium Chair for Optics East Symposium and Conference Chair at Photonics West, USA. He is author or co-author of more than 80 publications, conference contributions and patents. Since August 2008 Werner Weiershausen is working for the management board (in the role of CSO) of the Finnish startup company Luxdyne Ltd, Helsinki, on sub systems for optical access (FTTH, PON).
[email protected] Malte Schneiders has more than nine years experience in the area of Optical Transport. He received his Diploma degree in electrical engineering from the Technical University of Dortmund, Germany in 2001. During his employment at Deutsche Telekom Group he has contributed already to several strategic projects, as well as to national and international research activities on the optimization of optical transport networks and high-speed transmission systems. Malte has authored or
Author biographies
co-authored more than 40 publications in scientific journals and conference proceedings and five patents in this field of investigations.
[email protected] CHAPTER 12. OPTICAL AMPLIFIER FOR MAINTENANCE FRIENDLY FIBER NETWORKS (GLENN A. WELLBROCK AND TIEJUN J. XIA) Glenn Wellbrock is the Director of Optical Transport Network Architecture and Design at Verizon, where he is responsible for the development of new technologies for both the metro and long haul transport infrastructure. In addition to his 20+ years at Verizon (1984-2001 & 2004-present), Glenn worked at Marconi and Qplus Networks.
[email protected] Dr. Tiejun J. Xia is a Distinguished Member of Technical Staff at Verizon. He was a faculty member at the University of Michigan. He holds his Ph.D. degree from the University of Central Florida, M.S. degree from Zhejiang University, and B.S. degree from University of Science and Technology of China. He has published more than 100 technical papers and holds more than 40 granted or pending U.S. patents.
[email protected] CHAPTER 13. LOW COST OPTICAL AMPLIFIERS (BRUCE NYMAN AND GREG COWLE) Bruce Nyman is currently with Tyco Electronics SubCom where he works on next generation undersea systems. From 2005 to 2009 he was with Princeton Lightwave as Vice President of system solutions. Previously, he developed optical amplifiers and measurement equipment at JDS Uniphase and optically amplified undersea systems at AT&T Bell Laboratories. He received his doctorate from Columbia University and is an IEEE fellow.
[email protected] CHAPTER 14. SEMICONDUCTOR OPTICAL AMPLIFIERS FOR METRO AND ACCESS NETWORKS (LEO SPIEKMAN AND DAVID PIEHLER) Leo Spiekman is the Director of Amplifier Products at Alphion Corporation, Princeton Junction, NJ, where he is responsible for the development of
xvii
xviii
Author biographies
semiconductor devices for photonic amplification and switching, for applications in optical telecommunications and beyond. He has served in technical and leadership roles for several technical conferences, among which OFC and ECOC. He has (co-)authored over 60 contributed and invited papers, four chapters in books, and has three patents.
[email protected] David Piehler is an innovator and leader in the deployment and development of FTTH and HFC broadband access networks and in their underlying technologies. At Fields and Waves, he advises clients on technology and markets for nextgeneration broadband access at physical, network and services layers. He presently plays a leading role in the definition of next-generation networks, including OFDM-PON, 10G-PON and RFoG. Dr. Piehler received a Ph.D. in physics from the University of California at Berkeley for experimental work in nonlinear optics.
[email protected] CHAPTER 15. MARKET TRENDS FOR OPTICAL AMPLIFIERS (DARYL INNIS) Dr. Inniss is the Component Practice Leader of Ovum’s Telecom research. Ovum is an ICT market research firm and Dr. Inniss’ research includes optical components for telecommunication and enterprise networks. Prior to joining Ovum Dr. Inniss was a Technical Manager at JDS Uniphase, and at Lucent Technologies, Bell Laboratories. Dr. Inniss holds a PhD in Chemistry from UCLA and an AB from Princeton University.
[email protected] CHAPTER
Optical Amplifiers for Next Generation WDM Networks: A Perspective and Overview
1
Atul Srivastava*, John Zyskind** * OneTerabit, Morganville, NJ, USA, ** Oclaro, Inc., Acton, MA, USA
CHAPTER OUTLINE HEAD 1.1. Introduction ..................................................................................................... 1.2. Optical amplifiers: recent developments............................................................. 1.2.1. Wideband amplifiers ....................................................................... 1.2.2. Agile amplifiers .............................................................................. 1.2.3. Cost reduction and commoditization of amplifiers ............................. 1.2.4. Standardization of amplifiers ........................................................... 1.3. Optical amplifiers: present status ................................................................... 1.4. Chapter overviews ......................................................................................... 1.4.1. Chapter 2. ROADM-based networks (Brandon Collings and Peter Roorda) ...................................................................... 1.4.2. Chapter 3. Challenges and opportunities in future high-capacity optical transmission systems (Xiang Liu) ...................................... 1.4.3. Chapter 4. EDFAs, Raman Amplifiers and Hybrid Raman/EDFAs (John Zyskind and Maxim Bolshtyansky)....................................... 1.4.4. Chapter 5. Dynamic and static gain changes of optical amplifiers at ROADM nodes (Etsuko Ishikawa, Setsuhisa Tanabe, Masato Nishihara, and Youichi Akasaka)....................................... 1.4.5. Chapter 6. Mastering power transients: a prerequisite for future optical networks (Peter Krummrich) ................................... 1.4.6. Chapter 7. Spectral power fluctuations in DWDM networks caused by spectral-hole burning and stimulated Raman scattering (Joerg-Peter Elbers and Cornelius Fuerst) ..................................... 1.4.7. Chapter 8. Amplifier issues for physical layer network control (Daniel C. Kilper and Christopher A. White) .................................. 1.4.8. Chapter 9. Advanced amplifier schemes in long-haul undersea systems (Alan Lucero)................................................... 1.4.9. Chapter 10. Challenges for long-haul and ultra-long-haul dynamic networks (Martin Birk and Kathy Tse)........................................... 1.4.10. Chapter 11. Transport solutions for optically amplified networks (Werner Weiershausen and Malte Schneiders) ............................... Optically Amplified WDM Networks. DOI: 10.1016/B978-0-12-374965-9.10001-9 Copyright Ó 2011 Elsevier Inc. All rights reserved.
2 3 3 3 8 9 11 11 11 12 13
13 14
15 15 16 16 17
1
2
CHAPTER 1 Optical amplifiers for next generation WDM networks
1.4.11. Chapter 12. Optical amplifier for maintenance-friendly fiber networks (Glenn A. Wellbrock and Tiejun J. Xia) .................... 1.4.12. Chapter 13. Low-cost optical amplifiers (Bruce Nyman and Greg Cowle) ................................................... 1.4.13. Chapter 14. Semiconductor optical amplifiers for metro and access networks (Leo Spiekman and David Piehler)................. 1.4.14. Chapter 15. Market trends for optical amplifiers (Daryl Inniss) ....... Acronyms ............................................................................................................. Acknowledgements ............................................................................................... References ...........................................................................................................
17 18 19 19 20 21 21
1.1 INTRODUCTION Since its invention in 1987, the erbium-doped fiber amplifier (EDFA) [1,2] has revolutionized the telecommunications industry. Today the EDFA is widely viewed as mature technology, while new network applications drive new requirements for further enhancement of optical fiber amplifiers. One such emerging application is dynamic optical networking using remotely reconfigurable optical add/drop multiplexers (ROADMs). With the ability to express optical channels through these ROADM nodes, service providers can reduce per wavelength cost in the network. On the other hand, longer reach between regeneration is needed to derive benefits from this network architecture, making the optical amplifier a key component in these reconfigurable dynamic networks. Amplifier designs will also be influenced by the renewed need for increased capacity on a fiber. Deployment of wavelength division multiplexed (WDM) systems transmitting 40 Gb/s per WDM channel is already under way, and a great deal of current work is directed at 100 Gb/s transmission. Meeting the demands for higher capacity within the constraints of current fiber plant and filtering limitations of wavelength switches will require advanced modulation and detection formats that encode multiple bits per symbol and, in many cases, involve differential or coherent detection. In general, these modulation schemes increase robustness against chromatic dispersion (CD), polarization mode dispersion (PMD), and filtering impairments, but place additional demands on the noise performance of the optical amplifiers due to their increased susceptibility to nonlinear impairments and, frequently, increased optical signal-to-noise ratio (OSNR) requirements. On the other hand, the amplifier characteristics, such as the gain spectrum, its dependence on channel loading, the cross saturation and dynamic response of the amplifier gain, and the optical noise added by the amplifiers, must all impose constraints on the design and operation of advanced networks. These important network design issues are addressed in a number of chapters in this book as well as the chapter summaries that follow. This book is intended to address recent developments in amplifiers driven by emerging trends in network applications as well as the interrelationship between
1.2 Optical amplifiers: recent developments
the amplifiers and these emerging trends. Experts on the amplifiers, on the network equipment in which they are used, and on the service provider networks in which the equipment is deployed were asked to examine how the emerging network trends drive amplifier requirements, how amplifier characteristics guide or constrain the design and operation of networks with more complex optical connectivity and higher capacities, and how, as a result, the engineering of solutions to various network design and operation challenges increasingly require network-wide analysis and network-wide solutions. Application spaces ranging from low-cost, shortreach access networks to terrestrial networks to ultra-long-haul transcontinental undersea systems are considered. In the remainder of this chapter, the recent trends in optical amplifiers that are driven in large part by emerging trends in network applications are discussed, then summaries of the remaining chapters are provided.
1.2 OPTICAL AMPLIFIERS: RECENT DEVELOPMENTS Four major trends characterize developments in the amplifiers over the past decade. These include the increase in amplifier bandwidth to support larger numbers of channels for higher network capacity, integration of optics and fast electronics to provide performance agility for dynamic networks, significant cost reduction, and standardization of amplifier modules. These trends are summarized below.
1.2.1 Wideband amplifiers In response to the exponential growth in data traffic, optical networks were designed with a growing number of dense wavelength division multiplexing (DWDM) channels. State-of-the-art commercial DWDM systems include EDFAs able to support up to 160 channels with 50 GHz channel spacing spanning both C- and L-bands. However, the deployed networks predominantly consist of channels in the C-band. Some geographical areas such as Japan have deployed amplifiers to support transmission systems in the L-band over the dispersion shifted fiber spans. Moreover, the use of a hybrid design of EDFAs in conjunction with Raman amplifiers is prevalent in special networks with ultra-long spans.
1.2.2 Agile amplifiers There is a large market for optical amplifiers that until recently was driven by static point-to-point WDM system applications. However, the increasing unpredictability of traffic demand and the emergence of bandwidth-hungry applications such as video on demand have recently turned the industry’s focus to dynamic, reconfigurable optical networks based on ROADMs to route different wavelength channels. Sales of ROADMs are increasing with a compound annual growth rate exceeding 40%; by the end of 2010 the number of field deployed wavelength
3
4
CHAPTER 1 Optical amplifiers for next generation WDM networks
selective switches (WSS), a key component of modern ROADMS, is expected to exceed 100,000. Essentially all new metro, regional, and long-haul WDM system products developed by equipment manufacturers offer ROADM-based wavelength agility as a key feature. Similarly, all network deployments planned by Tier-1 carriers in the major industrial nations and many in emerging nations require reconfigurable wavelength agility to reduce operational expenses and increase service velocity. A recent report from Ovum-RHK (see Fig. 1.1 below) concluded that ROADMs and agile EDFAs lead the transition to a dynamic optical network. The report forecast that the growth of these two components will track each other closely with over 80% of 2010 module sales in both categories expected for agile components. This has created new challenges for the design of amplifiers including, for example, those related to dynamics and control of transients, spectral hole burning, and polarization-related impairments. Introduction of channels at higher data rates employing a coherent receiver, followed by high-speed digitization and signal processing, can pose additional constraints on the system tolerance due to their sensitivity to channel power fluctuation caused by transients. Increasingly the optical amplifier technical literature and commercial practice have dealt with such design challenges posed by dynamic, reconfigurable optical networks. Today specifications for new optical amplifier designs are almost always framed to support the requirements of reconfigurable optical networks. These developments have led to the realization that the optical amplifiers constituting the transmission path need to provide performance agility in order to provide high-quality service for the channel traffic propagating through the
FIGURE 1.1 Penetration of agile ROADM and EDFA modules as a percentage of revenue by product group, 2005e11. Source: Ovum-RHK
1.2 Optical amplifiers: recent developments
amplifier. In the network architecture, two types of events have to be considered: intentional reconfiguration or re-provisioning and unintended failures or faults leading to protection path switching. In both types of events the optical amplifiers in the transmission path need to have fast gain control, which is usually implemented via electronic control of the amplifier’s pump lasers. In order to effectively control the power transients in DWDM systems, it is important to understand the factors that influence the speed or the rate at which the surviving channel powers change during transients. This speed depends on both the EDFA characteristics and the number of EDFAs constituting the system [3]. In a single EDFA, the time constant of the power transients decreases with increases in saturated power output. In a system with long chains of EDFAs, the amplifiers strive to maintain the saturated output power levels. The time varying output of the first EDFA (after a fiber span loss) appears as an input to the second EDFA, which has time-dependent gain, and therefore the output power of the second EDFA changes at a faster rate. Consequently, with increasing numbers of amplifiers in the chain, the speed of transients becomes faster and faster, thereby requiring control on shorter and shorter time scales to limit the power excursion of surviving channels [4]. Proper transient control design needs to take into account the fastest single event in the network as well as the acceleration of this event due to the cascading of amplifiers. Another important characteristic for the control of the transients is the slew rate associated with the event causing a change in the channel loading of amplifiers which determines the rate of the surviving channels’ power change during transients. Events such as failure of a transmitter laser can happen over sub micro second time scales and connector pulls, which can happen by mistake, typically occur over hundreds of ms, whereas fiber breaks often happen over hundreds of milliseconds (ms) or even seconds, although in a rare worst case they can be much faster. Likewise, provisioning of channels can be implemented in a controlled fashion, one at a time, to minimize sudden changes in the power of the transmitting channels. The state-of the art switching technologies currently being implemented in ROADMs also have transition times of hundreds of milliseconds. Today’s predominantly data-centric transport network has evolved from the synchronous optical network (SONET)-based voice-centric one in the past. The availability of high-bandwidth OC-768 and higher rate port cards for the asynchronous transfer mode (ATM) switches and IP routers has led to the development of an intelligent optical layer network, where light wave paths provide connectivity between SONET terminal, ATM, and IP router devices. The function of the optical layer is to provide high-bandwidth wavelength service to the client layer. A DWDM system supports multiple wavelengths over the optical bandwidth of the optical amplifier, unlike the single-wavelength SONET systems of the past. The DWDM system is designed to support channels over the bandwidth of the in-line optical amplifiers. Though the restriction on the channel powers for acceptable quality of service is determined by the details of the system architecture such as
5
6
CHAPTER 1 Optical amplifiers for next generation WDM networks
OSNR and non-linear effects in the fiber, typically the transmission system is designed with about 10 dB power margin at the receiver to accommodate these variations. At high channel powers and channel counts, optical non-linear effects, four photon mixing (FPM), cross-phase modulation (CPM), and Raman effect can impair the service quality of the system and thereby further limit acceptable channel power excursion. In the event of a fiber cut or inadvertent disconnection of one of the tributaries carrying one or more wavelengths entering the optical line system, the SONET protection switching could be triggered for the express channels, and they could be routed through the protection path. This protection switching would occur if the express channel powers experienced either an upward or downward power surge due to transients in the EDFAs. The following sequence of events could occur in a SONET system. First, if the optical power of the channel either exceeds or dips below the set limits, the system would generate an alarm to the network control and management (NC&M) system. Second, if the power excursion exceeds the dynamic range of the receiver, either due to poor signal or failure in the receiver thresholding circuit, a LOS (loss of signal) defect will be detected when an all-zero pattern on the incoming SONET signal lasts 100 ms or longer [5]. Finally, the signal has to be switched to the protection path within 50 to 100 ms via SONET/SDH (synchronous digital hierarchy) or Optical 1:1 protection [6]. For the surviving express channels, it is desirable that the power transients are mitigated within the physical layer to restrict the power transients to the minimum possible under both the intentional add/drop event during network reconfiguration or accidental events so that unnecessary SONET protection switching can be avoided. In order to implement the fast gain control to limit the power transients to an acceptable level, it became essential to integrate the optical gain block assembly with fast electronics to form a gain controlled module. Controlled amplifier modules with 50 to 100 microsecond control loop time constant were developed six to eight years ago. An example of the performance of an amplifier module with ultra-fast gain control is shown in Fig. 1.2. When 99 out of 100 channels propagating through the amplifier are interrupted, the surviving channel would experience only about 2 dB of power overshoot. The fast gain control circuit returns the power of the surviving channel to within 3) of connectivity and rapid provisioning and restoration capability and the introduction of channel rates of 100 Gb/s and greater, the requirements of higher degree nodes with colorless, contentionless, and gridless features are being considered and defined. The challenge for the network equipment vendors lies in the identification and implementation of key optical component technologies in order to meet the current and future needs of the service providers.
1.4 Chapter overviews
This chapter starts with a brief history of the DWDM system capacity and network evolution and provides a review of the current status of optical mesh backbone network from point of view of a North American network operator. Requirements for the optical amplifiers in the present day ROADM based mesh networks are described to include optical link control, and key characteristics of static and dynamic amplifiers are discussed. Finally, the requirements for fully dynamic future mesh networks are described with the possibilities ranging from precabling provisioning to full photonic restoration.
1.4.10 Chapter 11. Transport solutions for optically amplified networks (Werner Weiershausen and Malte Schneiders) In recent years, the distinction between the system reach of metro and long-haul systems has diminished. Today’s metro systems can be easily extended to cover distances longer than 500 km. This development has profound impact on the architecture of the European national networks, where the size of networks is closer to that of current metro/regional networks. Moreover, since signal wavelengths can be extended to distances nearing 100 km without re-amplification, the requirement of a large number of central offices in the legacy national networks in countries like Germany has also diminished. It is therefore more cost effective to consider network architectures with fewer central offices, potentially eliminating up to 90% of the existing ones. Cost benefits will also be derived from the implementation of new technologies such as new modulation formats and coherent detection, particularly for high-data-rate channels. In this chapter, network architecture issues and physical impairments due to chromatic dispersion, polarization mode dispersion (PMD), and related mitigation techniques for these networks are described from the point of view of a European network operator. Simulation results are also presented to show a comparison of the system reach over different types of fiber for signals with return to zero (RZ) and carrier suppressed RZ (CSRZ) modulation formats and the use of Raman and EDFA hybrid amplifiers to optimize the network design.
1.4.11 Chapter 12. Optical amplifier for maintenance-friendly fiber networks (Glenn A. Wellbrock and Tiejun J. Xia) Managing necessary fiber maintenance activities is a significant challenge for network operators. Traffic must be rerouted from an active fiber to an inactive fiber, without reducing service availability or violating service level agreements by triggering protection switches. In SONET-based ring networks with point-to-point wavelength paths, it is relatively straightforward to switch at the synchronous optical network (SONET) level all of the affected traffic onto a protection fiber link. The deployment of optically transparent mesh networks based on ROADM nodes with multiple fiber degrees, where different wavelengths on a fiber span follow diverse paths through the network, make the task of identifying and switching all of the
17
18
CHAPTER 1 Optical amplifiers for next generation WDM networks
affected client traffic paths much more difficult. With large numbers of narrowly spaced DWDM channels and channel rates increasing from 10 Gb/s to 40 Gb/s and then to 100 Gb/s and beyond, the task of moving traffic at the electronic level is becoming all the more intractable. In Chapter 12 the authors review these emerging trends in network architecture and discuss the challenges these trends present for network maintenance. Finally, an elegant solution is proposed which avoids the complications entailed in optically transparent mesh networks with high capacity DWDM traffic. The principle is to switch the traffic at the fiber level at the amplifier nodes which terminate either end of the affected span so that all of the affected traffic and only the affected traffic is switched. This requires a new architecture for optical amplifier repeater nodes incorporating the necessary switching capability at the amplifier input and the amplifier output. Switching architectures and technologies, which must support the required switching speed to avoid triggering of protection switches or unacceptably long traffic interruptions, are reviewed.
1.4.12 Chapter 13. Low-cost optical amplifiers (Bruce Nyman and Greg Cowle) Cost reduction of optical amplifiers is of increasing concern because of continual pressure on the pricing of optical networking equipment, because of changes in applications and network architectures which are extending the range of applications of amplifiers beyond the line amplifier repeaters of the core network, and because the dominant EDFA technology is not as easily amenable to cost reduction through integration as other technologies such as semiconductors. This chapter examines the issues involved in lowering the cost of optical amplifiers focusing on single stage optical amplifiers because, by and large, they will be used in the highest volume, most cost sensitive applications, such as metro and access network line amplifiers, single-channel amplification for high speed, advanced modulation format channels, cable television (CATV) distribution booster amplifiers, and ASE sources for WDM passive optical networks (PONs). The alternative technologies for low-cost amplifiers, such as semiconductor optical amplifiers, and erbium-doped waveguide amplifiers (EDWAs) are covered. EDFAs, which are the dominant technology, comprise multiple components with different features and are based on different technologies. The challenges and opportunities of reducing the costs of the primary components of EDFAs and the labor costs of assembling EDFAs are discussed. EDWAs offer opportunities for cost reduction by integrating the features of many of the components required for optical amplifiers. However, the lower efficiency of converting pump-to-signal power in erbium-doped planar waveguides compared with erbium-doped fiber, poses an obstacle to the commercial realization of the potential cost advantages of EDWAs. A recent approach is the PLC erbium-doped fiber amplifier, in which many of the passive devices are integrated on a PLC but the gain is provided by an erbium-doped fiber. This approach combines the cost advantages of PLC integration with the
1.4 Chapter overviews
performance and pump efficiency of erbium-doped fiber and is especially advantageous for complex amplifier architectures requiring many optical components.
1.4.13 Chapter 14. Semiconductor optical amplifiers for metro and access networks (Leo Spiekman and David Piehler) In 1990s, semiconductor optical amplifiers (SOAs) were expected to emerge as a compact and low-cost alternative to EDFAs. However, because of the short time constant of their gain dynamics, the SOAs caused unacceptable levels of inter- and intra-channel distortions making them unsuitable for in-line amplification in DWDM systems. Schemes such as laser cavity formation to clamp the amplifier gain led to many years of research and development activity without significant commercial success. The ability of SOAs to amplify signals in the wavelength regions beyond the C- and L-band covered by the EDFAs did not give them much benefit because the commercial DWDM systems remain confined to these bands. The other proposed applications of SOAs such as switching, non-linear signal processing and wavelength conversion have also been confined to academic research. With the advent of PONs for access applications in recent years, however, there is renewed interest in SOAs to extend the total loss budget of such networks. This chapter addresses the architecture issues of PONs to show the benefits of including SOAs for network reach extension, for example. The standards of gigabit passive optical network (GPON), for example, specify a loss budget of 28 dB to be used in a combination of splitting loss and fiber loss. This limitation can be overcome by the inclusion of SOAs at the splitter as a mid-span amplifier. The ability of SOAs to amplify signals at the wavelengths 1550, 1310, and 1490 nm in PON standard is also discussed.
1.4.14 Chapter 15. Market trends for optical amplifiers (Daryl Inniss) At the peak of the telecom bubble in 1999e2000, the size of the optical amplifier market was projected to grow to $5 billion in five years. However, 10 years later the market size is more than one order of magnitude lower at $300 million, and is projected to remain on a plateau over the next five years. The market stagnation is caused by the continual demand for amplification cost reduction, which nearly compensates the 10% to 15% year-to-year growth in number of optical amplifier (OA) units deployed. Carriers are not willing to pay a premium for extra performance such as rapid gain control to suppress transients in the network. The market is therefore dominated by three or four incumbent suppliers, who also control the expensive in feeds and can offer integrated higher level solutions, such as a ROADM card incorporating WSS and OAs and optical performance monitoring (OPM). There has not been a new OA start-up in more than five years. There are some emerging bright spots such as the expected growth for low-cost, single-channel amps including SOAs for the PON application and high-bit-rate power boosters and pre-
19
20
CHAPTER 1 Optical amplifiers for next generation WDM networks
amps. The other area attracting market attention is the Raman amplifier, which is expected to rapidly grow in market size either as a stand-alone unit or in conjunction with the EDFA to support the higher OSNR requirement of the 100 Gb/s and higher data rate channels. This chapter presents an analysis of the market forces contributing to the present trends in the amplification segment. Current market drivers include the traffic growth in the core network, the move to narrower channel spacing in the metro area networks, and rapid introduction of ROADMs and 40 Gb/s and 100 Gb/s channels. The author also highlights the emerging market for the single-channel amplifiers in the access PON and high-power amplifiers in the CATV distribution fiber to the x (FTTx) networks. There is considerable pricing pressure on the amplifier suppliers, and the emergence of network equipment vendors from east Asia has posed even an greater challenge. The control of key components and ability to make higher functional modules combining amplifiers for example with dispersion compensating modules (DCMs), WSSs, etc. has prevented new entrants to the amplifier supplier market. At the same time there is a crowding of the amplifier vendor space and not yet enough consolidation. In summary, the author points out that while the amplifier market is in an evolutionary phase characterized by both steady growth and significant commercial challenges, there are opportunities for innovation and room for entrants with novel solutions.
ACRONYMS AGC ATM CATV CD CPM CSRZ DCM DWDM EDFA EDWA FPM FTTx GPON IEC LOS NC&M NEM OA
Automatic gain control Asynchronous transfer mode Cable television (or community antenna television) Chromatic dispersion Cross phase modulation Carrier suppressed return to zero Dispersion compensating module Dense wavelength division multiplexing Erbium-doped fiber amplifier Erbium-doped waveguide amplifier Four photon mixing Fiber to the x (x ¼ home, building, curb, .) Gigabit passive optical network (ITU-T G.984) International Electrotechnical Commission Loss of signal Network control and management Network equipment manufacturer Optical amplifier
Acknowledgements
OFDM OPM OSNR PDL PM-QPSK PLC PMD PON ROADM RZ SHB SOA SDH SONET SRS VOA WDM WSS
Orthogonal frequency division multiplexing Optical performance monitoring Optical signal-to-noise ratio Polarization dependent loss Polarization multiplexed quadrature phase shift keying Photonic light wave circuit Polarization mode dispersion Passive optical network Reconfigurable optical add/drop multiplexer Return to zero Spectral hole burning Semiconductor optical amplifier Synchronous digital hierarchy Synchronous optical network Stimulated Raman scattering Variable optical attenuator Wavelength division multiplexer Wavelength selective switch
ACKNOWLEDGEMENTS We are sincerely grateful to Tim Pitts and Melanie Benson of Elsevier for their gracious support and patience during the whole process of publication of this book. We are also thankful to the authors for their efforts in preparing the manuscript highlighting the recent developments in their respective fields. Finally, we hope the reader will enjoy reading this book and benefit from its contents.
References [1] E. Desurvire, J.R. Simpson, P.C. Becker, High-gain erbium-doped fibre amplifier, Opt. Lett. 12 (11) (1987) 888. [2] R.J. Mears, L. Reekie, I.M. Jauncey, D.N. Payne, Low-noise erbium-doped fibre amplifier operating at 1.54 mm, Electron. Lett. 23 (19) (1987) 1026. [3] R. Monnard, H.K. Lee, Suppressing Amplifier Transients in Lightwave Systems, Proceedings of the LEOS Summer Topical Meeting, Paper WE3 (2002). [4] Y. Sun, J.L. Zyskind, A.K. Srivastava, J.W. Sulhoff, C. Wolf, R.W. Tkatch, Fast Power Transients in WDM Optical Networks with Cascaded EDFAs, Electon. Lett. 33 (4) (1997) 313. [5] Telcordia GRE-253-Core, SONET Transport Systems: Common Generic Criteria, no. 4, December 2005. [6] P. Bonnenfant, Protection and Restoration in Optical Networks, Tutorial FF1, Optical Fiber Communications Conference, San Diego, CA, February 21e26, 1999. Optical
21
22
CHAPTER 1 Optical amplifiers for next generation WDM networks
[7]
[8] [9] [10]
Layer Protection and Restoration, Short Course SC113/153, Optical Fiber Communications Conference, Anaheim CA, March 17e22, 2001. R. Monnard, A.K. Srivastava, Optical Amplifier Transient Suppression Requirements for Dynamic DWDM Networks. Invited Paper Presentation: Optical Amplifiers and their Applications, Otaru, Japan, (2003) July 6e9, 2003. Ja´nos Farkas, Alberto Paradisi, Csaba Antal, Low-cost survivable Ethernet architecture over fiber, J. Opt. Netw. 5 (5) (2006) 398e409. Private communication from industry sources. International Electrotechnical Commission Documents, Optical Amplifiers - Test Methods - Part 4-1: Transient parameters – Two-wavelength method (61290-4-1) and Part 4-2: Broadband source method (61290-4-2) (to be published) Courtsey: IEC and David Menashe
CHAPTER
ROADM-Based Networks
2
Brandon C. Collings* and Peter Roorda** *
2 Applegate Drive, Robbinsville, NJ,
**
61 Bill Leathem Drive, Ottawa, Ontario
CHAPTER OUTLINE HEAD 2.1. Introduction ...................................................................................................... 23 2.2. Evolution of the ROADM component and network ................................................. 27 2.2.1. Wavelength blocker ......................................................................... 27 2.2.2. Planar light wave circuitry (PLC)-ROADM .......................................... 29 2.2.3. Wavelength selective switch............................................................. 30 2.3. Impact on optical amplifiers requirements........................................................... 35 2.4. Increased density and functional integration of ROADM technology ...................... 37 2.5. Emerging applications and uses of ROADM networks ........................................... 39 2.5.1. Colorless add/drop architectures....................................................... 41 2.5.2. Directionless add/drop architectures ................................................. 41 2.6. Summary........................................................................................................... 43 Acronyms ................................................................................................................. 44 References ............................................................................................................... 45
2.1 INTRODUCTION Early optical networks capitalized on the high bandwidth capacity and transmission reach offered by the combination of an optical frequency carrier and very low loss optical telecommunications fiber. As the need for bandwidth capacity continued to increase, the combination of multiple optical signals at distinct wavelengths onto a single transmission fiber, each carrying independent information streams, provided an economically attractive method to further increase the total information carried by a single fiber. Known as wavelength division multiplexing (WDM) and later dense wavelength division multiplexing (DWDM), this technique has driven the total bandwidth capacity of a single fiber from a relatively meager 155 Mb/s well into the multiple Tb/s capacities of today’s commercial systems, an increase of over four orders of magnitude. Given that DWDM technology transports independent information on separate wavelength channels, the use of wavelength-specific filtering to physically separate and route traffic within a network has proven to be a very cost effective approach to Optically Amplified WDM Networks. DOI: 10.1016/B978-0-12-374965-9.10002-0 Copyright Ó 2011 Elsevier Inc. All rights reserved.
23
24
CHAPTER 2 ROADM-based networks
managing the overall traffic topology within an optical network. It is preferable to using exclusively electrical switching fabrics for two general reasons: The first is that the costs of the optical filtering elements to separate and insert a single DWDM wavelength channel are typically significantly lower than the corresponding costs of terminating all the DWDM channels into the electrical domain, electrically grooming that traffic, and then regenerating all traffic bypassing the node. The second reason is that, unlike typical electronic receiving, switching, and regenerating equipment, optical filtering is agnostic to the protocol and line rate of the traffic carried by a particular wavelength. This results in simpler network operation and upgrade as new and different traffic protocols and line rates can be introduced without replacing any intermediate elements of the network infrastructure. Prior to the introduction of reconfigurable optical network technology, networks generally consisted of point-to-point optical transmission systems interconnecting electrical switching fabrics or ring networks with optical channel filters permanently deployed at each node of the ring to extract and insert wavelengths from and onto the ring. Figure 2.1 illustrates a transmission network with fixed wavelength multiplexing and demultiplexing terminal nodes with electronic based regeneration and bandwidth management situated at the center of each node. In early DWDM deployments in the mid-to-late 1990s, point-to-point topologies were the norm. Optical reach limitations typically resulted in the need to terminate all wavelengths for regeneration purposes. In shorter reach networks where reach was not a constraint, electrical access to the payload to allow sub-wavelength granularity bandwidth management also necessitated the termination of wavelengths into electrical cross-connects or synchronous optical networks (SONET) or synchronous digital hierarchy (SDH) add-drop multiplexers (ADMs). As the optical reach of systems was extended and the number of wavelengths supported per fiber increased, it became both possible and desirable for some wavelength channels to optically bypass some nodes to both simplify the node and eliminate opto-electronic termination equipment previously required to allow traffic to bypass the node. To enable this, fixed wavelength optical add-drop multiplexers (OADMs), shown in Figure 2.2, were introduced in rings or linear chains. These
FIGURE 2.1 Diagram of a typical point-to-point DWDM transmission system employing fixed wavelength multiplexers and demultiplexers, electronic regeneration, and sub-wavelength bandwidth management
2.1 Introduction
FIGURE 2.2 Diagram of a typical point-to-point DWDM transmission system employing fixed wavelength OADM multiplexers and demultiplexers to add and drop optical wavelengths
OADMs were constructed using fixed filters that essentially determined at the time of the network commissioning which wavelengths would locally drop and which would pass through at each node. In addition to being fixed, wavelength banding could be used to reduce loss and filter-based spectral degradation. The introduction of fixed OADMs provided the opportunity to save network cost by eliminating unnecessary optical to electrical to optical conversion, but carried with it a number of key limitations that ultimately limited their application. Network operators were required to carefully plan the network topology at the time of deployment based on how they expected the network traffic to evolve. If these predictions were not accurate and barriers to new bandwidth service deployments surfaced, inefficient workarounds or new system deployments became necessary, both costly consequences considering the initial network may have unused, but inaccessible, bandwidth elsewhere. This effectively reintroduced much of the cost savings that had been achieved by optical bypass. Furthermore, when channels were added into fixed DWDM networks, significant care was needed to control the power level and the manner by which the channel was introduced to optimize performance and ensure existing channels were not impacted. While this was quite manageable in point-to-point configurations, the complexity of the power engineering quickly increased as fixed OADMs were introduced, essentially limiting fixed OADMs to tightly constrained applications (specifically linear OADM chains or hubbed rings). The combination of forecast-intolerant initial filter deployments with complex and manual wavelength deployment procedures conspired to slow network capacity deployment. Prior to the age of consumer driven internet applications and the concurrent explosion of user bandwidth consumption [1], reasonably accurate network planning was possible. Today however, network operators are simultaneously faced with escalating bandwidth requirements, less predictable and more dynamic traffic topologies, and an increasingly competitive marketplace. Therefore, many have turned to reconfigurable optical add/drop multiplexer (ROADM) technology to provide an optical network infrastructure over which they can flexibly deploy
25
26
CHAPTER 2 ROADM-based networks
wavelengths generally between any pair of nodes with relative independence from how other active wavelengths in the network are already provisioned. Furthermore, this flexibility is available to the operator without requiring any interruption of existing traffic, a critical requirement of network operation. This flexibility allows the network operators to grow and adapt the traffic topologies of their networks as traffic demands emerge in a cost effective and operationally efficient manner. It also extends the operational life of a deployed network infrastructure as capacity increases in regions with unanticipated higher demands can be accommodated. In comparison, in fixed-wavelength topology systems, the number of wavelengths serving a given region is determined during the initial network deployment, and areas with higher than anticipated growth can become capacity bottlenecks, resulting in the need for inefficient solutions. Also, as ROADM networks provide an infrastructure that is effectively optically transparent to the line rate and protocol of the traffic passing through it, newly provisioned wavelengths can use higher bit rates, seamlessly support traffic of differing protocols, and take advantage of recent transmission technology advancements such as advanced modulation formats [2] and impairment mitigation strategies [3,4]. Historically, optical networks have often been constructed using ring topologies with electronic switch fabrics interconnecting traffic between optical ring networks as needed at junction points. Another attractive feature of modern ROADM networks is the ability to replace these electronic interconnection facilities with the capability of directly routing optical wavelength channels between optical rings without translating the signals through the electrical domain. More specifically, ROADM networks can optically interconnect wavelengths between different node degrees (fiber trunk pairs entering a network node), essentially creating an alloptical wavelength cross-connect capable of routing wavelengths between any pair of degrees (provided no two like wavelength signals are attempted to be routed to the same degree). Given that equivalent electrical switch fabrics and associated optical transponders are more expensive to deploy, power, spare, and maintain, avoiding the translation through the electronic domain frequently yields significant network cost savings. This multidegree optical interconnection capability enables more general mesh network topologies and a flatter network structure that is generally easier to manage and expand. However, capitalizing on the ability to optically route wavelength channels within the mesh network requires that the optical transmitted wavelengths both have sufficient reach and do not suffer significant impairments from propagating through the ROADM nodes [5]. Finally, ROADM networks generally offer an increased level of embedded monitoring of the optical transmission layer as well as the automation of optical power control and wavelength routing. Many ROADM networks incorporate optical channel power monitors which, along with the ROADM components’ ability to independently control each channel’s optical power, enable the ROADM network to continuously monitor and automatically optimize every channel’s power level. This results in a network that is simpler and more reliable to deploy as fewer manual adjustments are required; more robust over time as power levels are
2.2 Evolution of the ROADM component and network
continuously adjusted toward an optimum operating point; and easier to monitor and detect problems before they become critical. These features translate into decreased operational expenditures and the reduction of new bandwidth service deployment intervals from multiple months down to, potentially, days or less, allowing for more rapid customer capture and less time to revenue from new services.
2.2 EVOLUTION OF THE ROADM COMPONENT AND NETWORK 2.2.1 Wavelength blocker Optical reconfigurability with wavelength granularity was initially introduced into two-degree nodes (nodes with two bi-directional pairs of transmission fibers) using a typical node architecture as shown in Figure 2.3. A broadband optical power coupler on the inbound transmission fiber provides a copy of all inbound wavelengths to a demultiplexing filter structure, allowing any independent combination of wavelengths to be received at this node. A similar structure, used in the reverse direction and connected to a coupler on the outbound transmission fiber, enables the injection of any combination of wavelengths. The critical optical component in this design is the wavelength blocker (WB), a two-port device that can independently attenuate (block) each and any combination of wavelength channels. The WB is placed between the drop coupler of one degree of the node and the add coupler of the other degree with the responsibility of blocking any wavelength signal terminating at this node. By blocking a wavelength channel that has been dropped, a new signal at the same wavelength can be inserted in the outbound direction of the opposite side of the node, allowing wavelength channels to be reused on either side of the node. This overall node architecture allows the network operator to independently select which channels entering/leaving a node are to be dropped/added by directing the WB to block those wavelength channels and which are to be routed through the node by directing the WB to be transparent to those channels.
FIGURE 2.3 Diagram of 2-D ROADM node using a wavelength blocker
27
28
CHAPTER 2 ROADM-based networks
In general, WB components are composed of three functional stages as shown in Figure 2.4. In the first stage, the wavelength channels are spatially separated by a dispersive element such as a diffraction grating. Each of the wavelength signals then encounters a spatial array of optically attenuating elements with one or more independently controllable element(s) per wavelength channel. Finally, all the spatially dispersed wavelength signals are recombined and launched into a single output fiber. Thus, channels can be independently attenuated or blocked through adjustment of the respective attenuation elements. The ability to independently attenuate unblocked channels enables equalization of the optical power levels of channels expressed through a node. Channel power equalization provides greater reach of the optical signals as the power levels of weaker channels (caused by uneven optical amplifier gain, spectrally dependent fiber attenuation, and Ramaninduced inter-channel energy exchange) are renormalized. As the equalization is typically automated, continuous and driven by closed-loop feedback from direct channel power level measurements, this also provides greater system stability and reliability. The WB solves the two key problems associated with fixed OADMs. First, decisions on which wavelengths are add/dropped can be flexibly determined as the network grows, avoiding stranding of unused wavelengths inherent in banded and fixed-wavelength OADMs. Secondly, the means to automate the power control function on a per-wavelength basis allows more controlled introduction of new wavelengths into the system and vastly reduces the complexity of provisioning new wavelengths. The high-capacity long-haul and regional portions of the core network were the initial segments in which systems incorporating WBs were deployed. Given the significant bandwidth these networks carried and the growth they experienced, the ability to flexibly add new wavelength channels reduced the operational expense of supporting that growth and extended the systems’ ability to accommodate that growth. Also, as these systems typically propagated over long distances, the ability to accurately and dynamically maintain the channels’ optical power levels at the desired levels increased the performance and reliability of the transmission which results in lower operational costs.
FIGURE 2.4 Functional diagram of wavelength blocker component
2.2 Evolution of the ROADM component and network
2.2.2 Planar light wave circuitry (PLC)-ROADM An early alternative approach to ROADM implementation was to use planar light wave circuitry (PLC) technology. PLC has been extensively used for arrayed waveguide gratings (AWG) for multiplexing and demultiplexing elements. As active optical functions, such as space switches and attenuators, can also be constructed in solid state PLC technology using Mach-Zehnder interferometers (MZI), the natural evolution of ROADM component technology was to integrate per-wavelength optical routing and power control capabilities with the channel multiplexer and demultiplexer. A functional diagram of a typical PLC-ROADM component and node is shown in Figure 2.5. Similar to the WB architecture, an optical copy of the channels entering a node is directed to a channel demultiplexing structure so that any combination of channels may be locally received. The other copy is directed to the opposite degree of the node and separated according to wavelength using an AWG demultiplexer. Each channel is then directed to a dedicated 2x1 optical space switch with the other input to the switch constituting the respective add port for that channel. The position of each 2x1 switch determines if the locally added channel or the channel from the opposite side of the node will be directed outbound from the node. The output of each switch is passed through an independent optical attenuator followed by all channels being recombined into a single output by an AWG channel multiplexer. The MZIs typically allow relatively fast (w1ms) switching enabling optical protection switching applications. The channel demultiplexer for channels propagating in one direction is integrated with the demultiplexer-switch/attenuator array-multiplexer structure for the channel propagating in the opposite direction (see Figure 2.5) to form a single component that provides the combined functionality for one fiber degree as depicted in the figure. An identical component is used for the second fiber degree in order to maintain equipment separability between opposite degrees of the node, and therefore diverse risk groups for protection purposes. Integration and co-packaging of these elements enable reduced overall costs relative to WB architectures. In addition, optical power monitors can be readily integrated
FIGURE 2.5 Diagram of 2-D ROADM node incorporating PLC-ROADM component
29
30
CHAPTER 2 ROADM-based networks
into the PLC component allowing very high-speed and accurate channel power level monitoring at add, express, and drop locations with low incremental costs. However, similar to the WB, node architectures built from PLC-ROADMs are practically limited to two-degree implementations. Another critical characteristic of PLC-ROADM technology (not necessarily present in WB components) is the limited ability to design AWG multiplexers with wide and predominantly flat channel passbands. For metro networks where a given optical channel may need to propagate through 16 or more nodes, and thus PLC-ROADMs, the repeated filtering by the AWG multiplexers and demultiplexers progressively reduces the bandwidth available to the channel. While the cascadability performance is acceptable for 100 GHz channel spaced applications with 10 Gb/s wavelengths where less than 30 GHz of net bandwidth is required (primarily due to the unlocked wavelength drift of the transmitting laser), PLC-ROADMs with a 50 GHz channel spacing cannot support that necessary amount of bandwidth. As the industry looked toward higher channel line rates, the need to provide as much channel bandwidth as possible, through more than 16 nodes and with 50GHz channel spacing, became a severe limitation for PLC-ROADM devices and consideration of these devices for new network designs declined.
2.2.3 Wavelength selective switch The third generation of ROADM technology and the one that is dominant in current systems is the wavelength selectable switch (WSS). Figure 2.6 shows the basic functional diagram of a 1xN port WSS which is capable of independently routing each channel injected into the common port to any one of the N output ports. Typically, WSS components are optically bidirectional such that the channels from the N ports are selectively multiplexed according to origin port onto the single common (output) port. In addition, the WSS can impart a provisioned amount of attenuation independently to each wavelength channel or may block any channel. Note that when N¼1, the resulting WSS has the same functionality as a WB. The common node architecture construction using WSS elements is shown in Figure 2.7. Similar to the WB node architecture (see Figure 2.3), a copy of the
FIGURE 2.6 Functional diagram of WSS component (Section 3.d.i)
2.2 Evolution of the ROADM component and network
FIGURE 2.7 Diagram of a WSS-based 4-degree ROADM node using a multiplexing WSS and colored add/drop
incoming channels is demultiplexed allowing any combination of channels to be locally received. The WSS is positioned on the outbound side of each degree. The channels coming into each degree are further split into multiple fibers with a copy of each set of channels directed to one of the N ports of each WSS of every other degree (except the degree in which the channels entered the node). Channels that are to be locally added are multiplexed together and injected into one of the N ports of the WSS. The common port of the WSS is connected to the output transmission fiber of the respective degree. Thus, across the N ports of every WSS, a copy of every channel entering the entire node is present. For each degree and for each wavelength, the WSS is then provisioned to route wavelength signals from one of the N ports to its respective common port. Therefore, with this architecture, any wavelength entering a node can be routed to the output of any one or more other degrees, limited by the condition that only a single channel for each wavelength may leave via each degree. Channel power level equalization is implemented by each WSS, independently imparting the proper amount of attenuation to each channel routed through the WSS. As can be seen in Figure 2.7, the value of N for the WSS determines the maximum number of degrees the node configuration can support. For core metro and long-haul networks, values of N as high as 8 are currently of interest. However, for networks residing closer to the end consumer with less need for multidegree optical mesh connectivity, there is strong interest in WSSs with N¼2 simply for the purpose of reducing the cost of the WSS component yet maintaining the full channel granularity flexibility of the higher-degree configurations. Figure 2.8 shows an alternative node architecture where a WSS is installed into each degree with the common port of each WSS connected to the inbound transmission fiber of respective degree. Some of the N ports are then connected to an
31
32
CHAPTER 2 ROADM-based networks
FIGURE 2.8 Diagram of a WSS-based 4-degree ROADM node using a demultiplexing WSS and colorless add/drop
input port of N:1 power splitters/combiners (PS) in each of the other degrees. For channels that are to be locally received, some of the N WSS ports are designated as drop ports and the WSS is instructed to route only one wavelength channel to each of these ports. Each WSS is then configured to direct each channel to the desired degree or drop port, or to block it completely. Channels to be locally added are combined using an M:1 optical power combiner and the output connected to one of the N:1 power combiner inputs. Note that the provisioning of the WSSs and each add multiplexing stage must be coordinated such that only one signal per wavelength is directed to each of the N:1 couplers to avoid a collision of same-wavelength signals. Channel power level equalization is also executed by introducing the proper attenuation to each channel routed through each WSS. In this architecture, each of the drop ports is not permanently assigned a specific wavelength, but rather the wavelength to be dropped from each of these ports is determined by which wavelength is routed to that port by the WSS. Similarly, the add ports have no permanent wavelength assignment. Add/drop ports with this characteristic are know as “colorless” ports and are attractive as they increase the level of flexibility as the operating wavelength of a given colorless port can be assigned and changed by the network operating system at any time. Also, as typically only a subset of the full complement of channels entering a node are needed to be locally added/dropped, colorless ports allow for fewer add/drop ports to be physically presented (relative to AWG demultiplexers which typically possess ports for all channels) thereby saving valuable faceplate space and cost. Figure 2.9 shows the general functional construction of a WSS component which is typically similar to that of a WB. The wavelength signals inserted into the common port of the WSS are spatially separated by a wavelength dispersive device and directed to an array of actuators capable of deflecting each wavelength beam in a controllable angle orthogonal to the axis of wavelength dispersion. The deflected
2.2 Evolution of the ROADM component and network
FIGURE 2.9 Diagram of a general example of the internal design of a WSS
signals are then recombined in the wavelength dispersion direction resulting in N beams of channel groups with each beam comprised of wavelengths that received the same amount of angular deflection. Each beam is coupled into one of the N output fibers. Thus, each channel can be independently routed by adjusting the deflection angle of each actuator for each respective wavelength to the deflection angle corresponding to the desired output port. To operate the device in reverse, the same process is followed with the light propagating in the reverse direction. Per channel attenuation and blocking is generally obtained by adjusting the deflection angle of each channel actuator to detune the coupling efficiency with the output fiber. WSS components can use a variety of deflection angle actuator array technologies [6] including micro electro-mechanical systems (MEMS) mirror arrays [7], arrays of liquid crystal on silicon (LCoS) phase modulators [8], liquid crystal (LC) polarization based switches [9], digital light processing (DLP) mirror arrays [10], and combinations of multiple technologies [11]. For MEMS actuators, an array of tiltable mirrors is fabricated, each with a highly reflective coating. By tilting each mirror (typically one mirror per channel), the beam for each channel can be independently steered among the output fibers allowing the desired output fiber to be selected and coupling efficiency to that fiber to be controlled. LCoS actuators are composed of a large two-dimensional array of optical phase modulators. Angular deflection is achieved by establishing a linearly varying phase retardation profile in one direction of the two-dimensional array. This manipulation of the phase front of the optical beam causes the beam to be deflected by an angle proportional to the rate of change of the phase retardation across the array. The other axis of the array is aligned with the wavelength dispersion direction.
33
34
CHAPTER 2 ROADM-based networks
Actuators using LC technology use the polarization rotation capability of a layer of LC material to variably adjust the orientation of the polarization state of the passing light, followed by a birefringent wedge, to spatially separate the incoming beam into two separate beams with a splitting ratio dependent on the alignment of the polarization state of the light with the axis of the wedge. Therefore, by controlling the polarization state with the LC layer, the beam can be deflected into one of two directions, or split into both directions with a variable ratio. By cascading x sets of these beam steering arrangements with decreasing (or increasing) deflection magnitudes, the ability to deflect the initial beam into 2x discrete output ports is achieved. Similar to the other technologies, an array of such deflection actuators is aligned to the wavelength dispersion direction. Attenuation and blocking can be achieved by adding a stage to partially or fully deflect the beam in an unused direction. The DLP mirror array actuator is composed of a two-dimensional array of binary tiltable mirrors. One axis of the array is aligned in the dispersion direction. In the orthogonal direction, all mirrors are actuated to the same binary state deflecting the entire beam. As the dimension of the individual mirrors is smaller than the practical optical beam size; the beam covers a large number of individual mirrors. Given the binary nature of the DLP mirror tilt, N values greater than 2 may require multiple bounces off of different sections of the DLP array. Similarly, attenuation can be achieved by tilting a fraction of the mirrors covered by the beam; however, as the discarded power is reflected in the other binary direction, attenuation functionality requires an additional bounce off an independent section of DLP mirrors. The most critical performance parameters for the WSS are generally the spectral bandwidth and shape of each wavelength channel and the suppression of wavelength channels from being emitted from unintended ports. To support higher line rate wavelengths, a wide and square channel bandwidth is necessary to avoid bandwidth narrowing from the cumulative filtering resulting from multiple passes through WSSs within the network. Secondly, sufficient suppression of wavelength channels from passing to unintended ports is critically necessary as insufficient suppression can result in an accumulation of optical crosstalk and ultimately signal performance degradation [12]. When used to build a WSS, each switch engine technology has both strengths and weaknesses. MEMS mirror arrays provide channel bandwidths in excess of 40GHz (for a 50GHz channel spacing) and port isolation performance typically better than 45dB, and allow for short optical path lengths that translate into compact physical component volumes maintained as the number of ports is increased. However, as the channel wavelength filtering is defined by the MEMS mirror geometry, variable wavelength channel passbands is not practical and wavelength channels cannot be broadcast to multiple output ports. Also, as channel attenuation is implemented through slight misalignment of the coupling to the output fiber, the open-loop attenuation accuracy depends on the stability of the MEMS mirror. LCoS and LC based actuators can support both variable wavelength channel passbands, accurate open-loop attenuation, and channel broadcast to multiple output ports to some extent; however, port isolation is typically around or
2.3 Impact on optical amplifiers requirements
below 40dB and wide channel passband widths can require a more challenging optical system design. Also, longer optical path lengths are generally needed which results in larger component volumes as the port count increases or the channel spacing decreases.
2.3 IMPACT ON OPTICAL AMPLIFIERS REQUIREMENTS Reconfigurable optical networks require optical amplification to compensate the attenuation of the transmission fiber in generally the same manner as needed by earlier fixed wavelength networks. However, due to the significant differences in both the network and node architectures along with the increasingly dynamic utilization of the networks’ flexibility, the requirements placed on the optical amplifiers differ significantly. To implement the flexibility and photonic cross-connect capabilities that define a reconfigurable optical network, the typical ROADM node architecture requires two optical amplifiers (as shown in the figures in the preceding section). In general, one amplifier on the inbound fiber to the node is positioned and designed to compensate the loss of the preceding fiber span, whereas the second amplifier is typically positioned adjacent to the outbound fiber to compensate the splitting and component losses within the node and prepare the channel optical powers for launch into the transmission fiber. Given this reference node configuration and the loss characteristics of modern ROADM components, the gain required by the outbound amplifier is generally around 19 dB (9dB for 8-degree splitting loss, 6 dB for the WSS component and 3 dB for channel power equalization). As the loss within the node is reasonably constant, this outbound optical amplifier (OA) is designed to have a constant amount of gain or a reasonably narrow variable gain range. In contrast, the inbound amplifier often requires larger amounts of gain, often 25 dB or more, to compensate for fiber spans of 80 km or more. Furthermore, as the fiber span loss varies from span to span (and it is operationally desirable to design as few amplifier variants as practical within a given network system product offering), the inbound amplifier is typically designed to have variable gain to allow a single amplifier variant to support the corresponding range of span losses. Also, as most networks incorporate chromatic dispersion compensation on a per span basis, the losses of these compensation elements are typically offset by inserting the dispersion compensation element between two gain stages of the inbound amplifier. Because the dispersion compensation elements have significant insertion loss, adding them between the input amplifier and output amplifier together would significantly degrade the performance of the ROADM node for two reasons: first, because the loss between the amplifiers would become significantly larger and comparable to or even greater than the span loss, and second, because typical dispersion compensation elements are composed of dispersion compensating fibers in which significant nonlinear penalties are generated at low powers, the power launched from the input amplifier would need to be reduced, which would degrade the ROADM node noise performance.
35
36
CHAPTER 2 ROADM-based networks
For many network applications where fiber spans are longer, the gain required by the inbound amplifier is often larger than that required of the outbound amplifier, and as the nominal output power levels of both inbound and outbound amplifiers is often similar (for economic considerations), the degradation of a given channel’s optical signal-to-noise ratio (OSNR) is dominated by the noise performance of the inbound amplifiers. As ROADMs enable all-optical mesh networking, the need is growing for optical links to propagate through larger numbers of nodes and fiber spans before requiring expensive conversion into the electrical domain, thereby applying pressure to reduce the OSNR degradation a link experiences as much as possible. Historically, one key limitation of optically amplified transmission systems has been the flatness of the amplification gain spectrum. As the gain spectrum of erbiumdoped fiber optical amplifiers is naturally far from constant versus wavelength and changes depending upon the erbium ion inversion operating point of the amplifier, many techniques have been employed to mitigate this characteristic and thereby provide as consistently spectrally constant gain as possible. However, despite many amplifier improvements, in non-ROADM systems, non-flat gain remains a considerable source of system reach limitations. Several spectral equalization approaches have been developed to compensate the unequal spectra gain and thereby extend the link reach by periodically resetting those spectral areas or channels that have received too little or too much gain back to their desired nominal power levels. As current ROADM components have the ability to manipulate each channel independently, they generally have the ability to also independently impart a controllable amount of attenuation to each channel, thereby providing channel power level equalization at each ROADM node. This naturally incorporated channel equalization allows for frequent channel power level rebalancing and minimizes the excursion magnitude each channel may experience before being rebalanced and consequently minimizes the potential subsequent OSNR degradation and nonlinear penalties. In order to accomplish this channel power equalization, some means by which to measure the individual channel powers is necessary in order to assign the proper amount of equalization. This ability to measure and control individual channel power levels at each network node generally allows ROADM networks to incorporate highly automated channel power level management. This provides the network operator a number of advantages including increased and accurate visibility into the network’s health and current operating points, continuous and accurate channel power level optimization, and the elimination of manual power level adjustments during installation (which requires time and expertise, is not typically high resolution, is not dynamic, and is prone to errors). In addition, this automated power control functionality can also be harnessed to provide the power level controlled introduction and removal of individual channels and nodes within an operating network. By implementing these controls, events that trigger rapid changes to the optical power levels in the network resulting in detrimental and transient gain deviations within the amplifiers can be limited. Specifically, the power levels of newly provisioned channels can be slowly increased such that the gain control algorithms of the amplifiers (and the channel
2.4 Increased density and functional integration of ROADM technology
power equalizers) can adiabatically adapt resulting in very limited power excursions of pre-existing channels. Similarly, when adding new segments to an operating network, the gain of the newly commissioned line amplifiers can be slowly increased so that no rapid power fluctuations are experienced. Conversely, during the decommissioning of a channel or network segment, the automated power control management system can slowly reduce the corresponding power appropriately until its level is insufficient to perturb the remaining portions of the network when the transponder or amplifier is finally disabled or physically disconnected. Therefore, within a properly operating system, power transients are quite significantly mitigated for a large number of the events within the normal scope of network operations. However, these control systems cannot mitigate all power level transientcausing events that may occur, namely fiber or hardware equipment failure and improper fiber cable disconnection. But, in each of these cases, the consequence is a rapid decrease in optical power within the network. As it is reasonable to architect and design ROADM network systems and their optical power management systems such that events resulting in a rapid increase in optical power should be rare under proper operating conditions, power level transient mitigation within the optical amplifiers may limit their focus on controlling gain transients resulting from a rapid decrease in input optical power. The inclusion of per-channel power level measurement also provides the opportunity to employ more sophisticated optical amplifier gain control algorithms that use this knowledge of the current present wavelength channel population as well as their relative power levels to provide more accurate and spectrally constant gain. However, there are some considerable practical limitations to this capability as the components employed to measure the channel power levels (optical channel monitors) typically make periodic measurements that are slow relative to the desired transient control time scales. Therefore, this information can be used to fine-tune the amplifier gain control during periods where the power levels are effectively stable; however, as real-time information of the current situation is not available during the short time scale while a transient event is transpiring, the gain control improvements are somewhat limited.
2.4 INCREASED DENSITY AND FUNCTIONAL INTEGRATION OF ROADM TECHNOLOGY Earlier sections in this chapter have highlighted the evolution of ROADM networks as enabled by the expanding performance and capabilities or ROADM technology. An important additional trend is the move toward higher ROADM component densities and the functional integration of ROADM sub-elements. A ROADM node is built up of a number of functional modules (often referred to as “line cards” or “circuit packs”): amplifier modules act as pre-amplifiers for signals entering the node and booster amplifiers for signals launching onto the fiber, ROADM modules provide the key optical switching and power attenuation
37
38
CHAPTER 2 ROADM-based networks
functions, multiplexer and demultiplexer modules provide channel separation for add/drop, and optical channel monitor modules provide per channel power monitoring at select points in the node. These modules are deployed onto a telecom shelf that provides communications among the modules and from a shelf controller that manages the coordinated operation of the elements. Typically, the key modules required for a two-degree ROADM node can occupy a large portion of an equipment chassis. The space consumed is largely dictated by the physical size of the underlying optical components in the node. Amplifier and ROADM components use up a significant portion of the space on the motherboard of shelf modules. Specifically, for ROADM modules, the height of the components often results in shelf modules that consume multiple “slots” in the shelfdtypically two or three slots per ROADM module in a 20-slot shelf. For this reason, recent component development activities have focused on miniaturization of ROADM components with particular emphasis on new low-profile wavelength selective switches with component height to enable single-slot implementations. As enabling optical components shrink, combining multiple architecturally adjacent components into a single module is required to further improve density. While collapsing the entire ROADM node into a single module is problematic due to the resulting “single point of failure” for terminating traffic, the goal of combining the amplifiers, ROADM components, and optical channel monitor (OCM) for one degree of a ROADM node into a single module is attracting considerable attention within the industry. This goal requires not only miniaturization of the various components, but also requires that the packaging for each subelement be optimized. Figure 2.10 a) indicates the key optical functional blocks of one direction of a WSS-based ROADM node. Today this configuration is often separated into multiple modules which may include a two-slot WSS module, a pre-amplifier module, a booster amplifier module, an OCM module, and a shelf processor to coordinate control and management of the node. Figure 2.10 b) is a photograph of an example module with all these functions integrated. Achieving this density requires not only miniaturization of the optical sub-elements, but also novel packaging approaches and advanced mechanical and thermal design techniques. This type of higher level integration will allow not only density improvements but the cost reductions associated with elimination of the electronics and mechanical costs of housing each element in individual modules. Once the traditional demarcation points between optical functions have been blurred through integration, there is potential to optimize the combined erbium-doped fiber amplifier (EDFA)-WSSOCM control to improve responsiveness and accuracy versus a node where real-time control is limited to individual modules. Also, this integrated approach will reduce the operational complexities associated with multiple different cards including inventory management, node commissioning, and fault isolation. The integration conceived of in Figure 2.10 is likely only an initial step toward more fundamental integration that may be achievable with such approaches as monolithic planar waveguide circuits.
2.5 Emerging applications and uses of ROADM networks
(a)
(b)
FIGURE 2.10 a) Functional diagram and b) photograph of a highly integrated ROADM circuit pack incorporating two optical amplifiers, a WSS, a channel monitor, and associated passive optical components and control electronics
2.5 EMERGING APPLICATIONS AND USES OF ROADM NETWORKS For most operators, the initial ROADM network deployments were motivated by wavelength topology flexibility and the multidegree cross-connect (ring interconnect) capabilities offered by ROADM network systems. These features use the dynamic switching capabilities of the ROADM systems during the wavelength provisioning
39
40
CHAPTER 2 ROADM-based networks
process. However, in general, once a wavelength is deployed within the network, it remains unchanged relatively indefinitely. Network operators are now looking at leveraging the dynamic flexibility inherent within ROADM networks to capture more operational advantages by using that network flexibility to better and more efficiently respond to transient events within the network such as fiber breaks and equipment failures. In an operation physically similar to the configuration of a new wavelength, a wavelength channel that has encountered a failed segment in the network can be re-provisioned along a new physical route avoiding the failed segment and its connectivity restored. In many cases, this “re-provisioning,” or photonic restoration, can be accomplished exclusively through the network management system without the requirement of any physical intervention into the network and can typically be accomplished in an interval of seconds to minutes. Typically, this photonic restoration would operate as a secondary mechanism to a primary service protection switch (such as SONET/ SDH or L2/L3). Therefore, the photonic restoration capabilities enabled by the ROADM network’s flexibility allows the network operator to increase the overall service availability by quickly and efficiently restoring connections with minimal field operations. Furthermore, with an automated control plane (such as generalized multi-protocol label switching [GMPLS]), this restoration capability can be integrated into the network operating system such that restoration activities are automated and occur without any operator intervention. Beyond restoration, the ability to re-route in-service wavelengths enables operators to balance traffic loads within their networks and relieve emerging congestion points by redirecting wavelengths away from the congestion and through areas of lower traffic. It also lets operators proactively shift critical traffic away from areas with planned potentially invasive maintenance activities. If photonic layer restoration is used simply to increase network availability by restoring a secondary path of a protected service, the desire to achieve wavelength reroute intervals in a minimal amount of time may not be a high priority assuming the interval is naturally on the order of seconds to minutes. Therefore, the speed in which wavelengths need to be turned up within the network (as in the case of a rerouted wavelength) may be sufficiently slow with respect to the transient suppression capabilities of the modern optical amplifiers. However, if interest develops in photonic layer restoration as a primary mechanism for service recovery (potentially for lower cost, best effort service classes), then one would expect the rate in which a wavelength would need to be turned up within a network to be minimized and perhaps begin to approach speeds where the capabilities of modern optical amplifier transient suppression capabilities for rapid optical power increase and decrease become critically necessary. Common ROADM architectures deployed today as illustrated in Figures 2.3, 2.5, and 2.7 impose some wavelength routing limitations that effectively prevent the full flexibility required for wavelength recovery and dynamic reconfigurability. Specifically, the wavelength and the ingress/egress direction of the add/drop channels are permanently assigned. This is a direct result of the fact that the transmit/
2.5 Emerging applications and uses of ROADM networks
receive signal is connected to a wavelength-specific port on a multiplexer/demultiplexer that is connected to WSS and couplers assigned to a specific direction.
2.5.1 Colorless add/drop architectures Currently, ROADM networks are transitioning to support “colorless” add/drop ports which, unlike colored add/drop ports, do not have a permanently assigned wavelength channel but rather are provisioned as to which wavelength channel will be added/dropped. Architectures to achieve colorless add/drop switching were discussed in an earlier section and illustrated in Figure 2.8. Colorless ports are attractive as generally fewer total add/drop ports are needed, resulting in simpler operations and potentially more compact physical interfaces. Also, coupled with wavelength tunable transmitters, colorless ports allow the wavelength to be selected and provisioned remotely, further simplifying deployment and enabling the remote and rapid modification of a channel’s wavelength needed to support ROADM-based wavelength connection recovery and rerouting. Colorless ports are generally created by replacing the fixed wavelength demultiplexing and multiplexing elements with either a WSS (only a single channel provisioned per port) as shown in Figure 2.8, or a power splitter and tunable filter array. In some cases, a power combiner is used as the multiplexing element. To support the growing interest in systems with a greater number of colorless ports, 1xN WSSs where N is greater than 8 are required. For the WSS, this generally means either a greater range of angular deflection is needed from the switch engine in order to address more fiber ports, a greater density of ports per deflection angle unit (together with some optical system modifications to pack coverage of more ports within the same deflection angle range), or a combination of both, are needed. With a MEMS switch engine, either can be achieved without requiring a change of the spot size in the diffraction direction on the mirror itself, thereby preserving isolation and bandwidth performance as well as physical size. LCoS based engines have some limitations on the maximum deflection angle from the pixelated 0-to-2 dimensional phase modulator array but can achieve greater deflection accuracy by illuminating a larger number of pixels in the deflection axis of the array. However, this may require modifications to the optical system in order to accommodate the larger coverage. LC stack-based engines require additional stacks for additional deflection stages, requiring a longer beam focus through the stacks and consequently a wider spot size, which likely translates in a larger optical system length and generally WSS device.
2.5.2 Directionless add/drop architectures In conventional ROADM nodes, each degree contains a separate group of channel add/drop ports which are multiplexed/demultiplexed within that degree and permanently leave/enter the node through their respective degree (see Figure 2.7). This permanent association of an add/drop port with a particular transmission
41
42
CHAPTER 2 ROADM-based networks
direction limits the dynamic flexibility of the network as the route of a signal from an add/drop port cannot be remotely altered until that signal reaches an adjacent ROADM mesh node. Thus, a second currently emerging trend is the development of architectures that support “directionless” add/drop ports in which the degree, and thereby transmission fiber pair in which the signals enter/leave the node, is fully flexible and is provisioned through the operation system. Therefore, with directionless ports, the network operator is able to select, provision, and alter a signal pair’s route at truly any point within the network giving greater and more thorough flexibility [13]. An example of a node with directionless add/drop ports is shown in Figure 2.11. A copy of the incoming channels from each degree is injected into the ports of a 1xN WSS which selects which wavelengths from which degrees are routed to its common port. This port is connected to the common port of a separate 1xM WSS which demultiplexes these wavelengths into M respective colorless and directionless ports. Given the single fiber connection between the two WSSs, only one instance of each wavelength channel may be dropped per bank of ports. For multiplexing, the same demultiplexing assembly can be used in the reverse direction or a power coupling structure may be used. To relieve the potential application limitations presented by the wavelength blocking characteristic within each the add/drop banks of the architecture illustrated in Figure 2.11, architectures based upon a new type of WSS are being considered. An example is shown in Figure 2.12. This architecture incorporates a new type of
FIGURE 2.11 Four degree ROADM node with “directionless” add/drop switching
2.6 Summary
FIGURE 2.12 Directionless ROADM node incorporating non-blocking NxM WSS
WSS that has N multi-wavelength ports on one side and M ports on the other and has the functionality that wavelengths can be independently routed between any port from the group of N ports and any port from the group of M ports, provided no two signals of the same wavelength are routed to the same port. However, signals within the same wavelength channel, but entering through different ports from the group of N ports, can independently be routed to distinct ports from the group of M ports. Therefore, the add/drop bank created using this wavelength non-blocking NxM WSS can support multiple add/drop ports being provisioned to the same wavelength channel but associated with different node degrees, removing the wavelength blocking limitation and further increasing the flexibility and capability of the network. The ideal architectures and supporting technologies for colorless and directionless ROADM nodes remain a subject of ongoing development and research, but a consensus is emerging that these general approaches will yield the required improvements in network flexibility and responsiveness.
2.6 SUMMARY The inclusion of optical reconfigurability into optical networks has had a profound impact on the application, construction, and operation of optical transmission systems and has placed some additional requirements as well as removed some traditional burdens on the optical amplification. All-optical mesh networking enabled by the WSS has resulted in the need for optical wavelengths to propagate further through the network, yet overcoming greater total loss given the additional
43
44
CHAPTER 2 ROADM-based networks
loss introduced by the ROADM node elements, placing greater burden on the quality of the optical amplification. However, through automated and more accurate channel power level control, some stringent gain and transient flatness requirements can be relaxed. As the capabilities, functionalities, and complexities of optically reconfigurable networks continue to evolve, the need for optical amplification and the requirements placed on that amplification are likely to increase. The introduction of colorless and directionless add/drop ports is expected to facilitate a more dynamic use of optical networks. This evolution will likely result in greater emphasis on network stability and improved transient behavior, in turn requiring higher performance optical amplification capabilities.
ACRONYMS ADM AWG DLP DWDM EDFA GMPLS LC LCoS MEMS MZI OA OADM OCM OSNR PLC PS ROADM SDH SONET WB WDM WSS
Add/drop multiplexer Arrayed wavelength grating Digital light processing Dense wavelength division multiplexing Erbium-doped fiber amplifier Generalized multi-protocol label switching Liquid crystal Liquid crystal on silicon Micro electro-mechanical systems Mach-Zehnder interferometer Optical amplifier Optical add/drop multiplexer Optical channel monitor Optical signal-to-noise ratio Planar light wave circuitry Power splitter (combiner) Reconfigurable optical add/drop multiplexer Synchronous digital hierarchy Synchronous optical network Wavelength blocker Wavelength division multiplexing Wavelength selective switch
References
References [1] Cisco White Paper: Global IP Traffic Forecast and Methodology, 2006-2011 [2] P.J. Winzer, R.J. Essiambre, Advanced Optical Modulation Formats, Proc. IEEE 94 (5) (2006) 952e985. [3] M.G. Taylor, Coherent Detection Method Using DSP for Demodulation of Signal and Subsequent Equalization of Propagation, IEEE Photonics Technology Letters 16 (2004) 674e676. [4] J. McNicol, M. O’Sullivan, K. Roberts, A. Comeau, D. McGhan, L. Strawczynski, Electronic Domain Compensation of Optical Dispersion. Optical Fiber Communication Conference, Optical Society of America, (2005), paper OThJ13. [5] F. Heismann, P. Mamyshev, 43-Gb/s NRZ-PDPSK WDM Transmission with 50-GHz Channel Spacing in Systems with Cascaded Wavelength-Selective Switches, Optical Fiber Communication Conference and Exposition and The National Fiber Optic Engineers Conference (2009), paper OThC1. [6] P. Wall, P. Colbourne, C. Reimer, S. McLaughlin, WSS Switching Engine Technologies, Optical Fiber Communication Conference and Exposition and The National Fiber Optic Engineers Conference (2008), paper OWC1. [7] D. Marom, D.T. Neilson, D.S. Greywall, C.-S. Pai, N.R. Basavanhally, V.A. Aksyuk, D.O. Lo´pez, F. Pardo, M.E. Simon, Y. Low, P. Kolodner, C.A. Bolle, WavelengthSelective 1K Switches Using Free-Space Optics and MEMS Micromirrors: Theory, Design, and Implementation, J. of Lightwave Tech. 23 (2005) 1620e1630. [8] B. Fracasso, J. L. de Bougrenet de la Tocnaye, M. Razzak, C. Uche, Design and Performance of a Versatile Holographic Liquid-Crystal Wavelength-Selective Optical Switch, J. of Lightwave Tech. 21 (2003) 2405e2411. [9] J. Kelly, Application of Liquid Crystal Technology to Telecommunication Devices, Optical Fiber Communication Conference and Exposition and The National Fiber Optic Engineers Conference (2007), paper NThE1. [10] M.T. Knapczyk, L. Grave de Peralta, A.A. Bernussi, H. Temkin, Reconfigurable Add– Drop Optical Filter Based on Arrays of Digital Micromirrors, IEEE J. of Lightwave Tech. 26 (2008) 237e242. [11] J. Homa, K. Bala, ROADM Architectures and Their Enabling WSS Technology, IEEE Communications Magazine 46 (7) (2008) 150e154. [12] B. Collings, F. Heismann, C. Reimer, Dependence of the Transmission Impairment on the WSS Port Isolation Spectral Profile in 50GHz ROADM Networks wit 43Gb/s NRZ-ADPSK Signals, Optical FIber Communication Conference, (2009), paper OThJ3. [13] P. Roorda, B. Collings, Evolution to Colorless and Directionless ROADM Architectures, Optical Fiber Communication Conference and Exposition and The National Fiber Optic Engineers Conference (2008), paper NWE2.
45
This page intentionally left blank
CHAPTER
Challenges and Opportunities in Future High-Capacity Optical Transmission Systems
3 Xiang Liu
Bell Laboratories, Alcatel-Lucent, Holmdel, New Jersey
CHAPTER OUTLINE HEAD 3.1. Introduction .................................................................................................. 3.2. Recent developments in high-capacity transmission systems ........................... 3.2.1. High spectral-efficiency modulation formats and detection schemes. 3.2.2. Orthogonal frequency-division multiplexing .................................... 3.2.3. Recent high-capacity transmission demonstrations ......................... 3.3. Technical challenges in future high-capacity transmission .............................. 3.3.1. OSNR requirements ..................................................................... 3.3.2. Nonlinear fiber impairments.......................................................... 3.3.3. Linear fiber impairments (including optical amplifier impairments) .. 3.3.4. Implementation challenges ........................................................... 3.4. Estimating a “Shannon limit” for fiber optical systems..................................... 3.4.1. Background ................................................................................. 3.4.2. A capacity estimation framework ................................................... 3.4.3. Capacity estimation results and implications .................................. 3.5. Emerging technologies for increasing system capacity and reach .................... 3.5.1. Novel optical transmission fibers ................................................... 3.5.1.1. Ultralow-loss fiber ................................................................. 3.5.1.2. Large core area fiber ............................................................. 3.5.1.3. Shorter span length............................................................... 3.5.2. Novel optical amplification schemes.............................................. 3.5.2.1. Amplification to cover the full fiber transmission band............ 3.5.2.2. High-order distributed Raman amplification ........................... 3.5.3. Photonic integrated circuits .......................................................... 3.5.4. Novel coding techniques .............................................................. 3.5.5. Compensation of fiber nonlinearity ................................................ 3.5.6. Spatial multiplexing ..................................................................... 3.6. Conclusion.................................................................................................... Acknowledgments ................................................................................................. Acronyms ............................................................................................................. References ........................................................................................................... Optically Amplified WDM Networks. DOI: 10.1016/B978-0-12-374965-9.10003-2 Copyright Ó 2011 Elsevier Inc. All rights reserved.
48 49 49 52 54 57 57 59 60 60 61 61 62 65 68 68 68 68 69 70 70 71 71 71 72 73 73 74 74 75
47
48
CHAPTER 3 Challenges and opportunities
3.1 INTRODUCTION Optical transmission systems have been evolving dramatically over the past decade in order to meet the ever-increasing demand in telecommunication capacity. Dense wavelength-division-multiplexing (DWDM) is now widely used to support high capacity per fiber link, and further expansion in capacity is being achieved by increasing the data rate per wavelength channel. In most optical transmission systems, channel data rate has completed the migration from 2.5 Gb/s to 10 Gb/s, and 40 Gb/s is under active development. 100 Gb/s channel data rate is considered the next-generation standard for Ethernet (see, for example, IEEE P802.3ba 40 Gb/s and 100 Gb/s Ethernet Task Force, http://www.ieee802.org/3/ba/). The estimated communication traffic growth is 2 dB per year, or an increase of a factor of 100 in traffic in 10 years [1]. In addition to the demand for high capacity, optical transmission systems are desired to support long-haul optical transmission to effectively address metropolitan, regional, and national network applications. Moreover, the cost per bit, in terms of both capital and operational expenditure, needs to decrease with the increase of system capacity to support sustainable traffic growth. To satisfy these requirements for future high-capacity transmission systems, there are several technical challenges that we have to meet. First, high-spectralefficiency optical modulation formats have to be generated, and at high speed. Second, high-speed optical signals with these advanced modulation formats need to be received with optimized receiver sensitivity. Third, linear and nonlinear fiber transmission impairments, often critical to high-speed transmission, need to be mitigated. Finally, as we approach the Shannon information capacity limit [2] for nonlinear optical fiber transmission [3], the last resort may be an improved fiber link that offers reduced amplified spontaneous emission (ASE) noise, for a given fiber nonlinear penalty. Accompanying the above technical challenges are opportunities for innovating optical technologies that enable future high-capacity transmission systems to become a reality. In this chapter, we discuss promising technologies that may help address the challenges of realizing future high-capacity optical transmission systems. Particular emphasis is put on the system benefits of advanced optical amplification schemes. This chapter is organized as follows: In Section 2, we review the recent developments in high-capacity transmission systems. High spectral-efficiency modulation formats, as well as differential detection and digital coherent detection, are discussed. Multi-carrier modulation based on orthogonal frequencydivision multiplexing (OFDM) is briefly introduced, and compared with singlecarrier modulation. Some state-of-the-art high-capacity transmission demonstrations are highlighted. Section 3 describes the technical challenges in high-capacity transmission, in terms of the requirement on optical signal-to-noise ratio (OSNR) and the tolerance to linear and nonlinear fiber transmission impairments. Section 4 discusses the fundamental Shannon limit applicable for fiber-optic systems. Section 5 introduces a few emerging technologies for increasing system capacity and reach.
3.2 Recent developments in high-capacity transmission systems
Novel optical fiber types and amplification schemes, as well as photonic integrated circuits, are discussed. Emerging digital technologies such as coding and digital compensation of fiber nonlinearity are also discussed. Finally, Section 6 concludes this chapter.
3.2 RECENT DEVELOPMENTS IN HIGH-CAPACITY TRANSMISSION SYSTEMS 3.2.1 High spectral-efficiency modulation formats and detection schemes The evolution of high-capacity optical transmission systems is empowered by technological innovations in several key areas. A noticeable area of significant advance over the past few years is advanced modulation formats and detection schemes that enable high performance and high spectral-efficiency optical transmission [4e9]. The modulation format of an optical signal defines how the information is encoded at the transmitter, and the detection scheme specifies how the signal is received at the receiver. In forming an optical channel, its modulation format and detection scheme are naturally considered together. The choice of modulation format and detection scheme has strong system-level implications such as the OSNR requirement, the achievable spectral efficiency, and the tolerance to fiber chromatic dispersion (CD), polarization-mode dispersion (PMD), and fiber nonlinearity [10]. It is thus a common practice to select the optimum modulation format and corresponding detection scheme for a given set of system requirements. Most current DWDM optical transmission systems are primarily based on 10 Gb/s on-off-keying (OOK) channels on a 50 GHz channel grid with a spectral efficiency of 0.2 b/s/Hz. Capacity upgrade of these systems calls for 40 Gb/s and 100 Gb/s wavelength channels to be carried in the same system. To achieve this, several technical challenges need to be addressed. First of all, the optical spectral bandwidth of each 40 Gb/s or 100 Gb/s channel has to be similar to that of the 10 Gb/s OOK channel to be carried in the same DWDM system. This implies that more spectrally efficient modulation formats are required for channels with higher data rates. Second, the transmission distance of the 40 Gb/s and 100 Gb/s channels is preferred to be comparable to that of current 10 Gb/s channels. This means that advanced modulation formats and detection schemes are desired so that the signal tolerances to ASE noise, CD, PMD, and fiber nonlinearity are not compromised with the increase of per-channel data rate. Differentially coherent or self-coherent optical transmission based on differential phase-shift keying (DPSK) and direct detection [2e7,9] has recently emerged as an attractive vehicle for supporting high-speed optical transmission. The most basic DPSK format is differential binary phase-shift keying (DBPSK) carrying 1 bit/symbol. With partial-delay demodulation, DBPSK was shown to be capable of supporting 40 Gb/s transmission over a 50 GHz grid, corresponding to a spectral efficiency of 0.8 b/s/Hz [11,12].
49
50
CHAPTER 3 Challenges and opportunities
To further increase spectral efficiency, differential quadrature phase-shift keying (DQPSK) carrying 2 bits/symbol can be used. DWDM transmission of 10 107 Gb/s RZ-DQPSK channels over 2,000 km has recently been demonstrated with a net spectral efficiency of 1 b/s/Hz [13,14]. More phase levels can be used to carry more bits per symbol. For example, differential 8-ary phase-shift keying (D8PSK) [15] was used to carry 3 bits/symbol. The demodulation and retrieval of all the data tributaries of an m-ary DPSK signal become more complex with the increase of m. With the help of analog-to-digital conversion and digital signal processing (DSP), a simple yet universal digital self-coherent optical receiver for m-ary DPSK using only one pair of orthogonal demodulators was proposed [16]. The receiver sensitivity of direct-detection m-ary DPSK can be improved to approach that of coherentdetection through data-aided multi-symbol phase estimation (MSPE) [9,16]. Combining DPSK with pulse-amplitude modulation (PAM) [17] provides another means to carry more bits per symbol. The receiver sensitivity is optimized when the Euclidean distances between adjacent PAM levels are equal; they are also equal to the minimum differential-phase distance in the DPSK modulation [9]. Since DPSK acts on signal phase and PAM acts on signal amplitude, they can be independently received. The data tributaries associated with PAM can be received by a binary or multilevel decision circuitry depending on the number of the PAM levels. With the assistance of analog-to-digital conversion, D8PSKþPAM2 and D8PSKþ PAM4 formats, carrying 4 and 5 bits/ symbol respectively, have recently been demonstrated [17]. Figure 3.1 shows the constellation diagrams of four high spectral-efficiency DPSK-based modulation formats that have been experimentally demonstrated. All these formats are compatible with direct detection. The constellation diagrams are normalized such that the average signal power is unity. Evidently, with the increase of modulated phase and/or amplitude levels, the Euclidean distance between adjacent symbols decreases, resulting in a reduction in the tolerance to the ASE noise. We will discuss this issue in Section 2.2. Digital coherent detection has recently attracted extensive attention due to its capability to receive high-spectral-efficiency signals with high receiver sensitivity
FIGURE 3.1 Constellation diagrams of four high spectral-efficiency DPSK-based modulation formats demonstrated with direct detection, DQPSK, D8PSK, D8PSKþPAM2, and D8PSKþPAM4
3.2 Recent developments in high-capacity transmission systems
FIGURE 3.2 Constellation diagrams of four high spectral-efficiency modulation formats commonly used with digital coherent detection, QPSK, 8-QAM, 16-QAM, and 32-QAM
Sx
S
Rx
Sy
OLO
Ry
R PBS
90-deg Hybrid
PBS Signal
90-deg Hybrid
and to digitally compensate for transmission impairments such as CD and PMD [8,18e23]. With digital coherent detection, the optical field of a signal under detection is reconstructed in the digital domain, making it easy for compensation of linear impairments and data recovery. Figure 3.2 shows the constellation diagrams of popular modulation formats commonly used with digital coherent detection, qudrature phase-shift keying (QPSK) [8, 18e23] or 4-point quadrature amplitude modulation (QAM), 8-QAM [24], 16-QAM [25], and 32-QAM [26], carrying 2, 3, 4, and 5 bits/symbol per polarization, respectively. In digital coherent detection, polarization diversity is usually needed as the signal polarization state after transmission is unknown. Figure 3.3 shows a schematic of a polarization-diversity digital coherent receiver, consisting of an optical local oscillator (OLO), a polarization-diversity 28 optical hybrid, 4 balanced
Sx+Rx Sx-Rx
BD ADC
Sx+jRx Sx-jRx
BD
ADC
Sy+Ry Sy-Ry
BD
ADC
Sy+jRy Sy-jRy
ADC
BD
Ix Qx Iy
DSP
Qy
IIx,y ∝ real(Sx,y) Qx,y ∝ imag(Sx,y)
FIGURE 3.3 Schematic of a polarization-diversity digital coherent receiver. OLO: optical local oscillator; PBS: polarization-beam splitter; BD: balanced detector; ADC: analog-to-digital converter; DSP: digital signal processor
51
52
CHAPTER 3 Challenges and opportunities
detectors (BDs), 4 analog-to-digital converters (ADCs), and a DSP unit. The polarization-diversity optical hybrid mixes the incoming signal S with the reference source R generated by the OLO to obtain four pairs of mixed signals, (SxRx), (SxjRx), (SyRy), and (SyjRy), where j is the imaginary unit. The power waveforms of each pair of the output mixed signals are detected and compared by a BD followed by an ADC. The resultant four digital signals Ix,y and Qx,y are linearly related to the in-phase (I) and the quadrature (Q) components of each of the two orthogonal polarization components of the signal separated by the PBS at the signal port. These digital signals are provided to a DSP unit for further processing to determine the amplitude and phase of the unknown incoming signal S. Polarization-division multiplexing (PDM) is an effective means to double the spectral efficiency of a given modulation format. With the use of polarizationdiversity digital coherent receiver, PDM can be supported without additional complexity in receiver hardware, so digital coherent detection is naturally compatible with PDM. Indeed, most of recent demonstrations with digital coherent detection [23e26] were using PDM. Polarization de-multiplexing was performed in the digital domain by using algorithms such as the constant modulus algorithm (CMA) [9,22]. In addition, CMA-based equalization is capable of compensating for PMD, making digital coherent detection attractive for high-speed optical transmission where large system tolerance to PMD is desired. On the other hand, polarization de-multiplexing for PDM DPSK-based formats is not as convenient. In early DPSK experiments, polarization de-multiplexing was performed manually by using a polarization beam splitter [27e29]. Optical polarization de-multiplexing of a PDM signal is feasible in principle, albeit at the expense of added optical complexity. Electronic polarization de-multiplexing was demonstrated for a 100 Gb/s PDM-DQPSK signal, but the PMD tolerance is quite limited [30].
3.2.2 Orthogonal frequency-division multiplexing Orthogonal frequency-division multiplexing (OFDM) is a widely used modulation/ multiplexing technology in wireless and data communications [31]. OFDM was recently introduced to optical fiber communication [32e35]. Enabled by digital coherent detection, coherent optical OFDM (CO-OFDM) [33,35e40] brings similar benefits as single-carrier-based coherent systems described in the previous section such as high spectral efficiency and high receiver sensitivity, while additionally offering transmitter adaptation capability [41] and efficient channel estimation and compensation [42]. Figure 3.4 shows the schematic of a PDM-OFDM transceiver setup. The original data is first divided into x- and y-polarization branches, each of which was mapped onto many frequency subcarriers with a given modulation format, e.g., QPSK or 16-QAM. These data subcarriers, together with a few pilot subcarriers, are then transferred to the time domain by inverse fast Fourier transform (IFFT). To form a complete OFDM symbol, a cyclic prefix is added in order to accommodate the penalty from inter-symbol interference (ISI) that may be caused by fiber CD and PMD. Training symbols (TSs) with known content are inserted
3.2 Recent developments in high-capacity transmission systems
PDM-OFDM Transmitter Ix
PBS
IQ-Mod J-SPMC
TS Insertion
P/S
Symbol Mapping
ytributary
S/P
IFFT
Data
Prefix Insertion
DAC
Symbol Mapping
S/P
xtributary
To fiber link
Qx
DAC CDAC
Iy IQ-Mod Qy
DAC 56GS/s
Prefix Removal
Synchronization and FE
J-SPMC
S/P
FFT
Rough EDC
Channel Estimation
PA-CPEC PA-CPEC
ytributary
P/S
Data
Channel Compensation
Symbol Mapping Symbol Mapping
P/S
xtributary
I’x
ADC
Q’x
ADC
I’y Q’y
ADC ADC
PolarizationDiversit Optical Hybridy
From fiber link
PDM-OFDM Receiver
OLO
FIGURE 3.4 Schematic of a PDM-OFDM transceiver setup. S/P: serial/parallel conversion; P/S: parallel/ serial conversion; TS: training symbol; DAC: digital-to-analog converter; IQ-Mod: I/Q modulator; PBS: polarization beam splitter. ADC: analog-to-digital converter; FE: frequency estimation; J-SPMC: joint self-phase modulation compensation [40]; EDC: electronic dispersion compensation; PA-CPEC: pilot-assisted common phase error compensation.
periodically in the OFDM symbol sequence to facilitate synchronization and channel estimation. The time-domain digital samples are converted by two digitalto-analog converters (DACs) to two analog electrical signals, which are used to drive two optical I/Q modulators. The modulated optical signals are combined by a PBS to form a PDM-OFDM signal. The receiver hardware is the same as that required for single-carrier digital coherent detection. Symbol synchronization, frequency estimation (FE), and channel estimation are performed with the help of the TSs. The data subcarriers are recovered after fast Fourier transform (FFT), channel compensation, and phase compensation (PE). The original data is finally obtained by symbol mapping and serialization. A key feature of CO-OFDM is its DSP capability at the transmitter, in addition to that at the receiver, thereby allowing an end-to-end software defined transmission link. The transmitter DSP capability makes it easy to realize high-level modulation formats such as 32-QAM [26] and 64-QAM [43], and to adapt modulation format and data rate depending on link condition, as commonly done in wireless communications.
53
54
CHAPTER 3 Challenges and opportunities
A Complete OFDM Channel Δf
Sub-channel 2 Sub-channel 1 ΔfG=mΔf (m=1,2,3…)
Sub-channel N
Sub-channel 3
........ fi
fj
Frequency
FIGURE 3.5 Illustration of the spectrum of an OFDM channel consisting of N sub-channels with the orthogonality among all the subcarriers preserved to allow crosstalk-free sub-channel access.
Another attractive feature of CO-OFDM is that it allows individual access to closely spaced sub-channels without coherent crosstalk among these sub-channels [44]. Figure 3.5 shows an illustrated spectrum of an OFDM channel that consists of N sub-channels. When the guard-band spacing (DfG) between adjacent sub-channels is equal to a multiple of subcarrier spacing (Df), each sub-channel can be individually received without suffering coherent crosstalk from neighboring sub-channels because the orthogonality among all the subcarriers (1, . fi,. fj,.) is preserved. When the guard-band spacing is set to be equal to the subcarrier spacing, no spectral efficiency is lost in grouping the subchannels together. The sub-channel access feature allows one to overcome the bandwidth bottleneck of ADC and DAC circuits to form an ultra-high-speed optical channel, albeit at the expense of increased optical complexity needed for receiving all the sub-channels. In fact, this feature has recently been exploited to demonstrate over 1 Tb/s CO-OFDM transmission with sub-channel access at 107 Gb/s [45] and 121 Gb/s [46]. The benefit of OFDM in reducing the coherent crosstalk between sub-channels has also been demonstrated in multi-carrier OOK [47], commonly referred to as coherent WDM, and 2-carrier DQPSK [48] and PDM-QPSK [49]. In the 2-carrier PDM-QSPK demonstration [49], cyclic prefix (or guard interval) usually used in CO-OFDM was removed to further increase the obtainable spectral efficiency. The ISI due to CD and PMD was compensated at the receiver using CMA-based blind equalization, as commonly done in single-carrier digital coherent detection.
3.2.3 Recent high-capacity transmission demonstrations With the advances in high spectral efficiency modulation formats and detection schemes, the last few years have witnessed many record-breaking high-capacity transmission demonstrations. Table 3.1 summarizes some of the state-of-the-art high-capacity transmission results in order of achieved spectral efficiency (SE). A key performance indicator is the SE-distance product (SEDP), which is directly related to the transmission capacity-distance product for the same optical bandwidth allocation.
SE (b/s/Hz) 1.4 [50]
Format/detection
2 [51] 4 [24]
43G DQPSK and 107G PDMDQPSK /DD 111G 2-Carrier NGI-CO-OFDM /DCD 112G PDM-QPSK/DCD 114G PDM-8QAM/DCD
6.2 [25] 7 [26]
104G PDM-16QAM/DCD 65.1G PDM-OFDM-32QAM /DCD
2 [49]
Reach (km)
Fiber type
Optical amplification
1280
SSMF
EDFA
6248
PSCF
7040 580
LCF Low-loss SSMF SSMF SSMF
630 240
Number of channels
SEDP (km-b/s/Hz)
20
1,792
2nd-order DRA and EDFA DRA and EDFA EDFA
135
12,496
72 320
14,080 2,320
EDFA and DRA EDFA and DRA
10 8
3,906 1,680
DD: direct detection with real-time BER measurement; DCD: digital coherent detection with offline DSP; SSMF: standard single-mode fiber; PSCF: low-loss (0.16 dB/km) low-nonlinearity pure silica core fiber; LCF: large-core fiber with 120 mm2 effective area and 0.184 dB/km loss; EDFA: erbium-doped fiber amplifier; DRA: distributed Raman amplifier
3.2 Recent developments in high-capacity transmission systems
Table 3.1 State-of-the-art high-capacity transmission demonstrations
55
56
CHAPTER 3 Challenges and opportunities
With direct detection (DD), a SE of 1.4 bit/s/Hz was demonstrated by copropagating 42.7 Gb/s DQPSK and 107 Gb/s PDM-DQPSK channels in a same DWDM system with 50 GHz channel spacing [50]. Transmission over a 1,280 km standard single-mode fiber (SSMF) link including four reconfigurable optical add/ drop multiplexer (ROADM) passes was achieved. The optical amplification was solely based on cost-effective erbium-doped fiber amplifiers (EDFAs) in the C-band. The achieved SEDP was 1,792 km$b/s/Hz. With digital coherent detection (DCD), DWDM transmissions with higher SE were demonstrated. At 2 b/s/Hz SE, a transmission capacity-distance product of 84.3 Petabit/skm (13.5Tb/s6,248km) was achieved by employing 111 Gb/s no-guardinterval (NGI) CO-OFDM with two PDM-QPSK-modulated carriers and secondorder distributed Raman amplifiers in combination with EDFAs in the extended L-band [49]. The fiber link consisted of 80 km low-loss (0.16 dB/km) low-nonlinearity pure silica core fiber (PSCF) spans. The resulting SE-distance product was 12,496 km/b/s/Hz. In another experiment, 112 Gb/s single-carrier PDM-QPSK was used, and transmission over transoceanic distance (7,040 km) was demonstrated over a fiber link consisting of large-core fiber (LCF) spans with 120 mm2 effective area (Aeff) and 0.184 dB/km loss, achieving a record SEDP of 14,080 km$b/s/Hz [51]. At 4-b/s/Hz SE, a record capacity of 32 Tb/s was demonstrated by using 320 114 Gb/s PDM-8QAM channels on a 25 GHz channel grid [24]. The transmission fiber link consisted of seven 82.8 km spans of ultra-low-loss fiber with an average loss coefficient of 0.176 dB/km and was amplified only by EDFAs in the C and L bands. The achieved SEDP was 2,320 km$b/s/Hz. At 6.2-b/s/Hz SE, 10-channel 112 Gb/s PDM-16QAM transmission over 630 km of SSMF was demonstrated with a DWDM channel spacing of 16.7 GHz [25]. A SEdistance product as high as 3,906 km$b/s/Hz was achieved. The highest SE demonstrated so far for long-haul DWDM transmission is 7 b/s/Hz, which is based on PDM-OFDM with 32-QAM subcarrier modulation [26]. Eight 65.1 Gb/s PDM-OFDM-32QAM channels were transmitted over 3 80 km SSMF spans with DRAs and EDFAs. The achieved SEDP was 1,680 km$b/s/Hz. Note that in the first demonstration listed in Table 3.1, DD was used and the bit error ratio (BER) after transmission was measured in real time. However, in the rest of the demonstrations, DCD with offline DSP was used, and the BER was not measured in real time. This is mainly due to the unavailability of ADC at sufficiently high sampling speed to support these high data rates. At lower data rates, e.g., 40 Gb/s, real-time DCD was recently demonstrated [23] for PDM-QPSK. With two independent DCD-based receivers, the real-time detection of a 100 Gb/s two-carrier PDM-QPSK signal with 20 GHz carrier spacing was reported [52]. Real-time CO-OFDM receivers have also been recently demonstrated [53,54], but at low per-channel data rates ( Γinhom
Unsaturated
Γhom
Saturating signal Saturated
Γhom
A Single Transition energy
(b) Saturating signal
Γinhom >> Γhom
Unsaturated 2Γhom
Saturated
Γinhom Multiple Transition Energies
FIGURE 4.6 Schematic illustration of the mechanism of saturation of homogeneously and inhomogeneously broadened transitions. (a) Unsaturated (dashed curve) and saturated (solid curve) gain spectra of homogeneously broadened transition, with energy in arbitrary units on the horizontal scale and transition gain or gain coefficient on the vertical axis. All atoms have the same transition energy with a Lorentzian line shape centered at the transition energy and a linewidth of Ghom. The transition saturation is uniform across the lineshape, independent of the energy of the saturating signal. (b) Unsaturated (dashed curve) and saturated (solid curve) gain spectra of inhomogeneously broadened transition, with energy in arbitrary units on the horizontal scale and transition gain or gain coefficient on the vertical axis. Atoms at different site types and orientations have different transition energies with natural linewidth Ghom . The distribution of transition energies is assumed to be Gaussian and much wider than Ghom imparting to the transition an overall Gaussian line shape with width Ginhom >>Ghom. The saturating signal burns a hole in the gain spectrum the width of which is twice the homogeneous linewidth Ghom and which is centered at the saturating signal’s energy.
illustrates homogenous saturation. Saturation is homogeneous when all of the ions have identical gain spectra, are Lorentzian in shape, and have a width corresponding to the natural linewidth of the atomic transition. A saturating signal at any wavelength within the gain spectrum (for illustration a saturating signal at the center of the gain spectrum is shown) will result in the same gain
4.3 Gain spectra and DWDM applications
spectrum if the degree of inversion is the same. By contrast, Figure 4.6b illustrates the mechanism for inhomogeneous saturation. In this case the transition energies for atoms located at different types of sites in the glass matrix, which experience different electric fields from neighboring atoms of the silica matrix, are slightly shifted. The aggregate gain spectrum is the accumulation of these individual atomic gain spectra. A saturating signal interacts preferentially with atoms that have transition energy at the photon energy corresponding to the saturating signal’s frequency and less strongly with atoms the transition energy of which is separated from the saturating signal photon energy by more than the transition linewidth. As a result saturation usually, but not always, occurs preferentially in that portion of the gain spectrum in close proximity to the saturating signal’s wavelength, resulting in a spectral hole. The modeling of SHB in EDFAs has proven challenging because the Er3þ ion 4 I15/2 ground state and 4I13/2 first excited state have high angular momentum resulting in a large number of sublevels of each electronic energy level. These sublevels are separated by small energy shifts due to Stark splitting arising from electric fields the erbium ions experience in the glass matrix. As a result of this Stark splitting, SHB in EDFAs is considerably more complicated, and difficult to model, than the simple two-level system depicted in Figure 4.6. The details of the strengths and energies of transitions between the various sublevels of the two electronic states, which would be needed for an explicit model of the SHB in EDFAs are not well known and are difficult to determine. Desurvire [9] attempted to extract the transition energies and strengths as well as the homogeneous and inhomogeneous linewidths from low temperature spectroscopic measurements but even with these powerful experimental tools, because of the complexity of the level structure and transition characteristics (such as dependence of the erbium ion absorption spectrum on the direction of the local electrical field in relation to the optical polarization), it was not possible to extract a model with useful predictive power. Instead, a semi-empirical approach to modeling SHB in EDFAs based on the Giles Model of Equations 1 and 2 has been quite successful. [33] In this model it is recognized that the usual absorption and emission coefficients, a(l) and g*(l), respectively, represent averages of the actual atomic absorption and emission coefficients over the erbium ions in all the various sites in the glass matrix: aðlÞ ¼ ai ðlÞ Equation 10 g ðlÞ ¼ gi ðlÞ where indices i indicate the values for the ith ion and the angle brackets denote averaging over all the atoms i. The average absorption and emission coefficients can be determined without knowledge of the atomic energy level schemes or the strengths of the transitions among the different sublevels purely from optical
99
100
CHAPTER 4 EDFAs, Raman amplifiers and hybrid Raman/EDFAs
measurements of fibers. The values for the ith atom can be written in terms of a deviation function xi(l): ai ðlÞ ¼ aðlÞ,ð1 þ xi ðlÞÞ gi ðlÞ ¼ gðlÞ,ð1 þ xi ðlÞÞ
Equation 11
where the deviation function is the same for the emission and absorption coefficients and has the properties hxi ðlÞi ¼ 0 xi ðlÞ > > vz ¼ ½ðN~2 þ n Þ,ðaðlj Þ þ gðlj ÞÞ aðlj ÞPðlj Þ > > > > > X > vN~2 > > ssp ¼ N~2 ½ðN~2 þ n j Þ,ðaðlj Þ þ gðlj ÞÞ aðlj ÞPðlj Þ=hnz > > < vt j
> X > vn j j > > ¼ n 1 þ ðaðlk Þ þ gðlk ÞÞPðlk Þ=hnz ssp > > vt > > k > > X > > > ðN~2 ,ðaðlk Þ þ gðlk ÞÞ aðlk ÞÞ Gðlj ; lk ÞPðlk Þ=hnz : k
Equation 16 where N~2 is some z-dependent variable (very similar to the erbium inversion, but not exactly equal to it), nj is some other z-dependent variable (similar to the deviation of the inversion of i-th ion from the inversion averaged over ions at given z-location, but also not exactly equal to it); finally, ssp is the spontaneous lifetime of the 4I13/2 erbium first excited state (which is assumed to be the same for all erbium ions). In Figure 4.12, the trace immediately above the experimental trace shows the results of using this model to include effects of SHB. The magnitude of the offset between initial and final states as well as the time profile of the transient are faithfully reproduced. For the dynamic model, if SHB is to be included in the calculation, the only additional variable that is needed is the same correlation function G(li,lj), which is used for steady state modeling of SHB. Thus this correlation function completely characterizes not only the steady state case but also the dynamics of SHB. The simulated evolution of a spectral hole is shown in Figure 4.13 for an EDFA operated with 16 dB gain when a channel at 1530 nm with an input power of -6 dBm is added to a surviving channel at 1533 nm with an input power of e 19 dBm. Note
109
CHAPTER 4 EDFAs, Raman amplifiers and hybrid Raman/EDFAs
0.5
Add -13 dBm
Surviving -19 dBm
0.0
Hole Depth (dB)
110
-0.5
T=0
-1.0 T=500 μs -1.5
-2.0 1525
1530
1535
Wavelength (nm)
FIGURE 4.13 Dynamic behavior of SHB. Simulated SHB profiles for the output power of a surviving channel with input power 19 dBm at a wavelength of 1533 nm in a gain controlled EDFA with gain setting of 16 dB after a signal with input power 6 dBm at a wavelength of 1530 nm is added at timez0 are shown in 50 ms increments [47].
that at T¼0, before the addition of the strong channel at 1530 nm, the SHB spectrum has a minimum near the wavelength of the 1533 nm channel. After the addition of the 1530 nm channel the minimum moves to a shorter wavelength near 1530 nm. The evolution from the initial to the final SHB spectrum occurs over a longer time than it takes the total power to reach its steady state value. This is because the gain control drives the pump to restore the average inversion level more quickly than the inversion levels of different ions at various types of sites can relax to the new equilibrium. These results are based on EDFAs pumped at 1480 nm where the upper pump electronic state is the same as the upper lasing electronic state; the only difference is that the upper pump state is higher in energy in the Stark split manifold of the 4I13/2 first excited state. In this case the relaxation from the pump state to the upper lasing transition (see Figure 4.1) is essentially instantaneous. For 980 nm pumping, on the other hand, the upper lasing transition is a different electronic state: the 4I11/2 second excited state. The lifetime of the 4I11/2 second excited state for nonradiative decay to the 4I13/2 first excited state, which is on the order of 1 to 10 ms, introduces a delay between the absorption of a photon and the population of the upper lasing transition. This can affect the time profile of the transient event as shown by Bolshtyansky et al. [48] who demonstrated that, especially when the transient stimulus is fast, the time profile of the transient response of a gain controlled EDFA can be affected. In particular, the overshoot and undershoot are best modeled if the effects of both SHB and the finite lifetime of the 4I11/2 second excited state are included, especially when the transient stimulus is fast.
4.5 Conclusions
4.5 CONCLUSIONS Optical amplifiers are key elements in high-capacity, wavelength-routed reconfigurable DWDM networks. Networks are evolving from point-to-point systems terminating all channels at the ends of each link to ROADM-based wavelength routed networks which provide increased flexibility and reduced cost. At the same time channel capacity is increasing with higher bit rates and more complex modulation schemes to meet growing bandwidth demand. As a result of both of these trends, the requirements of transparency are becoming more stringent. Ideally the optical amplifiers, combined with the transmission fiber and DCMs, would provide transparent optical lines that faithfully transport the signals regardless of the path length of the channel through the network and regardless of the channel loading on the link or any temporal changes in channel loading. Any impairments added by the amplifiers, most significantly ASE-related optical noise, must be minimized to support both the longer channel path lengths offered by ROADM-based systems and the high channel capacities needed to meet growing bandwidth demand. EDFAs, which provide high gain and output power over a large optical bandwidth adding relatively little optical noise, do all this remarkably well. However, there are limitations that must be managed through amplifier design, network architecture, and network operation. One of the critical requirements for DWDM ROADM-based networks is that the gain spectrum of each amplifier and the link as a whole be flat independent of channel loading. Today’s DWDM amplifiers are gain flattened and are designed to operate at the inversion level at which the gain is flat or has an intended level of tilt. However, for channels with long paths through the network which can traverse 10 or 20 amplifiers or more, the achievable gain ripple (typically specified as 1 to 1.5 dB) is not adequate. In addition, SHB gives rise to a dependence of the gain spectrum on which channels are actually populated. The recent advances in understanding and modeling of SHB in EDFAs are important in designing amplifiers that will meet the demanding requirements of reconfigurable networks. As is described in Chapter 2, ROADMs generally have the capability to adjust the signal power in each channel at each ROADM node to rebalance the channel powers. This permits compensation both of the accumulated gain ripple and of channel population dependent SHB, making possible longer channel paths in the network. In a reconfigurable wavelength network with diverse channel paths, transient effects arising from network reconfiguration and from network faults, such as fiber cuts and connector pulls, will result in changes in amplifier input power, and thus in the noise performance and nonlinear impairments, of bystander channels through cross saturation in the amplifiers. These cross saturation transients can be quite rapid in chains of networks. Fortunately, advances in designing high speed gain control, as described in Chapter 5, go a long way toward mitigating these effects. However, where large changes in channel loading occur abruptly in links with long chains of amplifiers, power excursions could still be large enough to induce bit errors during the transient. During network reconfigurations (commissioning, decommissioning,
111
112
CHAPTER 4 EDFAs, Raman amplifiers and hybrid Raman/EDFAs
or rerouting channels), the powers of the affected channels can be ramped sufficiently slowly so that the amplifier’s gain control can track the changes in loading and limit the power excursions of surviving channels. Most fault conditions, such as connector pulls and most fiber cuts, will occur sufficiently slowly (>100 ms) so that the amplifier’s fast gain control will be able to limit channel power excursions to acceptable levels. Finally, both the longer paths without regeneration that channels follow in reconfigurable networks and the higher channel bit rates that will be developed to meet growing bandwidth demand, will drive a demand for improved line system noise performance. With noise figures typically of 6 dB or less, EDFAs already deliver good noise performance, and there is little opportunity for improvement without significant changes in network architecture. The introduction of coherent systems, which can provide dispersion compensation through electronic signal processing, could obviate the need for in-line optical compensation of chromatic dispersion. If in-line DCMs were eliminated and in-line amplifiers were not required to compensate for their high loss, there would be a significant benefit in terms of gain dynamic range, but the noise figure reduction would not be large. The use of distributed Raman amplification, probably in conjunction with EDFAs in a hybrid Raman/EDFA configuration, could significantly improve the line system noise performance. Raman amplification is already used by many system suppliers to compensate for the loss of the occasional long span, but basing standard in-line amplifier design on distributed Raman amplification to improve noise performance would increase line system cost. Alternative options for improving line system noise performance, such as reducing the transmission span length between amplifier sites or using new fiber designs with larger effective area to make higher channel launch powers possible without incurring nonlinear impairments, are possible, but they would incur significant costs and entail significant changes in system architecture. Such significant changes, if they do occur, will not be rapid.
LIST OF ACRONYMS ASE DCF DCM DRA DWDM EDF EDFA GFF ISI NZDSF
Amplified spontaneous emission Dispersion compensating fiber Dispersion compensating module Distributed Raman amplification or amplifier Dense wavelength division multiplexing Erbium-doped fiber Erbium-doped fiber amplifier Gain flattening filter Intersymbol interference Non-dispersion shifted fiber
References
OSC OI OSNR PD PS RIN ROADM SDP SHB SNR SOA VOA WDM
Optical supervisory channel Optical isolator Optical signal-to-noise ratio Photodiode Power splitter Relative intensity noise Reconfigurable optical add/drop multiplexer Site dependent pumping Spectral hole burning Signal-to-noise ratio Semiconductor optical amplifier Variable optical attenuator Wavelength division multiplexer
References [1] E. Desurvire, J.R. Simpson, P.C. Becker, High-gain erbium-doped fibre amplifier, Optics Letters 12 (11) (1987) 880e890. [2] R.J. Mears, L. Reekie, I.M. Jauncey, D.N. Payne, Low-noise erbium-doped fibre amplifierr operating at 1.54 mm, Electron. Letters 23 (19), 1026e1028 [3] N.S. Bergano, Undersea Amplified Lightwave Systems Design, Optical Fiber Telecommunications IIIA, Academic Press, San Diego, CA, 1997, pp. 302e335. [4] J.P. Kunz, C. Fan, Terrestrial Amplified Lightwave System Design, in: I.P. Kaminow, T.L. Koch (Eds.), Optical Fiber Telecommunications IIIA, Academic Press, San Diego, CA, 1997, pp. 265e301. [5] M.D. Feuer, D.C. Kilper, S.L. Woodword, ROADMs and Their System Applications, in: T. Li, A.E. Willner, I.P. Kaminow (Eds.), Optical Fiber Telecommunications VB, Academic Press, Burlington, MA, 2008, pp. 293e344. [6] J.L. Zyskind, Y. Sun, A.K. Srivastava, J.W. Sulhoff, A.L. Lucero, C. Wolf, et al., Fast Power Transients in Optically Amplified Mulitwavelength Networks, Proc. Optical Fiber Communications Conference, Optical Society of America, Postdeadline Paper PD31, (1996). [7] A.K. Srivastava, Y. Sun, Advances in Erbium-Doped Fiber Amplifiers, in: I.P. Kaminow, T. Li (Eds.), Optical Fiber Telecommunications IVA, Academic Press, San Diego, CA, 2002, pp. 174e212. [8] J.L. Zyskind, J.A. Nagel, H.D. Kidorf, Erbium-Doped Fiber Amplifiers for Optical Communications, in: I.P. Kaminow, T.L. Koch (Eds.), Optical Fiber Telecommunications IIIB, Academic Press, San Diego, CA, 1997, pp. 13e68. [9] E. Desurvire, Erbium-Doped Fiber Amplifiers: Principles and Applications, John Wiley & Sons, New York, 1994. [10] C.R. Giles, E. Desurvire, Modeling of Erbium-Doped Fiber Amplifiers, J. Lightwave Technol. 9 (2) (1991) 271e283. [11] Y. Sun, J.L. Zyskind, A.K. Srivastava, Average Inversion Level, Modeling and Physics of Erbium-doped Fiber Amplifiers, IEEE J. Sel. Topics in Quantum Electronics 3 (4) (1997) 991e1007.
113
114
CHAPTER 4 EDFAs, Raman amplifiers and hybrid Raman/EDFAs
[12] P.C. Becker, N.A. Olsson, J.R. Simpson, Erbium-Doped Fiber Amplifiers: Fundamentals and Technology, Academic Press, San Diego, CA, 1999. [13] C.R. Giles, E. Desurvire, J.R. Simpson, Transient gain and cross talk in erbium-doped fiber amplifiers, Optics Letters 14 (16) (1989) 80e882. [14] E. Desurvire, Analysis of transient gain saturation and recovery in erbium-doped fiber amplifiers, IEEE Photonics Letters 1 (8) (1989) 196e199. [15] R.H. Stolen, E.P. Ippen, Raman gain in glass optical waveguides, Appl. Phys. Lett. 22 (1973) 276e281 (R.H. Stolen, private communication). [16] Y. Emori, S. Namiki, 100 nm bandwidth flat Gain Raman Amplifiers Pumped and Gain-Equalized by 12-wavelength-channel WDM High Power Laser Diodes, in: Proc. Optical Fiber Communications Conference, Optical Society of America, Postdeadline Paper 19, (1999). [17] S. Namiki, Y. Emori, Ultrabroad-Band Raman Amplifiers Pumped and Gain-Equalized by Wavelength-Division-Multiplexed High-Power Laser Diodes, IEEE J. Sel. Topics in Quantum Electronics 7 (1) (2001) 3e16. [18] K. Rottwitt, A.J. Stentz, Raman Amplification in Lightwave Communication Systems, in: I.P. Kaminow, T. Li (Eds.), Optical Fiber Telecommunications IVA, Academic Press, San Diego, CA, 2002, pp. 174e212. [19] I. Mandelbaum, M. Bolshtyansky, Raman amplifier model in single-mode optical fiber, IEEE Photonics Letters 15 (12) (2003) 1704e1706. [20] C. Fludger, A. Maroney, N. Jolley, An Analysis of the Improvement in OSNR from Distributed Raman Amplifiers using Modern Transmission Fibres, in: Proc. Optical Fiber Communications Conference, Optical Society of America Paper FF2, 2001, p. 100. [21] H.H. Kee, C.R.S. Fludger, V. Handerek, Statistical Properties of Polarisation Dependent Gain in Fibre Raman Amplifiers, in: Proc. Optical Fiber Communications Conference, Optical Society of America, Paper WB2, (2002), p. 180. [22] C.R.S. Fludger, V. Handerek, R.J. Mears, Pump to Signal RIN Transfer in Raman Fiber Amplifiers, J. Lightwave Technology 19 (8) (2001) 1140e1148. [23] P.B. Hansen, L. Eskildsen, A.J. Stentz, T.A. Strasser, J. Judkins, J.J. DeMarco, et al., Rayleigh Scattering Limitations in Distributed Raman Pre-Amplifiers, IEEE Photonics Technology Letters 10 (1) (1998) 159e161. [24] S. Kinoshita, Y. Sugaya, H. Onaka, M. Takeda, C. Ohshima, T. Chikama, Low-Noise and Wide Dynamic Range Erbium-Doped Fiber Amplifiers with Automatic Level Control for WDM Transmission Systems. In Proc. Opt. Amplifiers and Their Applications, Monterey, CA, 1996, pp. 211e214. [25] Y. Sun, J.B. Judkins, A.K. Srivastava, L. Garrett, J.L. Zyskind, J.W. Sulhoff, et al., Transmission of 32-WDM 10 Gb/s Channels Over 640 km Using Broad-Band, Gain Flattened Erbium-Doped Silica Fibers, IEEE Photonics Technology Letters 9 (12) (1997) 1652e1654. [26] F. Forghieri, R.W. Tkach, A.R. Chraplyvy, Fiber Nonlinearities and Their Impact on Transmission Systems, in: I.P. Kaminow, T.L. Koch (Eds.), Optical Fiber Telecommunications IIIA, Academic Press, San Diego, CA, 1997, pp. 196e264. [27] H. Ono, M. Yamada, Y. Ohishi, Gain-flattened Er-doped fiber amplifier for a WDM signal in 1.57-1.60 mm wavelength region, IEEE Photonics Tech. Lett. 9 (1997) 596e598. [28] Y. Sun, J.W. Sulhoff, A.K. Srivastava, A. Abramov, T.A. Strasser, P.F. Wysocki, et al., A Gain-Flattened Ultra Wide Band EDFA For High Capacity WDM Optical
References
[29]
[30]
[31]
[32] [33] [34]
[35]
[36]
[37] [38]
[39]
[40] [41] [42]
[43]
[44]
[45] [46]
Communications Systems, In Proc. European Conference on Optical Communications, 54 (1998). A.K. Srivastava, J.L. Zyskind, J.M. Sulhoff, J.D. Evankow, Jr, M.A. Mills, Room Temerature Spectral-Hole Burning in Erbium-Doped Fiber Amplifiers. In Proc. Optical Fiber Communications Conference (1996) 33e34. E. Rudkevich, D.M. Baney, J. Stimple, D. Derickson, G. Wang, Nonresonant SpectralHole Burning in Erbium-Doped Fiber Amplifiers, IEEE Photonics Technology Letters 11 (5) (1999) 542e544. A.N. Pilipetskii, D. Kovsh, D.G. Foursa, S.M. Abbot, M. Nissov, Spectral-HoleBurning in Long-Haul WDM Transmission. In Proc. Optical Fiber Communications Conference, Paper FM3 (2004). A. Siegman, Lasers, University Science Books, Mill Valley, CA, 1986 (Chapter 30). M. Bolshtyansky, Spectral Hole Burning in Erbium-Doped Fiber Amplifiers, J. Lightwave Technol. 21 (4) (2003) 1032e1038. M. Bolshyansky, G. Cowle, Spectral Hole Burning in EDFA under Various Channel Load Conditions. In Proc. Optical Fiber Communications Conference, Paper OTUH5 (2009). M. Bolshyansky, G. Cowle, Spectral Hole Burning Compensation in Raman/EDF Hybrid Amplifier. In Proc. Optical Fiber Communications Conference, Paper JWA15 (2008). M.J. Yadlowsky, L.J. Button, Pump-Mediated Inhomogeneous Effects in EDFAs and Their Impact on Gain Spectral Modeling. In Proc. Optical Fiber Communications, Paper TuG5 (1998) 35e36. M. Yadlowsky, Pump Wavelength-Dependent Spectral-Hole Burning in EDFA’s, J. Lightwave Technology 17 (9) (1999) 1643e1648. P.N. Kean, S.J. Wilson, M. Healy, R. Di Muro, N.E. Jollye, F. Davis, Pump Induced Inhomogeneity of Gain Spectra in Conventional and Extended-Band EDFAs. In Proc. Optical Fiber Communications Conference, Paper WA4 (1999) 10e12. M. Bolshtyansky, P. Wysocki, N. Conti, Model of Temperature Dependence for Gain Shape of Erbium-Doped Fiber Amplifier, J. Lightwave Technology 18 (11) (2000) 1533e1540. C.R. Giles, E. Desurvire, J.R. Simpson, Transient Gain and Cross-talk in Erbiumdoped Fiber Amplifiers, Optics Letters 14 (16) (1989) 880e882. E. Desurvire, Analysis of Transient Gain Saturation and Recovery in Erbium-Doped Fiber Amplifiers, IEEE Photonics Technology Letters 1 (1989) 196e199. A.K. Srivastava, Y. Sun, J.L. Zyskind, J.W. Sulhoff, EDFA Transient Response to Channel Loss in WDM Transmission System, IEEE Photonics Letters 9 (3) (1997) 386e388. Y. Sun, J.L. Zyskind, A.K. Srivastava, L. Zhang, Analytical Formula for the Transient Response of Erbium-Doped Fiber Amplifiers, Applied Optics 38 (9) (1999) 1682e1685. J.L. Zyskind, Y. Sun, A.K. Srivastava, J.W. Sulhoff, A.L. Lucero, C. Wolf, et al., Fast Power Transients in Optically Amplified Multiwavelength Optical Networks. In Proc. Optical Fiber Communications Conference, Postdeadline paper PD31 (1996). E. Ishikawa, S. Tanabe, M. Nishihara, Y. Akasaka, (Chapter 5 of this book). C. Tian, S. Kinoshita, Analysis and control of transient dynamics of EDFA pumped by 1480- and 980-nm lasers, J. Lightwave Technology 21 (8) (2003) 1728.
115
116
CHAPTER 4 EDFAs, Raman amplifiers and hybrid Raman/EDFAs
[47] M. Bolshtyansky, N. King, G. Cowle, Dynamic behavior of spectral hole burning in EDFA. In Proc. Optical Amplifiers and Their Applications, Paper OTuB2 (2006). [48] M. Bolshtyansky, N. King, G. Cowle, Dynamic Behavior of Spectral Hole Burning in EDFA with 980 nm Pumping. In Proc. Optical Fiber Communications Conference, Paper OMN2 (1997).
CHAPTER
Dynamic and Static Gain Changes of Optical Amplifiers at ROADM Nodes
5
Etsuko Ishikawa,* Setsuhisa Tanabe,** Masato Nishiharax, Youichi Akasakaz *
Fujitsu Limited,
**
Kyoto University, x Fujitsu Laboratories LTD, z Fujitsu Laboratories of America Inc.
CHAPTER OUTLINE HEAD 5.1. EDFAS in ROADM-BASED systems .................................................................. 5.1.1. System model of ROADM ........................................................... 5.1.2. Dynamic change in optical amplifiers .......................................... 5.1.3. Static changes in optical amplifiers ............................................. 5.2. Optical power surge in Edfa and its suppression............................................ 5.2.1. The reason for power excursion ................................................... 5.2.2. Feedback control ....................................................................... 5.2.3. Feedback and feedforward control ............................................... 5.3. Spectral hole burning phenomena in silica-based edf .................................. 5.3.1. Experiment................................................................................ 5.3.1.1. Characteristics of EDF samples ........................................... 5.3.1.2. GSHB measurement scheme at 77K ................................... 5.3.2. Results ..................................................................................... 5.3.2.1. Wavelength dependence of the second hole character......... 5.3.2.2. Wavelength dependences of the main and second holes...... 5.3.2.3. Er3þ ion concentration and saturation signal input power dependence of the second hole........................................... 5.3.3. Discussion ................................................................................ 5.3.3.1. Characteristics of the gain spectral holes at low temperature and room temperature ........................... 5.3.3.2. Saturating wavelength dependence of the main and second hole depths ...................................................... 5.3.3.3. Saturating wavelength dependence of the main and second hole widths ...................................................... 5.3.3.4. Er3þ ion concentration dependence of the second hole depth 5.3.3.5. Signal wavelength dependence of the second hole depth ..... 5.3.4. Conclusion ................................................................................ 5.4. Principles of gain spectral hole burning in erbium-doped fiber amplifiers ....... 5.4.1. Experimental ............................................................................. 5.4.2. Results and Discussion .............................................................. Optically Amplified WDM Networks. DOI: 10.1016/B978-0-12-374965-9.10005-6 Copyright Ó 2011 Elsevier Inc. All rights reserved.
118 118 119 119 120 120 120 122 125 126 127 128 128 128 130 130 133 133 135 136 136 137 140 140 141 141
117
118
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
5.4.2.1. Multigain hole structure in erbium-doped fiber..................... 5.4.2.2. Relation between the multihole GSHB structure and the Stark energy structure of Er3þ ion ........................... 5.4.2.3. Effect of Er3þ concentration................................................. 5.5. Positive gain change of gain spectral hole burning observed at room temperature 5.6. Numerical model of GSHB in EDFA ................................................................ 5.6.1. Description of the GSHB numerical model ................................... 5.6.2. Calculated results of the GSHB simulation under various conditions 5.7. Summary..................................................................................................... Acronyms ........................................................................................................... References .........................................................................................................
141 143 144 146 147 147 150 152 153 153
5.1 EDFAS IN ROADM-BASED SYSTEMS Reconfigurable optical add-drop multiplexer (ROADM) systems were originally developed to realize remote control of wavelength paths from a central office in order to respond to changes in traffic demands. The simplified path control enabled by the ROADM is also used for other purposes, such as automated path changes in response to fiber failures on the active path with generalized multiprotocol label switching (GMPLS) [1]. In the future, this function may enable optical packets to be routed to their destinations dynamically in the optical domain. The re-routing of wavelength paths will cause the channel count on related fiber paths to increase or decrease dynamically, resulting in performance changes for those devices whose performance is channel-count dependent. However, usually optical amplifiers and their related components are the only devices in the system with such a dependency [2]. Among optical amplifiers, erbium-doped fiber amplifiers (EDFAs) are by far the most widely deployed, and therefore understanding the behavior of these devices is very important. This chapter will review the characteristics of EDFAs which might have an impact on ROADM systems. It describes transient power surges in Section 2 and the spectral hole burning (SHB) phenomenon which is inherent in erbium-doped fiber (EDF) in Sections 3 to 5. The modeling of SHB in ROADM systems will also be discussed in section 6.
5.1.1 System model of ROADM Metro systems usually consist of several to many ROADM nodes, arranged in a ring or mesh structure or some combination of the two. Figure 5.1 shows an example of a ring structure with five ROADM nodes and a total of 32 channels. Thirty-one of the channels are routed from Node A to Node D, while the remaining channel is added at Node B and dropped at Node C. If a fiber failure occurs between Node A and Node B, or channels 1 through 31 are rerouted to Node E due to a change in customer demand, the optical power between Node B and Node C will be reduced to 1/32 of its original value. However, the optical amplifiers between Node B and Node C will attempt to
E
D A
31ch C
Ch hannel countts between B and C
5.1 EdfaS in roadm-Based systems
32ch working
1ch
Fiber cut
B
Time
1ch FIGURE 5.1 System model of ROADM-based ring network.
maintain the same total power as for the case of 32 channels present. Table 5.1 summarizes the phenomena in optical amplifiers after a change in the channel count.
5.1.2 Dynamic change in optical amplifiers Without any pump power control mechanisms, the single remaining channel would be amplified 32 times (up to 15 dB) more than required. To prevent this power surge, optical amplifiers on ROADM systems should be equipped with pump power control features. The optical power surge is the most obvious phenomenon because it is a dynamic change. The rest of the phenomena listed in Table 5.1 are static changes. Table 5.1 Dynamic and static gain changes in ROADM network Italic: Dynamic change
Bold: Static Change
Discrete Amplification (EDFA)
Optical Power Surge Spectral hole burning (SHB) Gain deviation by GEQ
Distributed Amplication (SRS between singnals)
Gain tilt
5.1.3 Static changes in optical amplifiers SHB and gain deviation introduced by the gain equalizer (GEQ) are phenomena that occur in EDFAs, and gain tilt due to stimulated Raman scattering (SRS) and wavelength dependent loss of transmission fiber are generated in transmission fibers. These effects degrade the amplifier gain flatness over wavelength range occupied by the wavelength division multiplexed (WDM) signals. The effect of gain deviation is
119
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
independent of the remaining channel allocation. However, the gain difference due to SHB depends on the channel allocation, and the gain characteristics over wavelength could change locally depending on the remaining signals’ wavelength locations. In this case, fixed gain flattening devices such as thin film optical filters cannot compensate the gain deviations. The accumulation of this unexpected gain deviation due to SHB will have an impact on system performance. Therefore, ROADM systems should have features to address SHB for different channel allocations.
5.2 OPTICAL POWER SURGE IN EDFA AND ITS SUPPRESSION 5.2.1 The reason for power excursion The principle of optical amplification using an EDFA is energy transfer from excited Er3þ ions to signals through stimulated emission. Figure 5.2 shows the relevant energy levels of the erbium ion. A pump laser with a wavelength of 980 or 1480 nanometers excites Er3þ ions, which decay rapidly to the upper lasing levels. Transitions between levels 4I13/2 and 4I15/2 provide gain for the wavelength range from about 1530 to 1570 nanometers through stimulated emission. Assuming the pump laser operates at a constant rate, any change in the number of amplified channels will affect the gain of other channels. For a sudden drop in the channel count, the remaining channels will experience a corresponding increase in their output power and thus in their gain.
5.2.2 Feedback control Several optical and electronic technologies have been developed for maintaining constant signal power (gain)/channel in EDFAs under varying channel loads [3e9]. Optical mitigation such as gain clamping is advantageous because it eliminates situation dependent controls; however, it is costly and complex to implement. On the other hand, electronic mitigation is cheaper and easier to configure once conditions 4F
2
Ene ergy (104 cm-3)
120
7/2
980nm
4I
1
4I
11/2
13/2
980nm 1480nm
1550nm 4I
15/2
FIGURE 5.2 Energy levels in an EDFA and scheme of photon consumption by signal.
5.2 Optical power surge in EDFA and its suppression
EDF filter filt
Input I t
Input monitor
Output monitor
Pump Laser
AGC Feedback Loop
FIGURE 5.3 Configuration of EDFA with only feedback function.
are programmed. The electrical feedback circuit adjusts the EDFA pump power according to the difference between the measured and required power or gain. Automatic control of the signal power is called auto power control (APC) and control of the gain is called auto gain control (AGC). Figure 5.3 shows a schematic of the simplest electrical AGC, which employs only a feedback function. This control is sufficient for maintaining the optimal pumping power in networks without ROADMs. However, it may not be good enough for networks that frequently add and drop channels. Figure 5.4(a) shows an example where the channel count changes from 40 channels to one channel. Figure 5.4(b) shows the transient response of the remaining channel power during the channel dropping processes [10,11]. A residual population inversion persists for several hundred micro-seconds while the feedback circuit adjusts the pump power and the excited-state population adjusts according to the new pump rate. As a result, the remaining signal has more than 8 dB of excess
(a)
39chs were Dropped or Re-routed
(b)
working
1ch (1546.12nm)
Fiber cut
Time
output power (r.u.)
Chan el count
40channel to 1channel dropping due to fiber cut 0.9 0.8 8.91dB 0.7 1ch . 4 h 0.5 0.4 0.3 0.2 . -3.7dB 0 -0.0004 0.0001 0.0006 0.0011
0.0016
0.0021
time (s)
FIGURE 5.4 Power surge with only feedback function (a) Input signal stimulus time profile; (b) Remaining channel output power versus time.
121
122
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
gain just after the 39 channels are dropped. This power surge causes a higher bit error rate and may eventually damage the receiver as it grows even larger in the downstream EDFAs.
5.2.3 Feedback and feedforward control The origin for the power surge is the effective lifetime (w100 ms) of the erbium excited states, which makes it difficult to realize fast feedback loops. To realize a practical and low cost EDFA control solution, both electrical feedforward and feedback have been investigated [10,11]. As shown in Figure 5.5, both feedback and feedforward can be used to achieve fast AGC. The feedforward circuit reacts only to changes in input power. The fact that it does not react to output power changes makes it very fast. The amplitude of the feedforward control signal is determined by the relation between the AGC-required pump power and total input power, which can be measured. Figure 5.6 shows the required power of for a 980 nm pump for various signal channel counts and channel allocations. The relationship between the pump power and the number of channels is roughly linear at the two different levels of individual signal input power considered (12 dBm/ch and 22 dBm/ch) for the same gain. The differing values of AGC-required pump power correspond to different channel allocations as shown in Figure 5.6. In the field, information on the number of channels and the power level of each channel is not always available to the control circuit. However, the total signal input power and the AGC-required pump power are linearly related as shown in Figure 5.7. The lines for average channel powers of 12 dBm and 22 dBm merge into a single line. Therefore, for feedforward control of the EDFA, no information on the average channel power, the EDF Input
Input monitor
Pump Laser
Feedback Loop
Feedforward
FIGURE 5.5 Configuration of EDFA with feedback and feedforward functions.
Output monitor
5.2 Optical power surge in EDFA and its suppression
400 -12dBm/ch
-22dBm/ch
Pump power (r.u.)
350 channel allocation
300 250 200 150 100 50 0 0
10
20
30
40
50
number of channels
FIGURE 5.6 Required pump power (for feedforward control) as a function of channel count with different channel allocations and different powers per channel as indicated.
number of channels, or the exact channel allocation is necessary. The only important parameter is the total input power, which can be measured by the input monitor of the EDFA and the gain setting. Thus there is a useful relationship between total input power and feedforward control signal amplitude. The feedforward control signal is based on a nominal pump power, which is the average pump power required by all the possible channel allocations at the same total input power. Because these different channel allocations require slightly different pump powers, the difference between this pump power and the feedforward signal is provided by the feedback control of the AGC circuit. Under the control of both the feedforward and feedback systems, if the input channel number changes suddenly, the input monitor detects this event and adjusts the pump power to the corresponding nominal value 400 -12dBm/ch
Pump p power (rr.u.)
350
-22dBm/ch
300 250 200 150 100 50 0 0
1
2
3
Input power (r.u.)
FIGURE 5.7 Relation between the AGC-required pump power and the total input power for 980-nm pumped EDFA.
123
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
immediately. The feedback control then makes a fine adjustment to the pump power and compensates the pump power difference caused by different channel allocations. In a practical sense, the response time of the feedforward control can be faster than 1 ms, and the time constant of the feedback loop can be much longer. The control algorithm can be understood as follows: In an AGC-EDFA, the population inversion, and thus the gain in the gain medium, is assumed constant. If the input conditions change suddenly, the pump power is changed accordingly by the feedforward control immediately, keeping the population inversion constant. In a purely feedback-controlled EDFA, the feedback loop monitors the gain error. Therefore it has to wait until the gain changes before adjusting the pump power. Because the speed at which the control circuit can adjust the pump power is slow, and there is always a phase delay between the gain and the pump power, gain excursion occurs. The performance of an EDFA with AGC as a combination of electrical feedforward and feedback is shown in Figure 5.8. The gain excursion is quite small (0.31 dB) for a 980 nm pumped amplifier and would be similar for an EDFA pumped at 1480 nm. Similar performances are achieved for different numbers of dropped and survived channels. The control has little wavelength dependence and has high dynamic range. It also works for slow channel adding/dropping processes. In a practical design of the AGC circuit with combination of electrical feedforward and feedback, the relationship between the feedforward pump control and the input power can be determined during the design of the EDFA optical circuit. However, pump laser aging may change this relationship. The effects of aging can be compensated by monitoring the pump power and adjusting the relation between the pump-control signal and the pump power, which is always monitored in the EDFA control circuit. Because the control is fully electrical, no extra optical components are needed. 40ch-1ch drop in 100ns (fiber-cut) 0.4 Gain & Population inversion
0.35
Output (r.u.)
124
0.3 0.25
40ch
0.2 1ch
0.15
Pump power
0.1 0.05 0 -0.001
0.31dB
0
0.001
0.002
0.003
Time (s)
FIGURE 5.8 Transient response from 40ch to 1ch with both of feedback and feedforward control. The lower solid line labeled 0.31 dB is the output power of the single channel which is not dropped.
5.3 Spectral hole burning phenomena in silica-based EDF
5.3 SPECTRAL HOLE BURNING PHENOMENA IN SILICA-BASED EDF Spectral hole burning is a fundamental material property of the optical interactions of erbium-doped fibers used in EDFAs. We have characterized SHB by measuring and analyzing the gain spectra of EDFs with different Er concentration and conditions. The SHB in general describes the hole formation in the absorption spectrum of a material, which was site-selectively excited by narrow light sources such as a laser. On the other hand, the gain spectrum hole burning (GSHB) describes the energy selective degradation of signal gain in the gain spectrum of an EDF that was irradiated with a strong saturating signal. This degradation can be attributed mainly to the decrease in the population inversion and/or depopulation of excited levels, by the signal that induces the stimulated emission from upper states to lower states. Based on this phenomenological difference, we will call the SHB phenomena in the gain spectra of EDF GSHB. In this section, the GSHB in silica-based Er3þ-doped fiber, EDF are explained phenomenologically, based on the results of measurement for different Er concentration at 77 K in both the C-band and L-band wavelength regions. The deepest hole was observed in the EDF with the lowest Er concentration. When a saturation signal of longer wavelength than 1530 nm was input into an Er3þ-doped fiber, two holes can be observed at around 1530 nm and the wavelength of a the saturating signal. The depth of the second hole at around 1530 nm becomes shallower with increasing Er3þ concentration. The dynamics of gain spectral hole burning is discussed with relation to the effect of energy transfer between Er3þ ions. GSHB in the EDFA has been a serious problem for long-haul optical transmission systems for over a decade [12]. GSHB causes the gain deviation or hole in spectral region when a strong signal is input into EDFA. It has been known that an additional hole (second hole) can be burned at around 1530 nm independent of the wavelength of the saturating signal, where a main hole is burned [13]. So far, the second hole depth of GSHB observed around 1530 nm in an EDF has been reported to be about 1 dB at room temperature when a strong saturation signal was input into EDF [14]. The physical mechanism of the holes in the gain spectra is complicated and the formation process of the second hole has not been clarified. On the other hand, today, the transmission distance of long-haul optical system is much longer than ever, and can include over 150 EDFAs. Based on these facts, the more precise prediction of the gain spectral shape after transmission is becoming much more important. GSHB has been one of the major problems, which makes it difficult to predict the amplified gain spectra in the C-band wavelength region (1530e1565 nm) [15e17]. Accurate prediction of GSHB is required to manage the optical gain in a design. Therefore, many numerical and experimental measurements have been performed to clarify and model the mechanism of GSHB [18e20]. However, no complete physical interpretation of GSHB has been formulated yet. In order to approach the basic physical mechanism of GSHB from different perspective, we
125
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
investigated the Er3þ content dependence of the hole shape and depth of GSHB in silica-based EDF at 77 K. In this study, we performed GSHB measurements on four silica-based EDF at 77 K to observe the second hole depth and the shape more clearly the Er3þ concentration and to study its dependence on. We will discuss the effect of the energy transfer between Er3þ ions on the hole formation mechanism.
5.3.1 Experiment Figure 5.9 illustrates a subtracting technique used to measure gain spectral hole burning. Figure 5.9(a) shows the experimental setup. A saturation signal, gainlocking signal, and probe signal generated from the light sources are coupled by two optical couplers. The saturation signal is the origin of the GSHB and usually has high power. The gain-locking signal is used to fix the population inversion of the EDF being measured by maintaining the gain of the gain-locking signal constant. The probe signal is coupled to measure the gain spectrum by scanning its wavelength. The power of the gain-locking signal and the probe signal must be small enough that they do not induce GSHB. The coupled signals are input to a polarization scrambler to remove the effect of polarization hole burning. The output
(a)
Probe signal
Gainlock Gain -lock signal
EDFA EDF
Optical coupler
Saturation Signal
Optical coupler
980 nm / 1550 nm coupler
Polarization scrambler
Isolator
Isolator
Optical spectrum analyzer
980 nm pump LD
(c)
15
Without saturation signal With saturation signal
10 5 0
Gain deviation (dB)
(b) Gain (dB)
126
Saturation Gain - locking
Probe
-5 1520
1530
1540
1550
Wavelength (nm)
1560
1570
0 -1 -2 -3 -4 1520
Gain change due to SHB
1530
1540
1550
1560
Wavelength (nm)
FIGURE 5.9 Spectral subtraction technique for measurement of SHB. (a) Experimental setup, (b) measured gain spectra, (c) gain change due to SHB.
1570
5.3 Spectral hole burning phenomena in silica-based EDF
signals from the polarization scrambler are input to the EDF and the amplified signals are measured by an optical spectrum analyzer. The gain spectrum of the EDF is obtained by measuring the difference in the input power and the output power of the probe signal whose wavelength is scanned. While the gain spectrum is measured, the pump power of the EDF is controlled to maintain a constant gain for the gain-locking signal in order to fix the population inversion. The gain spectra with the saturation signal (GSHB-induced spectrum) and without the saturation signal (reference spectrum) are measured (Figure 5.9[b]), and the difference between the two gain spectra corresponds to the gain change due to GSHB (Figure 5.9[c])
5.3.1.1 Characteristics of EDF samples Four silica-based EDF samples with different Er2O3 content were used as the gain medium for EDFA. The Er3þ ion concentrations of samples were 130 ppm, 280 ppm, 700 ppm, and 1600 ppm. Optical parameters for each EDF are shown in Table 5.2. The fiber length of each EDF was determined to include same total number of Er3þ ions in EDF, which leads to the equivalent attenuation spectra for all the EDF samples. The absorption spectra of EDF in the wavelength region from 1420 nm to 1640 nm are shown in Figure 5.10. Table 5.2 Parameters for silica-based EDF samples Sample
Er3+ ion (ppm.wt)
Fiber Length (m)
n1,3mm
EDF EDF EDF EDF
130 280 700 1600
30.0 15.7 4.8 2.9
1.469 1.470 1.468 1.470
I II III IV
Loss(dB/div)
EDF EDF EDF EDF
1450
1500
1550
1600
Wavelength(nm)
FIGURE 5.10 Absorption spectra of EDF samples with different Er3þ ion content in the wavelength region from 1420 nm to 1640 nm.
127
128
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
5.3.1.2 GSHB measurement scheme at 77K Figure 5.11 shows the experimental setup of GSHB measurement at 77 K. Each EDF was cooled directly by liquid nitrogen in a dewar. Probe and gain-locking signals were from coupled C-band tunable laser (Santec, TSL210) and L-band tunable laser (Santec, TSL 210) sources combined by two 3dB couplers to provide wavelength range from 1520 nm to 1600 nm. Each EDF was forward pumped by a 979 nm laser diode (LD) (FITEL), and the pump power of LD was set to maintain a constant gain at the gain locking signal when the saturation signal was input into EDF. The gain locking signal wavelength was 1560 nm. Temporarily 1580 nm signal was used as gain locking signal when the saturation signal was at 1560.5 nm. The power of the gain locking signal was 30 dBm. The probe signal wavelength was scanned from 1520 nm to 1620 nm at a power of 30 dBm. To examine the saturating wavelength dependence of the second hole and main hole width, the wavelength of saturating signal was varied from 1530 nm to 1610 nm. To investigate the saturating power dependence of the second hole, GSHB was measured at the saturating power of 5.3 dBm and 0 dBm. The GSHB spectrum was calculated by subtracting gain spectrum without the saturating signal from that with the saturating signal. The gain spectrum without saturation signal had the same gain of 11.1 dB at 1531.8 nm, which corresponds to the second hole wavelength, to keep the same population inversion ratio of each fiber and to compare the degradation of population inversion ratio caused by GSHB.
FIGURE 5.11 GSHB measurement setup at 77 K (TLS: Tunable Laser Source, PS: Polarization Scrambler, ATT: Variable attenuator, OSA: Optical Spectrum Analyzer, CPL: Coupler).
5.3.2 Results 5.3.2.1 Wavelength dependence of the second hole character Figure 5.12 shows GSHB spectra with various saturating signal wavelengths at 77 K for EDF I (Er3þ:130 ppm). The saturation signal power was 0 dBm. There are both a primary hole at each saturating wavelength and a secondary hole at 1530 nm. These are deeper and sharper than those seen at room temperature. At 77 K, the center wavelength of the second hole was independent of the saturating signal wavelength
5.3 Spectral hole burning phenomena in silica-based EDF
4 3
ΔGain n(dB)
2
1540.5nm 1550.5nm 1560.5nm 1570.5nm 1580.5nm 1590.5nm
77K E 130 Er3+:130ppm EDF Length:30m Psat:0dBm
1 0 -1 -2 -3 -4 1520
1530
1540
1550
1560
1570
1580
1590
1600
Wavelength(nm)
FIGURE 5.12 GSHB spectra with various saturating signal wavelengths of at 77 K for EDF I (Er3þ:130 ppm).
while the center wavelength of the main hole varied with the saturating signal wavelength. This characteristic of the second hole has also been reported at room temperature in the C-band wavelength region [15]. As the saturating signal wavelength became longer, the second hole depth was found to be decreased at 77 K. Figure 5.13 shows the relationship between the second hole depth and the saturating signal wavelength for the cases where the saturating signal power is 0 dBm and 5.3
Se econd h hole de epth(dB B)
0
Psat.-5.3dBm Psat.:0dBm
-1
Psat.:-5.3dBm
-2
Psat.:0dBm -3
EDF -4
1540
1560
77K (130ppm) 1580
1600
Wavelength(nm)
FIGURE 5.13 Relationships between the second hole depth and the saturating signal wavelength for an EDF I. (Ps ¼ 5.3dBm, 0dBm).
129
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
dBm. As shown in Figure 5.13, the second hole depth at the saturating signal power of 5.3 dBm monotonically decreased as the wavelength of saturation signal became longer. On the other hand, the wavelength dependence of the second hole depth at the saturating signal power of 0 dBm seems to show the maximum peak around 1580 nm as the saturation signal wavelength became longer. The origin of the peak around 1580 nm of the second hole depth is not clear at present and needs more consideration.
5.3.2.2 Wavelength dependences of the main and second holes Figure 5.14 shows the relationships between the main and second hole widths and the saturating signal wavelength for the 0 dBm saturating signal power case. The values of the main and second hole widths were obtained by Gaussian fitting. As a result, the main hole width at 1530 nm was estimated to be 2.25 nm. The main hole width, DlMain at 1.53 mm has been reported to be about 2.4 nm at 77 K for the silica-based EDF and follow a T1.73-law in the 20e77 K temperature range [12]. This shows good agreement with our result. On the other hand, the second hole width DlSecond at 1.53 mm was estimated to be about 3.79 nm, which is broader than that of the main hole at 1.53 mm. In addition, the main hole width was found to become broader even at 77 K as the saturating signal wavelength became longer whereas the second hole width did not show the saturation signal wavelength dependence and was about 3.85 nm. 5 Main hole Second hole
Hole width h(nm)
130
4
3 77K Psat:0dBm EDF (130ppm) 2
1540
1560
1580
1600
Wavelength(nm)
FIGURE 5.14 Relationships between the main and second hole widths and the saturating signal wavelength for the case where the saturating signal power is 0 dBm.
5.3.2.3 Er3þ ion concentration and saturation signal input power dependence of the second hole Figure 5.15(a) and (b) show GSHB spectra for four EDF samples with different saturating signal power (Psat¼5.3 dBm, 0 dBm). The wavelength of saturating signal is 1550.5 nm. When the input power of saturation signal was 5.3 dBm, the second hole showed almost the same shape and depth regardless of the Er3þ ion
5.3 Spectral hole burning phenomena in silica-based EDF
(a)
1
ΔGain(dB B)
0 Er 3+:1600ppm Er 3+:700ppm Er 3+ :280ppm 3+ Er :130ppm
-1 -2 -3 -4 1520
Psat=-5. 3dBm 1530
1540
1550
1560
1570
Wavelength(nm) g ( )
(b)
1
ΔGain(dB)
0 3+
-1 -2
280ppm
-3 -4 1520
Er :1600ppm :1600ppm Er 3+:700ppm Er 3+:280ppm Er 3+:130ppm
1600ppm 700ppm Psat=0dBm
130ppm 1530
1540
1550
1560
1570
Wavelength(nm)
FIGURE 5.15 GSHB spectra for four EDF samples with different saturating signal power ((a) Psat ¼ 5.3dBm, (b) 0dBm).
concentration (Figure 5.15(a)). On the other hand, in the case of the saturation signal power 0 dBm, the second hole depth showed the Er3þ ion concentration dependence (Figure 5.15[b]). Figure 5.16 shows the Er3þ ion concentration dependence of the second hole depth at 77 K (Psat¼5.3 dBm, 0 dBm). In the case of EDF I (Er3þ:130 ppm), the second hole depth became deeper as the saturating signal input power increased. However, the second hole depth of EDF IV (Er3þ:1600 ppm) shows little power dependence at 77 K. With decrease of the Er3þ ion concentration in EDF samples, the second hole depth for saturating signal power 0 dBm became deeper. Figure 5.17(a) and (b) show the second hole spectra of GSHB with different saturating signal input power for EDF I (Er3þ:130 ppm) and EDF IV (Er3þ:1600 ppm), respectively. Figure 5.18 shows the relationships between saturating signal input power and the second hole depth for EDF I (Er3þ:130 ppm) and EDF IV (Er3þ:1600 ppm). The difference between the second hole depths of two samples did not occur when the saturating signal power was small. On the other hand the Er3þ ion concentration dependence of the second hole depth became more evident when the saturating signal power of 0 dBm was input into EDF.
131
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
-1
S Second d hole depth((dB)
77K sat:1550.5nm
-2
-3
P sat.:-5.3dBm P sat.:0dBm -4 4
0
500
1000
1500
Er3+ ion(ppm.wt)
FIGURE 5.16 Er3þ ion concentration dependence of the second hole depth at 77 K (Psat ¼ 5.3dBm, 0dBm).
(a)
1
3+
EDF Iv (Er :1600ppm)
Δ ΔGain(dB)
0 -1 -2
Psat :-10dBm Psat :-5dBm Psat :0dBm
-3 3 -4 1520
1530
1540
1550
1560
1570
Wavelength(nm)
(b)
1
EDF I (Er3+:130ppm) 0
ΔGaiin(dB)
132
-1 -2
Psat :-10dBm Psat :-5dBm Psat :0dBm
-3 -4 1520
1530
1540
1550
1560
Wavelength(nm)
FIGURE 5.17 GSHB spectra with different saturating signal power ((a) EDF I, (b) EDF IV).
1570
5.3 Spectral hole burning phenomena in silica-based EDF
Second d hole depth(dB)
0
77K :1550.5nm sat
-1
-2
EDF I -3
EDF IV
-4
-5
-10
-5
0
Psat.Input power(dBm)
FIGURE 5.18 Relationship between the second hole depth and the saturating signal power for EDF I (Er3þ:130ppm) and EDF IV(Er3þ: 1600ppm).
5.3.3 Discussion In our measurements, very deep and sharp second and main holes were observed at 77 K compared with those at room temperature. The second hole and main hole depth became deeper as the wavelength of saturating signal became shorter. In addition, the main hole width was found to become broader at 77 K as the saturating signal wavelength became longer, whereas the second hole width did not show the saturation signal wavelength dependence. Moreover, the second hole depth showed the Er3þ concentration dependence and was found to become deeper in the silica-based EDF that had lower Er3þ concentration, although four EDF samples have the same total Er3þ ion number. Therefore, we discuss the physical dynamics of hole formation process, mainly focusing on the anomalous characteristics of Er3þ concentration dependence of the second hole.
5.3.3.1 Characteristics of the gain spectral holes at low temperature and room temperature The origin of deep second and main holes at low temperature most likely lies in the change of the thermal relaxation rate to the Boltzmann distribution over the Stark manifolds of Er3þenergy levels. Since the thermal relaxation to the Boltzmann distribution at low temperature is achieved more slowly, the population density of ions in the excited state 4I13/2 remains almost unchanged as in the initial state until the induced emission from the 4I13/2 level occurs. This suggests that Er3þ ions in the Stark manifolds within the 4I13/2 level can be depopulated more stably at 77 K than that at room temperature. On the other hand, at room temperature, the thermal equilibrium over the Stark manifolds within the 4I13/2 and the 4I15/2 level is achieved
133
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
1
293K
ΔG Gain(dB)
0
-1
77K
-2
Psat:0dBm
-3
λsat sat:1540.5nm -4 1520
1530
1540
1550
Wavelength(nm)
FIGURE 5.19 SHB spectra at 77K and room temperature.
very fast before the induced emission from the 4I13/2 level to the 4I15/2 level occurs. Therefore, the Boltzmann distribution in the excited state is built up in very short time at room temperature, which leads to the shallow second hole at room temperature as mentioned above. Figure 5.19 shows the SHB phenomena at low temperature (77 K) and room temperature (293 K). The GSHB spectra in Figure 5.19 were obtained by the subtracting technique that was described in Figure 5.11. The saturation signal, gainlocking signal, and probe signal were input to the EDF. The large gain change was observed at the wavelength around the saturation signal (main hole) and around 1530 nm (second hole) for both temperatures. As the temperature increased from
Emission Line Broadening Low Temperature Gain
134
Room Temperature
1530nm Wavelength (nm)
FIGURE 5.20 Line broadening of multi-hole structure of Er3þ ions at room temperature.
5.3 Spectral hole burning phenomena in silica-based EDF
77 K to 293K, both of the main and second holes became shallower. The depth of the main hole and the second hole decreased by 2.6 dB and 1.0 dB, respectively. In addition, the main hole and the second hole overlapped each other and gain deviation was observed in a broad wavelength region at 293 K. The temperature dependence of the hole observed in Figure 5.19 is due to the thermal broadening of the emission line width of the Er3þ ions, i.e., the peak intensity and line width became small and broadened with the temperature increase. A schematic figure of the temperature evolution of gain spectral shape is shown in Figure 5.20. While clear structures of each transition are well separated at 77 K, overlapping of these broadened lines at room temperature make the spectral shape unclear. In a commercial system, the EDFAs are operated at near room temperature. Therefore, understanding of the behaviors of the SHB at room temperature will be important to estimate the effect of the SHB in the commercial system. The numerical model and the behavior of the SHB at room temperature are described in detail in Section 6.
5.3.3.2 Saturating wavelength dependence of the main and second hole depths The main and second hole depths decreased as the wavelength of saturation signal became longer. These characteristics of the main and second hole depths can be explained by the energy structure model as shown in Figure 5.21. The solid bars
(a)
4
(b)
I 13/2
Second hole 4
4I
Main hole
I 15/2
Shorter wavelength
13/2
Second hole 4
Main hole
I 15/2
Longer wavelength
FIGURE 5.21 Depopulated energy state region of the Er3þ ion energy structure by the saturating signals. ((a) shorter wavelength, (b) longer wavelength).
135
136
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
show the region over the Stark manifolds that can contribute to the formation of the second hole. The dashed bars show those that can contribute to the formation of the main hole. In addition, the length of solid bars and that of dashed bars correspond to the energy of the second and main hole, respectively. The population of the halftoned Stark manifolds can be depopulated by a saturating signal. Figure 5.21(a) and 5.21(b) show the burned areas by the saturating signal whose wavelength is longer and shorter, respectively. As shown in Figure 5.21, the population of the Stark manifolds within the 4I13/2 state decreases when the wavelength of saturating signal becomes longer . Therefore, one reason for the decrease of the second and main hole depth is considered to be due to the decrease of the population of the Stark manifolds within the 4I13/2 state.
5.3.3.3 Saturating wavelength dependence of the main and second hole widths The main hole width was found to become broader as the saturating signal wavelength became longer, whereas the second hole width did not show the saturation signal wavelength dependence as mentioned above. These characteristics of the main and second hole widths give important information on the relaxation process between the Stark manifolds relating the formation of the main and second holes because the relaxation process is directly correlated with the line width. Judging from the saturating signal wavelength-independent character of the second hole, the relaxation process relating the formation of the second hole is not supposed to change when the saturating signal wavelength was varied. On the other hand, the difference between the main hole widths in the C-band and the Lband region as shown in Figure 5.14 is possibly caused by the difference between the relaxation processes in the C-band and the L-band region. Actually, it has been reported that the temperature dependence of the homogeneous broadening in the C-band region shows T1.5 dependence over the temperature range 30e150 K, which indicates that both direct one-phonon process (which leads to T1 temperature dependence) and Raman process (which leads to T2 temperature dependence) are included in the relaxation process of the transitions in the C-band wavelength region. On the other hand, the temperature dependence of the homogeneous broadening in the L-band region has been reported to follow the temperature law, T (which indicates that direct one-phonon process is dominant). In addition, the homogeneous line width in the L-band region is broader than that in the C-band region at low temperature. (T < 120K) [21]. Taking these into account, the observed main hole width broadening in the L-band wavelength region at 77 K can be from the homogeneous contributions to the transitions in the L-band wavelength region.
5.3.3.4 Er3þ ion concentration dependence of the second hole depth
For an explanation of the Er3þ ion concentration dependence of the second hole depth, we propose the contribution of Er3þEr3þ energy transfer mechanism to thermal relaxation rate as a hole formation mechanism. The rate of relaxation is
5.3 Spectral hole burning phenomena in silica-based EDF
delayed at low temperature because of the decrease of the number of thermally activated phonons that contribute to the relaxation. On the other hand, the direct energy transfer between Er3þ ions such as migration is not largely affected by the change in temperature. In the case of electric dipole interaction, the dependence of energy transfer probability on the Er-Er separation has been well known to be inversely proportional to sixth power of the average distance between Er3þ ions [22]. Therefore, the effect of the direct energy transfer among Er3þ ions is more likely dominant in the EDF samples that have higher Er3þ ion concentration. In the case of the EDF samples with higher Er3þ ion content, the delayed thermal relaxation rate at 77 K is expected to be compensated by the contribution of the direct Er3þ-Er3þ energy transfer mechanism to thermal relaxation rate. The migration process between donor and accepter has been reported to contribute to make the homogeneous distribution of Er3þ ions excited to 4I13/2 state by energy transfer [23]. Therefore, the second hole depth of the EDF sample with higher Er3þ ion content such as EDF IV (Er3þ:1600 ppm) is considered not to change even in high saturation signal power because of the rapid energy transfer among Er3þ ions. On the other hand, in the case of the EDF sample with lower Er3þ ion concentration like EDF I (Er3þ:130 ppm), the second hole depth becomes deeper because of the slow energy transfer mechanism among Er3þ ions.
5.3.3.5 Signal wavelength dependence of the second hole depth Figure 5.22(a), (b), and (c) show the saturating signal wavelength dependences of the second hole depths at 77 K for three EDF samples when the wavelength of saturating signal was varied from 1540.5 nm to 1640.5 nm. The saturating signal power was 0 dBm in all measurements. As shown in Figure 5.22(a), (b), and (c), very deep and sharp second holes at around 1530 nm were observed especially when the wavelength of saturating signal was close to that of the second hole compared with those at room temperature. In addition, the third hole at 1520 nm was also observed in the three EDF samples. The depth of the second hole was about four times deeper than that of the third hole when the main hole was burned at 1540 nm. So far, this third hole has not been reported to exist in the shorter wavelength region than 1530 nm but reported to be the uplift of the gain. However, this uplift should be considered not to be intrinsic character of the gain spectral hole burning but to arise from the difference between the population inversion ratio of the reference gain and that of the gain, which was obtained when the burning signal was introduced. We believe that the third hole burned at 1520 nm reflects a transition between upper Stark manifold in the 4I13/2 and lower one in the 4 I15/2 level. Figure 5.23 shows the relation between the holes in a GSHB spectrum and their corresponding transitions in the Er3þ energy level. Assuming the energy gap between each Stark manifold sublevel is constant (3), the number of combinations of electronic transitions is a maximum for a transition energy of Ec, which is the center of gravity of the 4I13/2-4I15/2 transition. There exist several possible transitions, the energy of which corresponds to Ec 3 and Ecþ3. The latter case corresponds to the third hole, whose wavelength is shorter than that of the second
137
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
(a)
(b) 1640.5nm
(c) 1640.5nm
1630.5nm
1630.5nm
1620.5nm
1620.5nm
1620.5nm
1610.5nm
1610.5nm
1610.5nm
1590.5nm 1580.5nm 1570.5nm 1560.5nm 1550.5nm
1600.5nm
ΔHole depth(dB/div)
1600.5nm
1590.5nm 1580.5nm 1570.5nm 1560.5nm 1550.5nm
1530 1550 1570 Wavelength(nm)
1630.5nm
1600.5nm 1590.5nm 1580.5nm 1570.5nm 1560.5nm 1550.5nm
1540.5nm
1540.5nm
1510
EDF
1640.5nm
EDF
ΔHole depth(dB/div)
EDF
ΔHole depth(dB/div)
1510
1530 1550 Wavelength(nm)
1570
1540.5nm
1510
1530 1550 Wavelength(nm)
1570
FIGURE 5.22 GSHB spectrum of (a) Sample I, (b) Sample II, and (c) Sample III at 77K.
1
ε
4I 13/2
0
45cm- 1 ~ 10.5 nm
Gain(dB)
138
-1 -2
Ec - ε
-3 Ec + ε 4I 15/2
Ec
77K Psat .:0dBm λ sat :1530.5nm
-4 Ec
-5 1510 1520 1530 1540 1550 1560 1570 Ec + ε Ec Ec - ε
Gain( , z) ∝∫[
e(
) N 2( , z)z)
Wavelength(nm) z)]dz (λ) N1 ( , z)]dz
a
FIGURE 5.23 Origin of spectrum broadening.
hole. Because the number of combinations of these transitions should be less than that of the transition of Ec, the hole depth corresponding to these transitions becomes shallower than that of the second hole observed at 1530 nm. So far, the second hole depth of GSHB observed at around 1530 nm in an EDF has been reported to be about 1dB at room temperature even for a strong saturating signal [20]. This is because the thermal relaxation assisted by
5.3 Spectral hole burning phenomena in silica-based EDF
phonons occurs more rapidly at room temperature than at low temperature, and inhibits the formation of the gain spectral hole. The wavelengths of the second and third holes were independent of the saturating signal wavelength (1540.5 nm e 1640.5 nm). The similar characteristics were reported for the second hole at room temperature in the C-band wavelength region [14,15]. The second and third hole depths became shallower as the saturating signal wavelength became longer. These characteristics of the two holes indicate the decrease in the possible transition between the Stark components of excited and ground levels of Er3þ ion. In much longer wavelength region than 1590.5 nm, the depth of the second hole became deeper again in three EDF samples. The depth of the s second hole was found not to decrease monotonically when the wavelength of the saturating signal becomes longer. In addition, the depth of the second hole at around 1530 nm became deeper in the silica-based EDF that had higher Er3þ ion concentration. This Er3þ ion concentration dependence of the second hole depth has been found in our pervious GSHB study using the saturations signal wavelength of 1550.5 nm. This Er3þ ion concentration dependence was also observed in the case of the third hole, which was deeper in EDF I than in the other EDFs. Er3þ ion concentration dependence of the holes is supposed to be induced by the contribution of Er-Er interaction to the relaxation process of excited Er3þ ions. In this study, the Er3þ ion concentration dependence of the second hole depth was verified to hold in both the C-band and the L-band wavelength regions. Figure 5.24 shows the relationship between the second hole depth and the saturating signal wavelength for three EDF samples when the saturating signal power is 0 dBm. As shown in Figure 5.24, the second hole depth of the saturating signal power 0 dBm decreased as the wavelength of saturating signal became longer up to around 1590 nm. This
Second (dB)hole depth
0 -1
77K Psat :0dBm
-2 -3 sample (130ppm)
-4
sample (700ppm) sample (1600ppm)
-5
1540 1560 1580 1600 1620 1640
Wavelength(nm)
FIGURE 5.24 Relationship between the second hole depth and the saturating signal wavelength for three EDF samples with different Er3þ ion concentrations.
139
140
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
character of the second hole depth at 77 K is consistent with that observed at room temperature in the C-band wavelength region [13], which has been reported to become monotonically shallower as the wavelength of saturating signal became longer. This character of the second hole depth may suggest that the decrease of the transitions from the Stark manifolds within the 4I13/2 state to those within the 4I15/2 state that contribute to the emission corresponds to around 1530 nm. On the other hand, the second hole depth at 77 K showed the local peak around 1590 nm and started to decrease. At wavelengths longer than 1610 nm, the second hole depth stopped decreasing and increased again for three EDF samples. This character of the second hole at 77 K is now under consideration in connection with the real Stark energy structure of Er3þ, where the assumption of equal energy gap (¼3) does not hold, in order to clarify the formation mechanism of the second hole in real EDFs.
5.3.4 Conclusion The saturating power, the wavelength, and Er3þ ion concentration dependence of GSHB in silica-based EDFA at 77 K were reported. The depth of main and second hole decreased as the wavelength of the saturating signal became longer. The main hole width was found to become broader in the L-band region than that in the C-band region. On the other hand, the second hole width did not show the saturation signal wavelength dependence in the C-band and the L-band regions. In addition, the second hole depth of GSHB showed the Er3þ ion concentration dependence and increased with reduction of the Er3þ ion concentration under a high saturating signal condition. We propose the contribution of the energy transfer mechanism among Er3þ ions to the relaxation mechanism of GSHB.
5.4 PRINCIPLES OF GAIN SPECTRAL HOLE BURNING IN ERBIUM-DOPED FIBER AMPLIFIERS The origins of the gain spectral hole structure are explained based on the relation between the multigain spectral hole structure, which consists of spectral holes, and the Stark energy structure of the Er3þ ion. The GSHB is the phenomenon that causes the gain deviation in the wavelength region when some strong signals are input into erbium-doped fiber amplifier (EDFA) [12]. As shown in the previous Section 3, an additional hole (“second hole” or “nonresonant hole”) can be burned at around 1530 nm independent of the wavelength of the saturating signal, where the main hole is burned [13,15]. The effect of these holes is large enough to degrade the gain flatness. Therefore, GSHB in EDFA has become a serious problem for the precise design of the long-haul optical transmission system. In order to predict the gain spectral hole characteristics, many numerical and experimental measurements have been performed to clarify the origin and the mechanism of GSHB [20,24]. However, no complete physical interpretation of GSHB has been formulated yet.
5.4 Principles of gain spectral hole burning in Er-doped fiber amplifiers
There have been many reports of hole burning phenomena in rare-earth doped glasses for applications in high density optical memory. The glass is an inorganic amorphous solid, usually consisting of oxides, which can accommodate allio-valent ions such as trivalent rare-earths, mostly lanthanide, Ln ions in silica or germania glasses. Because doped Ln3þ ions need some charge compensation locally, nonbridging oxygen ions are formed from the glass network. However, these Stark level structures of doped lanthanide ions are not clearly observed at room temperature because of the line broadeing due to the inhomogeneous site variations and the Raman process at elevated temperatures. Therefore, we investigated the gain spectral hole characteristics in silica-based EDFA at 77 K using a saturating signal to clarify the origin of the second hole burned at around 1530 nm. As a result, in addition to the second hole structure, a multihole structure consisting of 11 gain spectral holes was observed in the wavelength region from 1420 nm to 1640 nm. So far, gain spectral hole has been burned at around 1530 nm and at the wavelength of saturation signal. Also, the effect of the gain spectral hole burning on the gain characteristics has been considered to be confined to the narrow wavelength region. However, from our experimental results at 77 K, it was shown that the gain spectral hole structure can have influence on the gain spectrum shape in a much wider wavelength region than expected. Based on the multihole structure in the gain spectrum obtained at 77 K, we estimated the Stark energy structure of Er3þ ion and verified that the estimated Stark energy structure of Er3þ ion showed good agreement with those evaluated by the spectroscopic fluorescence and absorption measurements. The origin of the gain spectral hole was clarified based on the relation between the multihole structure in the gain spectra and the Stark energy structure of Er3þ ion. In addition, Er3þ ion concentration dependence of multigain spectral hole structure is also shown.
5.4.1 Experimental Two silica-based EDF samples with Er3þ concentrations of 130 ppm and 530 ppm were used as the gain medium in GSHB measurements. The refractive indices of EDF samples, measured at 1.3mm with a reflectometer (Ando, AQ7413), were about 1.470.
5.4.2 Results and Discussion 5.4.2.1 Multigain hole structure in erbium-doped fiber Figure 5.25 shows a GSHB spectrum using a saturating signal of 1530.5 nm in the wavelength region from 1510 nm to 1640 nm. The saturation signal power was 0 dBm. We observed a multigain spectral hole structure that consists of eleven holes. The wavelengths of these gain spectral holes were 1490.6 nm, 1504.6 nm, 1519.7 nm, 1530.4 nm, 1539.7 nm, 1550.29 nm, 1557.7 nm, 1580.3 nm, 1596.5 nm, 1615.6 nm, and 1631.5 nm.
141
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
1
0
Hole depth(dB)
142
-1
-2 77K
-3
λ sat :1530.5nm -4
Psat :0dBm 3+
Silica-based EDF(Er 130ppm) -5 1460 1480 1500 1520 1540 1560 1580 1600 1620 1640
Wavelength(nm)
FIGURE 5.25 Multi-gain spectral hole structure of Er3þ ion in the wavelength region from 1460 nm to 1640 nm.
The gain spectral hole in EDF is considered to be formed due to the depopulation of Er3þ ions in the Stark levels within the 4I13/2 manifold by a strong saturating signal. Strong saturation signal generally induces the degradation of the degree of population inversion, which leads to the degradation of gain. If a strong saturating signal (lsat ¼ 1530.5 nm) is input into EDF, the depopulation of the lower Stark levels within the 4I13/2 manifold occurs, which leads to increase in the population of the lower Stark levels within the 4I15/2 manifold. The difference of gain can be expressed in terms of the electronic populations of initial and final level by the relation: Z (1) DGainfexp ðDN2j se DN1j sa Þdz where se and sa are the emission and absorption cross sections of Er3þ ion, respectively. DN1j and DN2j are the difference of the energy between the jth Stark level within the 4I13/2 and the lowest Stark level within the 4I15/2. In GSHB measurements, the saturation signal induces the decrease of DN2j and the increase of DN1j, which leads to the decrease of the gain of the EDF. Therefore, these 11 gain spectral holes shown in Figure 5.25 indicate the degradation of the degree of the population inversion at each wavelength. The difference between the gain spectral hole depths in the longer and shorter wavelength region than 1520 nm was observed. One of the reasons of this difference in the gain spectral hole depth can be the difference in the number of Er3þ ions which contribute to the emission and absorption at each wavelength.
5.4 Principles of gain spectral hole burning in Er-doped fiber amplifiers
5.4.2.2 Relation between the multihole GSHB structure and the Stark energy structure of Er3þ ion Gain spectral hole in EDF is considered to be formed due to the depopulation by a strong saturating signal of Er3þ ions of which the Stark levels within the 4I13/2 manifold are populated. At 77 K, it is enough to consider the depopulation of the lowest Stark levels within the 4I13/2 because most of the excited Er3þ ions are populated in the lowest Stark sublevels within the 4I13/2 at such low temperature. In addition, when the saturation signal of 1530.5 nm is input into EDF, the depopulation of the lowest Stark sublevel within the 4I13/2 is considered to have significant influence on the formation of the gain spectral hole structure because the energy of 1530.5 nm-photon almost corresponds to the energy gap between the lowest Stark level within the 4I13/2 manifold and that within the 4I15/2 manifold. Therefore, 11 gain spectral holes shown in Figure 5.26 could be burned mainly due to the depopulation of the lowest Stark level within the 4I13/2 manifold. Based on these considerations, the Stark energy structure of Er3þ ion in EDF I 3þ (Er : 130 ppm) can be estimated. Figure 5.26 shows the estimated Stark energy structure of Er3þ ion. Table 5.3 shows the comparison of the estimated Stark level energies with those obtained in different spectroscopic method for Er: silica glasses 6950 6900 6850 6800 6750 6700
6708.8
6650
6646.3
Energy (cm-1)
6600 6550
6580.2 6534.2
6500 450 400
404.9
350
344.6
300 250 200 150 100 50 0
FIGURE 5.26 Estimated Stark energy structure of Er3þ ion.
270.5 206.3 114.5 83.8 39.5 0
143
144
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
Table 5.3 Estimated Stark energy structure of Er3+ ion DE (cm-1)
DE (cm-1) Ref.6
DE (cm-1) Ref.7
4
I15/2
0 39.5 83.8 114.5 206.3 270.5 344.6 404.6
0 26–34 51–59 125–133 201 268 – –
0 25 55 95 – – – 445
4
I13/2
(6534.2) 0 – 46.1 – 112.0 174.5 – –
(6540) 0–8 – – – 96–104 163–171 222-230
(6515) 0 35 45 70 – – – 305
[24,25]. The estimated Stark energy structure of Er3þ ion shows almost good agreement with that obtained by the other spectroscopic measurements of fluorescence and absorption in erbium-doped fibers [24,25]. Therefore, the origin of the gain spectral hole structure should be the Stark energy structure of the 4I13/2 and the 4I15/2 of Er3þ ion.
5.4.2.3 Effect of Er3þ concentration Figure 5.27 shows a comparison of GSHB spectra of EDFs with different Er concentration. As described in previous sections, clear structures of multiholes are observed in the EDF of 130 ppm, whereas the number of clear holes decreased in the spectrum for higher Er concentration (530 ppm). Deepest holes are observed at 1540 nm and 1530 nm, which correspond to the main (signal wavelength) and the second holes, respectively. The hole at around 1520 nm, which was named as “the third hole” above in Section 3, is still observed in EDF (530 ppm). However, the holes at longer wavelength regions, especially in the L-band region, were almost completely diminished. It is clear that depth of all the holes became shallower in the EDF with higher Er concentration. These observations can be explained qualitatively by taking into account the different transient behaviors of populations of the Stark manifolds in the Er3þ: 4I13/2 and 4I15/2 levels. As has been discussed, the origin of gain spectral holes can be considered to be a result of transient depopulation of some Stark levels in the 4I13/2 level by a strong saturation signal. When the Er concentration in the silica glass matrix is low, each Er3þ ions shows spectral response almost independent of the other Er3þ ions, because the interaction between 4f electronic states is caused
5.5 Positive gain change of gain spectral hole burning
(a) Hole depth(dB)
1 0
-1 -2 -3 -4
1520
1540
1560
1580
1600
1620
1640
Wavelength(nm)
(b) Hole depth(dB)
1 0
-1 3+
-2
Er 530ppm sat :1540.5nm
-3
Psat :0dBm
-4
1520
1540
1560
1580
1600
1620
1640
Wavelength(nm)
FIGURE 5.27 Gain spectral hole structures in EDFs with different Er3þ ion content (a) Er3+: 130ppm, (b) Er3þ: 530ppm.
by dipole-dipole, and is weak. However, with increasing Er concentration, the interaction, the strength of which depends on the inverse sixth power of the distance, usually arises from the nearest neighbors, and ceases to be negligible. After the energy interaction, usually resulting in the energy exchange between an Er in the ground state (4I15/2) and another in the excited (4I13/2) state, the population states approach the Boltzmann distribution in both energy levels. This situation of thermal equilibrium does not hold for the Er3þ ions that were just burned by input of a strong saturation signal. It is difficult to estimate the transient time of this energy transfer, but it should certainly be shorter in EDF with higher Er concentration. The effect of this ion-ion interaction certainly causes reduction of the hole depth, especially for the holes at a longer wavelength, where the deviation from the thermal equilibrium state by saturation signal cannot be large because of narrower Boltzmann distribution. It is considered that usage of EDF with higher Er content would reduce the GSHB issue; however, it would also reduce the gain and power conversion efficiency of EDFA.
145
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
5.5 POSITIVE GAIN CHANGE OF GAIN SPECTRAL HOLE BURNING OBSERVED AT ROOM TEMPERATURE In a previous section, we explained the origin of the GSHB of the EDF by observing the gain spectral hole characteristics at 77 K. However, for practical systems it is important to understand the GSHB behavior at room temperature, which is nominally the ambient temperature of most field deployed EDFAs. In this section, we explain the positive gain change of GSHB observed by applying the subtracting technique which is one of the unique behaviors of GSHB at room temperature. Figure 5.28 plots an example of a positive gain change of GSHB observed at room temperature by using the subtracting technique depicted in Figure 5.9. The wavelength and the input power of the saturation signal are 1540 nm and -10 dBm, respectively and the wavelength of the gain-locking is 1560 nm. In this figure, the positive gain change is observed at a wavelength between the main hole and the second hole. This positive gain change occurs because of the gain shift due to the gain-locking operation in the subtracting technique. When the gain at the gain-locking wavelength is affected by GSHB and decreases, the pump power is controlled to increase the gain of the gain-locking signal, then the gain shift corresponding to the compensation of the gain change at the gain-locking wavelength occurs. As a result, the positive gain change is observed at the wavelength at which the gain change of the gain shift is larger than the gain change of GSHB. At 77 K, the holes of the multihole structure are separated, as shown in Figure 5.25 and the additional gain shift due to the gain-locking operation can be avoided by choosing the gain-locking wavelength at the wavelength that is located between the holes. However, at room temperature, the gain change due to GSHB is broadly spread over the gain spectrum, and it is difficult to choose the gain-locking in which the GSHB gain change is sufficiently small. Therefore, at room temperature, it is difficult to separate the original gain change of GSHB and the gain change 0.4 Positive gain change
Gain change (dB)
146
0.2 0 -0.2
Second hole
-0.4 -0.6 -0.8 1520
Main hole Gain-lock signal
Saturation signal 1530
1540
1550
1560
Wavelength (nm)
FIGURE 5.28 Positive gain change observed by applying the subtracting technique at room temperature.
5.6 Numerical model of GSHB in EDFA
due to the gain shift of the gain-locking operation and it must be treated as one gain change. In the next section, we will describe the numerical model of GSHB that can simulate both gain changes at the same time.
5.6 NUMERICAL MODEL OF GSHB IN EDFA GSHB of the EDFA is one of the origins of the static gain change that degrades the performance of the ROADM system, and the effect of GSHB must be taken into account when estimating the impact of the static gain change of EDFA on the system. The effect of GSHB depends on the operating conditions of the EDFA, including factors such as input signal power, the number and wavelengths of signals, and gain. It is important to estimate the GSHB gain change under the predictable conditions of the optical amplifiers deployed in ROADM systems. A software simulation based on an accurate and simple model will be a flexible and cost effective tool to estimate the GSHB gain change compared with the hardware simulation. However, as we explained in a previous section, the GSHB is caused by the Stark level structures of Er3þ ions and is too complicated to model analytically. When the analytical model is used in the simulation, it is difficult to acquire the accurate simulation results in practical calculation time. Therefore, a simple and accurate numerical model of GSHB is desiable to enhance the effectiveness of the software simulation. In this section, we explain a simple GSHB numerical model that covers the wide operating conditions and enables accurate estimation of the static gain change of ROADM systems.
5.6.1 Description of the GSHB numerical model Several studies have attempted to model the GSHB phenomenon of EDFAs [26,27]. Desurvire proposed an inhomogeneous gain model to describe the GSHB [26]. In that inhomogeneous gain model, the gain spectrum of the EDF is modeled by superposing the laser transitions between each pair of Stark split levels. The rate equations are written for each laser transition, and GSHB is described as the saturation of certain laser transitions. The inhomogeneous gain model describes the phenomenon theoretically; however, it is difficult to acquire accurate physical parameters for each laser transition, and a long calculation time is necessary for accurate simulation. A simple numerical model of GSHB was proposed by Aizawa, et al. [27]. They proposed calculating the GSHB gain change of the multichannel amplification from the previously measured results of the single-channel amplification. However, their model only simulates the main hole, and the effect of the second hole is not included. Prior measurement must be carried out on the simulating EDFA under certain conditions. Therefore, an additional measurement is needed every time there is a change in the calculation conditions or the EDFA configuration, such as the length of the EDF or the pumping direction.
147
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
The GSHB models proposed previously have problems in accuracy, simplicity, or calculation conditions. We describe a new GSHB numerical model based on the homogeneous gain model, which is a conventional EDF model without GSHB phenomenon [15,28]. The homogeneous gain model assumes that the laser transitions of Er3þ ions are broadened and saturated homogeneously [26]. The gain spectrum of Er3þ ions is determined by a population inversion ratio that is the atomic population density ratio of upper level (4I13/2) and lower level (4I15/2) of the quasitwo-level laser system. To model the fiber waveguide of EDF, the fiber is divided into infinitesimal slices, and the propagation equation is solved. The gain spectrum is calculated from the population inversion at each slice of the fiber. The homogeneous gain model is commonly used because it is simple and accurate under the conditions in which the effects of GSHB can be ignored.
(a)
(b)
Superposition of two Gaussian
Δz
Positive Gaussian
EDF
Saturation P s(z ) Signal
Inversion ratio
Negative Gaussian
λ sat
~1530 nm n ave
P s(z +Δz )
Wavelength
P s(z +Δ z ) - P s(z ) = G (n ave + Δn SHB(λ ))P S(z )
(c)
Peak of Δ n SHB C (a.u.)
Second hole
Peak of Δ n SHB C (a.u.)
148
0
-0.01
-0.02
-0.03 1530
1535
1540
1545
1550
Main hole
0
-0.01
-0.02
-0.03
0
Wavelength (nm)
2
4
6
Total power at z (mW)
(a) Introduction of the population inversion fluctuation at each longitudinal position of EDF, (b) wavelength dependence of the population inversion, (c) power dependence and wavelength dependence of peak value C of the main hole.
FIGURE 5.29 Numerical simulation of SHB phenomena by introducing the wavelength-dependent population inversion fluctuation.
5.6 Numerical model of GSHB in EDFA
To model the inhomogeneity of the GSHB, a wavelength-dependent fluctuation of the population inversion DnSHB ðz; lÞ is introduced to the homogenous gain model (Figure 5.29(a)). The population inversion fluctuation DnSHB ðz; lÞ represents the saturation of the certain laser transition caused by the GSHB. Here, DnSHB ðz; lÞ is calculated by the signal conditions of each slice and added to a wavelength-independent average population inversion nave ðzÞ of the homogeneous gain model. The propagation equation of channel i is written as dPi ðzÞ ¼ ½ðgi þ ai Þfnave ðzÞ þ DnSHB;i ðz; li Þg ðai þ lÞPi ðzÞ; dz where Pi ðzÞ is the power of the channel i at the longitudinal position of z, li is the wavelength of the channel i, gi and ai are the gain and absorption coefficients at li , and l is fiber loss. Figure 5.29(b) shows the wavelength dependence of DnSHB;i ðz; lÞ of the singlechannel amplification. Here, DnSHB;i ðz; lÞ models the gain-change shape of GSHB at room temperature and has two components corresponding to the main hole and the second hole of the GSHB. The main hole component is modeled by a negative Gaussian function, whose center wavelength is li . The second hole component is modeled by superposition of the negative and positive Gaussian functions at the wavelength around 1530 nm. DnSHB;i ðz; lÞ is expressed by following formula. ðlli Þ2 DnSHB;i ðl; zÞ ¼ Cðli ; Pi ðzÞ; Ptotal ðzÞÞexp lnð2Þ ðWmain =2Þ2 P ðllsec;j Þ2 Dj ð nave ðzÞ; Pi ðzÞ; Ptotal ðzÞÞexp lnð2Þ þ 2 j ¼ 1;2
ðWsec;j =2Þ
Ptotal ðzÞ is the total power of the signals at the longitudinal position z, Cðli ; Pi ðzÞ; Ptotal ðzÞÞDj ð nave ðzÞ; Pi ðzÞ; Ptotal ðzÞÞ are the peak values of the Gaussian functions, Wmain and Wsec;j are the widths of the Gaussian functions, and li and lsec;j are the center wavelengths of the Gaussian functions. The peak value of the main hole Cðli ; Pi ðzÞ; Ptotal ðzÞÞ depends on the signal power and wavelength, and the peak values of the second hole Dj ð nave ðzÞ; Pi ðzÞ; Ptotal ðzÞÞ depend on the signal power and average population inversion nave ðzÞ. The example of the wavelength dependence and the total power dependence of the peak value Cðli ; Pi ðzÞ; Ptotal ðzÞÞ are depicted in Figure 5.29(c). The wavelength and total power dependence of Cðli ; Pi ðzÞ; Ptotal ðzÞÞ shows that the depth of the main hole increases as the wavelength of the signal becomes shorter or the total power of signals increases. The total population inversion fluctuation of the multichannel amplification is modeled by superposition of the population inversion fluctuation of each channel (Figure 5.30). Because the GSHB numerical model is based on the conventional homogeneous gain model, the calculation is simple and fast. The calculation parameters are determined from the experimental results of the input power dependence, signal wavelength dependence, and gain dependence.
149
150
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
Inversion ratio
Saturation signal (wavelength λ i)
Superposition of each channel
nave Wavelength
Inversion ratio
Population inversion of Channel i amplification
Inversion ratio
Saturation signals
Multi-channel amplification Saturation signal (wavelength λ j)
nave Wavelength
nave Wavelength
Inversion ratio of amplification of two channels
Population inversion of Channel j amplification
FIGURE 5.30 SHB model of multi-channel amplification.
5.6.2 Calculated results of the GSHB simulation under various conditions Calculated and measured results were compared under various conditions that reflect the operating conditions of optical amplifiers of ROADM systems [28]. The 7.2 m-long EDF, whose Er concentration was 1900 ppm, was forward pumped by a 980 nm LD, and the GSHB phenomenon was measured by using the subtracting technique depicted in Figure 5.9. The same calculation parameters were used for the calculation of all conditions. Figure 5.31 shows the measured and calculated results of an GSHB gain change of the single channel amplification. Symbols and lines are the measured and calculated results, respectively. The dependence of the GSHB gain change on signal wavelength is plotted in Figure 5.31(a). The gain change of the signal wavelength decreased as the saturation signal wavelength increased. When the signal wavelength was 1530 nm, the main and second holes overlapped, and the gain change became especially large. Figure 5.31(b) shows the dependence on input power. As the input power increased, the depth of both main and second holes increased. The dependence on gain, shown in Figure 5.31(c), was measured by changing the gain of the gain-locking signal, whose wavelength was 1560 nm. As the gain increased, the gain change near 1530 nm changed from negative to positive. These figures show that the calculated results of the GSHB numerical model agreed well with the measured results. Next, we show the GSHB gain change of multichannel amplification. Figure 5.32(a) shows the experimental setup. The GSHB gain change was measured by observing the difference in the gain spectrum of 15 dBm/ch input power and 45 dBm/ch input power. The pump power of the EDF is controlled to fix the gain of the gain-locking signal, whose wavelength was 1560 nm. The
5.6 Numerical model of GSHB in EDFA
(b)0.2
0
-1 1550 nm
1541 nm
λ sat= 1530 nm
-2
-3 1525
Symbols: Measured Lines Calculated
1535
1545
1555
Wavelength (nm)
Gain change (dB)
Gain change (dB)
(a)
P sat = -20 dBm
0 -0.2
-10 dBm
-0.4 2.3 dBm -0.6 -0.8 -1 1525
Symbols: Measured Lines Calculated 1535
1545
1555
Wavelength (nm)
(c) Gain change (dB)
1 GGL = 29.5 dB
(a) Wavelength dependence, (b) input power dependence, (c) gain dependence.
0.5 25 dB 0 -0.5 20 dB -1 1525
1535
Symbol: Measured Line Calculated 1545
1555
Wavelength (nm)
FIGURE 5.31 Calculated and measured results of SHB phenomena under various conditions.
multichannel signal had eight saturation signals with three different channel allocations: (1) all eight channels were located on a short wavelength; (2) four channels were located on a short wavelength and the other four on a long wavelength; and (3) all eight channels were located on a long wavelength (Figure 5.32[b]). The EDFA configuration was the same as in Figure 5.31, and the same calculation parameters were used for the calculation. Figure 5.32(c) shows the measured and calculated results of the GSHB gain change of multichannel amplification. The difference between the measured results and the calculated results was less than 0.6 dB and agreed well. In this section, we explained the simple GSHB numerical model based on the conventional homogeneous gain model. The GSHB phenomenon is modeled by introducing the wavelength-dependent fluctuation of the population inversion. The calculation results of the numerical model agreed well with the measured results under various conditions. By introducing the GSHB numerical model to the system design, the designer of the system can easily estimate the impact of GSHB on the ROADM system.
151
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
(a)
(b) Saturation signals
Saturation signals (Psat= -15 or –45 dBm/ch)
EDFA EDF
AWG
DFB LD DFB LD
3 dB coupler
980 nm / 1550 nm coupler Isolator
Isolator
DFB LD
Optical spectrum analyzer
1525
1535
1545
1555
Wavelength (nm)
980 nm pump LD
LD source Gain-lock signal
1525
(1560 nm, PGL=-30 dBm, GGL= 20 dB)
1535
1545
1555
Wavelength (nm)
Gain change (dB)
(c) 0.5 1.4 dB 0 -0.5 Gain-lock -1 Saturation signals signal -1.5 (8ch + 0ch) -2 1525 1535 1545 1555 1565
Gain change (dB)
Wavelength (nm)
Gain change (dB)
152
0.5 1.0 dB 0 -0.5 Saturation signals -1 (4ch + 4ch) -1.5 -2 1525 1535 1545
1525
1535
1545
1555
Wavelength (nm)
(a) Experimental setup, (b) channel allocations, (c) measured and calculated results.
Gain-lock signal 1555
1565
Wavelength (nm) 0.5
0
0.3 dB -0.5 -1 Saturation signals -1.5 (0ch + 8ch) -2 1525 1535 1545
Gain-lock signal 1555
1565
Wavelength (nm)
: Measured
: Calculated
FIGURE 5.32 Channel allocation dependence of SHB phenomena.
5.7 SUMMARY In this chapter, we reviewed the dynamic and static gain changes of EDFAs when the channel count or allocation was dynamically changed by re-routing of the ROADM system. The AGC circuit with the combination of feedforward and feedback was effective for the suppression of the dynamic gain change. For the static gain change, we focused especially on GSHB and described the origin and the numerical model of GSHB. Currently, a hardware simulation is needed to investigate the dynamics of the ROADM system. However, introduction of the GSHB numerical simulation described in this chapter makes it possible to numerically analyze the behavior of the dynamics, and easier to investigate.
References
ACRONYMS AGC APC EDF EDFA GEQ GMPLS GSHB LD OSA ROADM SHB SRS WDM
Auto gain control Auto power control Erbium-doped fiber Erbium-doped fiber amplifier Gain equalizer Generalized multiprotocol label switching Gain spectral hole burning Laser diode Optical spectrum analyzer Reconfigurable optical add-drop multiplexer Spectral hole burning Stimulated Raman scattering Wavelength division multiplexed
References [1] K. Grobe, Applications of ROADMs and control planes in metro and regional networks, OFC/NFOEC2007, Paper NTuC1 (2007). [2] E. Desurvire, Spectral noise figure of Er3þ-doped fiber amplifiers, IEEE Photon. Technol. Lett. 2 (3) (1990) 208. [3] S. Kinoshita, C. Oshima, H. Itoh, T. Kobayashi, Y. Sugaya, T. Okiyama, Wide dynamic-range WDM optical amplifiers for 32 x 10 Gb/s, SMF transmission systems, OAA1998, Paper WA2 (1998). [4] K. Motoshima, N. Suzuki, K. Shimizu, K. Kasahara, T. Kitayama, T. Yasui, A channel-number insensitive erbium-doped fiber amplifier with automatic gain and power regulation function, J. Lightwave Technol. 19 (2001) 1759. [5] C. Wang, G.J. Cowle, Optical gain control of Erbium-doped fiber amplifiers with a saturable absorber, OAA1999 Tech. Dig. 110 (1999). [6] Y. Liu, M.F. Sharma, Krol, Dual-cavity optical gain control for EDFA’s and EDFA cascades, OAA1999 Tech. Dig. 115 (1999). [7] J.F. Massicott, S.D. Willson, R. Wyatt, J.R. Armitage, R. Kashyap, D. Williams, et al., 1480nm pumped erbium-doped fiber amplifier with all optical automatic gain control, Electron. Lett. 30 (1994) 962. [8] M. Fukutoku, M. Jinno, Pump power reduction of optical feedback controlled EDFA using electrical feedforward control, OAA1998 Tech. Dig. 32 (1998). [9] S. Sergeyev, E. Vanin, G. Jacobsen, Gain-clamped dynamics in EDFA with combined electronic feed-forward-Optical feedback control, OFC2002 Tech. Dig. 518 (2002). [10] C. Tian, S. Kinoshita, Novel solution for transient control of WDM amplifiers using the combination of electrical feedforward and feedback, CLEO’02 Tech. Dig. 430 (2002).
153
154
CHAPTER 5 Gain changes of optical amplifiers at ROADM nodes
[11] C. Tian, S. Kinoshita, Analysis and control of transient dynamics of EDFA pumped by 1480- and 980-nm lasers, J. Lightwave Technol. 21 (8) (2003) 1728. [12] E. Desurvire, J.L. Zyskind, J.R. Simpson, Study of spectral dependence of gain saturation and effect of inhomogeneous broadening in erbium-doped aluminosilicate fiber amplifiers, IEEE Photon. Technol. Lett. 2 (4) (1990) 246e248. [13] E. Rudkevich, D.M. Baney, J. Stimple, D. Derickson, G. Wang, Nonresonant spectralhole burning in erbium-doped fiber amplifier, IEEE Photon. Technol. Lett. 11 (1999) 542e544. [14] N.S. Bergano, C.R. Davidson, M.A. Mills, P.C. Corbett, S.G. Ecangelides, B. Pederson, et al. In Proceedings of Optical Fiber Conference, OFC97, (1997), PD16. [15] M. Nishihara, Y. Sugaya, E. Ishikawa. Characterization and new numerical model of spectral hole burning in broadband erbium-doped fiber amplifier. In Proceedings of Optical Amplifiers and Their Applications, OAA (2003) Tud3. [16] J. Chun, S. Yong Kim, C.J. Chae. Performance of all optical gain-clamped EDFAs with different feedback wavelength for use in multiwavelength optical networks. In Proceedings of Optical Fiber Conference, OFC97, TuE5, (1997) 22e23. [17] G. Luo. Relaxation oscillations and spectral hole burning in laser automatic gain control of EDFAs. In Proceedings of Optical Fiber Conference, OFC97, WF4, (1997) 130e131. [18] M. Bolshtyansky, Spectral hole burning in erbium-doped fiber amplifiers, IEEE Photon. Technol. Lett. 21 (4) (2003) 1032e1038. [19] J.W. Sulhoff, A.K. Srivastava, C. Wolf, Y. Sun, J.L. Zyskind, Spectral-hole burning in erbium-doped silica and fluoride fibers, IEEE Photon. Technol. Lett. 9 (12) (1997) 1578e1579. [20] I. Joindot, F. Dupre, Spectral hole burning in silica-based and in fluoride-based optical fibre amplifiers, Electron. Lett. 33 (1997) 1239. [21] L. Bigot, A. Jurdyc, B. Jacquier, L. Gasca, D. Bayart, Resonant fluorescence line narrowing measurements in erbium-doped glasses for optical amplifiers, Phys. Rev. B 66 (2002) 1. [22] D.L. Dexter, A theory of sensitized luminescence in solids, J. Chem. Phys. 21 (1953) 836e850. [23] V.P. Gapontsv, N.S. Platonov, Migration-Accelerated Quenching in Glasses Activated by Rare Earth Ions, in: Dynamical Processes in Disordered Systems, 51, Material Science Forum, 1990. [24] C.C. Robinson, Multiple sites for Er3þ in alkali silicate glasses (I). The principal sixfold coordinated site of Er3þ in silicate glass, J. Non-Cryst. Solids 15 (1) (1974) 1e9. [25] E. Desurvire, J.R. Simpson, Evaluation of 4I15/2 and 4I13/2 Stark-level energies in erbium-doped aluminosilicate glass fibers, Opt. Lett. 15, (10) (1990) 547e550. [26] E. Desurvire, Erbium-doped Fiber Amplifiers Principles and Applications, John Wiley & Sons, Inc., New York, 1994. [27] T. Aizawa, T. Sakai, A. Wada, R. Yamauchi, Effect of Spectral-Hole Burning on MultiChannel EDFA Gain Profile, OFC’99, WG-1 (1999). [28] M. Nishihara, Y. Sugaya, E. Ishikawa, Impact of spectral hole burning in multichannel amplification of EDFA, OFC2004, FB1 (2004).
CHAPTER
Mastering Power TransientsdA Prerequisite for Future Optical Networks
6 Peter Krummrich
TU Dortmund (University of technology)
CHAPTER OUTLINE HEAD 6.1. Introduction to power transients in optical networks...................................... 6.1.1. Electronic gain control ............................................................... 6.1.2. Remaining effects...................................................................... 6.1.2.1. Gain control error due to gain spectrum ripple ..................... 6.1.2.2. Gain variations due to spectral hole burning ........................ 6.1.2.3. Stimulated Raman scattering in the transmission fiber ......... 6.2. Overview of transient suppression concepts .................................................. 6.2.1. Linearized system ...................................................................... 6.2.2. Replacement signal.................................................................... 6.2.3. Retilting element ....................................................................... 6.2.4. Individual channel power control ................................................. 6.3. Compensation of power transients introduced by SRS .................................... 6.3.1. Numerical investigation of SRS tilt compensation......................... 6.3.1.1. Description of the numerical model and simulations ............ 6.3.1.2. Simulation results and discussion ........................................ 6.3.1.3. Transient propagation in transparent photonic networks....... 6.3.2. Experimental investigation of SRS tilt compensation..................... 6.4. Compensation of transients from all sources ................................................. 6.4.1. Individual channel power control ................................................. 6.4.2. Fill signals ................................................................................ 6.4.3. Linearized system ...................................................................... 6.5. Summary and conclusions ........................................................................... Acronyms ........................................................................................................... References .........................................................................................................
156 157 162 163 164 165 168 168 169 170 171 171 172 172 173 179 183 188 188 189 191 192 193 194
Transparent optical networks are currently the most cost efficient and power efficient solution for the terrestrial transport of multi-Terabit data traffic over distances from a few hundred up to a few thousand kilometers [1e3]. They should enable flexible activation and deactivation of wavelength channels for easy operation and adaptation to changing traffic patterns. Moreover, component failures or fiber cuts should not affect the performance of channels not passing through the distorted element. Optically Amplified WDM Networks. DOI: 10.1016/B978-0-12-374965-9.10006-8 Copyright Ó 2011 Elsevier Inc. All rights reserved.
155
156
CHAPTER 6 Mastering power transients
Optical power transients constitute a major obstacle, both for intentional channel switching flexibility and for resilience in case of failures. They have to be suppressed below acceptable limits to guarantee the unaffected operation of surviving channels.
6.1 INTRODUCTION TO POWER TRANSIENTS IN OPTICAL NETWORKS To be able to reduce the magnitude of optical power transients in transparent photonic networks, it is necessary to understand their origins and characteristics. This section addresses how and where the transients, i.e., fluctuations of the time averaged channel powers on timescales ranging from microseconds to milliseconds, are generated. Looking for the source of channel power transients usually leads to erbium-doped fiber amplifiers (EDFAs). They contribute a major source of transients in an optical transmission system, but as will be discussed later in this section, they are not the only source. The dynamic behavior of EDFAs was studied intensively over many years [4e81], with a strong peak around the year 1997. Without any compensation, transients generated in EDFAs will probably dominate over the ones from all other sources. So, generation of transients in EDFAs is addressed first. The fact that the power of a given channel at the output of an optical amplifier in a wavelength division multiplexing (WDM) system depends on the power of the other channels follows from the rule of energy conservation. Optical amplifiers are usually operated in saturation to achieve good pump to signal power conversion efficiency. Generation of the required several dozen to a few hundred milliwatts of pump power requires expensive pump lasers. Operation with less pump power reduces component cost as well as electrical power consumption for driving the laser and cooling the device. Figure 6.1 depicts gain G and total output power POUT, Total versus total input power PIN, Total in the two operating regimes of optical amplifiers. In linear operation, the signal power at the amplifier output is proportional to the power at the amplifier input. As a consequence of the conservation of energy, the linear regime can obviously not extend to continuously increasing input powers. At some point, the output power would exceed the sum of the input power and the pump power, resulting in more energy leaving the device than entering it. Such a net flow of energy out of the amplifier can occur for a short amount of time, if energy was stored in the device, but not in the steady state. For constant pump power and high small signal gain values, the output power has to deviate from proportional growth with the input power as soon as it reaches a significant fraction of the pump power in steady state operation. The corresponding reduction of the amplifier gain provides an indication for the onset of saturation. The coupling of WDM channel powers at the amplifier output results from operation in saturation and conservation of energy. This can be understood by looking at a simple example. An optical amplifier is operated with a single input channel. The channel power at the amplifier output corresponds to one-third of the
6.1 Introduction to power transients in optical networks
Linear regime
Saturated regime
log POUT,Total
log G
log PIN,Total
FIGURE 6.1 Operating regimes of optical amplifiers
pump power. If the gain was constant, activation of three additional channels with the same amplifier input power and gain as the initially present channel would result in a total output power exceeding the pump power. Consequently, the amplifier gain has to decrease when more channels are added to keep the output power below the pump power. The gain reduction decreases the output power of the channel which was already present before the other channels were added. For amplifier operation in saturation with constant pump power, the output power of a given WDM channel depends on the number of active channels. Activation or deactivation of channels results in power fluctuations of channels present before the switching event. These power fluctuations can grow when propagating through a cascade of amplifiers due to their nonlinear behaviour and result in detrimental power transients at the input of a receiver.
6.1.1 Electronic gain control The explanation of the origin of power transients leads directly to an approach to reduce their magnitude. If operating with constant pump power results in power changes of initially present channels after activation or deactivation of channels, adaptation of the pump power to the number of active channels may reduce the power fluctuations. As the relation between the required pump power and the channel loading at different wavelengths turns out to be rather complex, it is not easy to directly set the pump power. Instead, using a control loop provides an easier solution to determine the necessary amount of pump power. The target is to keep the power of a given channel at the amplifier output constant when other channels are switched. This cannot be achieved by simply monitoring
157
158
CHAPTER 6 Mastering power transients
the total power at the amplifier output, as the resulting power changes with the number of active channels. Individual monitoring of the channel powers that have to be kept constant would provide a solution, but it requires rather complex fast wavelength selective monitoring devices. An easier solution can be realized by constant gain control. If all channels experience equal and constant gain, adding channels will increase the total amplifier input power and the total output power, but the ratio of the total powers stays constant. Moreover, the output power of the initially present channels does not change as long as its input power and gain remains constant. With the quoted assumptions, controlling the pump power to keep the ratio between the total amplifier output power to the total input power constant provides the desired constant output power of initially present channels. As this concept can be implemented rather easily and cost effectively and results in good power fluctuation reduction performance even in case of fast changes of the total input power, it is realized in the majority of amplifier gain blocks found in WDM transmission systems. Figure 6.2 depicts a block diagram of the electronic realization of the gain control concept. Pump power is generated by a laser diode. Its output signal is coupled into the signal path by a WDM coupler and launched into the erbium-doped fiber. A tap coupler following the amplifier input diverts a small fraction of the total input power to a first photo diode. A small fraction of the total output power is fed to the second photo diode by the tap coupler before the amplifier output. Both photo diodes convert the optical powers to electrical signals and send them to a controller. The controller tries to keep the ratio of the total output and input signal powers constant by adjusting the injection current of the pump laser. Electronic gain control has been introduced in early generations of WDM transmission systems since approximately 1997. At this time, systems had a maximum
Tap coupler 1
WDM coupler
Erbium doped fiber Tap coupler 2
Pump laser
Photo diode 1
FIGURE 6.2 Electronic gain control
Controller
Photo diode 2
6.1 Introduction to power transients in optical networks
capacity around eight channels per fiber and direction covering only a fraction of the C-band and provided point-to-point links between terminals with a transparent reach of a few hundred kilometers. Channel power transient suppression performance of electronic gain control was very well suited for these systems. However, optical transport networks have evolved significantly since the early days. The current generation of terrestrial long-haul systems with a transparent reach of more than 1000 km carries up to 80 channels per fiber and direction, covering the entire C-band in a wavelength range from approximately 1530 nm to 1565 nm with a channel spacing of 50 GHz. Channels are launched into the fiber with power around 0 dBm, resulting in a total launch power of approximately 20 dBm. Metro systems with a transparent reach up to a few hundred kilometers use the same wavelength range, but a wider channel spacing of 100 GHz, resulting in a smaller channel count of up to 40. Total fiber launch powers are usually lower than 20 dBm in these systems. Transient suppression means reduction of power swings below acceptable limits. So what are these limits in long-haul and metro systems? Finding an answer is not an easy task, because it depends on many system parameters. First of all, we need to find a criterion that can be used to differentiate whether a given power transient can be tolerated or not. A good choice may be the bit error ratio (BER) of an affected channel after forward error correction (FEC). As long as the BER after FEC remains below 10-15 during the power fluctuation, the system is able to maintain error free operation of the channel despite the transient. Translation of this requirement into limits for the channel power swing magnitude, speed, and shape at the receiver turns out to be quite difficult. The BER after FEC during power fluctuations depends on many parameters like the characteristics of the FEC, the modulation format, the characteristics of the decision level control, the implementation of equalizers, if they are applied, the tolerance of the system against nonlinear effects, and the optical signal to noise ratio (OSNR) margin. How does a fast time-averaged channel power fluctuation affect the decision process in the receiver? An undershoot of the channel power results in a reduction of the OSNR. The reduced signal to noise ratio will increase the number of wrong symbol decisions and has a potential to cause a growth of the number of remaining bit errors after FEC decoding. Moreover, the decision level control may not be able to follow, resulting in a non-optimum decision level setting during the transient event. A fast overshoot of the channel power increases the OSNR but may also increase the impact of nonlinear effects in the transmission fiber. The latter can increase the number of wrong symbol decisions, for example by distorting the signal shape. As can be concluded from the discussion, there is no general answer to the question of the acceptable limits for channel power transients. The limits will depend on many characteristics of the system transponders, the dispersion compensation scheme, and even the transmission fiber. As a very rough estimate for a power swing with nearly the same power before and after the transient and a duration in the order of 1 ms, a power undershoot up to approximately 2 dB and a power overshoot up to approximately 3 dB may be tolerable in a long-haul system.
159
160
CHAPTER 6 Mastering power transients
Metro systems should generally be more tolerant to power overshoots due to the less pronounced impact of nonlinear effects. The longer transparent reach, wider wavelength band and higher total launch power of current WDM system generations compared with the early ones have turned channel power transient suppression into a more difficult task. However, even stronger challenges arise from the different topology and usage patterns of transparent photonic networks. It is important to note these differences to understand power transient issues of current and future optical networks. The transition from point-to-point configurations used in early WDM systems to meshed transparent photonic networks is depicted in Figure 6.3. Before 2000, WDM transmission systems in most cases provided mere data transport pipes between synchronous optical network (SONET) or synchronous digital hierarchy (SDH) equipment. All switching as well as adding or dropping of channels was done in the electrical domain. A change of the WDM channel number in these years was a rare event occurring in intervals of weeks or months. The most likely reason for channel number alterations was the installation of new transponders and activation of additional channels to increase the link capacity. As the power of the new channel can be increased slowly and the time of the activation can be moved to service intervals, handling power transients caused by these events was rather easy. Slightly higher challenges were created by fast channel deactivation due to transponder or component failures as well as disconnection of the wrong patch cable at a patch panel during service activities. Power transient performance provided by electronic gain control was usually also sufficient to handle these unintentional events.
FIGURE 6.3 Transition from point-to-point links to meshed transparent photonic networks
6.1 Introduction to power transients in optical networks
Fiber cuts are a major cause of a fast deactivation of many channels. However, these events did not create challenges with respect to power transient suppression. In a point-to-point link, there are no surviving channels that have to be protected against transients if the fiber is broken. The same applies to amplifier failures. This situation changes completely if channels are added or dropped along the link. The motivation for adding or dropping optical channels arises from the fact that optical filters can provide this function at much lower cost than optical to electrical conversion followed by switching in the electrical domain and reconversion to the optical domain (OEO). The lower part of Figure 6.3 shows an example for a meshed transparent photonic network using optical channel switching. Wavelength channels originating and terminating at different points in the network are depicted in different line styles. A fiber cut in the span marked by the dark gray arrow would interrupt the channel depicted with the dashed line. Deactivation of this channel in the link where it propagated together with the channels depicted with the solid and the dotted line may have an impact on the power stability of the remaining channels. The difficulty of suppressing optical power transients depends on the magnitude of power changes. Larger changes of the total power usually result in stronger fluctuations of surviving channel powers. In consequence, the challenge of protecting a small number of surviving channels against power transients grows with an increasing number of interrupted or added channels. The probability that a small number of surviving channels has to be protected against fast deactivation of a large number of channels is much higher in transparent photonic networks than in point-to-point links. In a point-to-point link, any component failure along the link, including fiber cuts, will interrupt all channels, leaving no surviving channels to protect. A component failure in the terminal will usually affect only a single channel. The reason failure scenarios with critical channel ratios occur more frequently in transparent photonic networks with optical add/drop multiplexer (OADM) nodes or optical crossconnects (OXC) can be understood by taking a look at typical configurations. Figure 6.4 shows an example of an OXC of degree 3 with local add/drop channels. For each fiber link, only one of the two directions is shown. The links on the left hand side and the one above the OXC carry channels toward the node, whereas the link on the right hand side carries channels away from the node. Local add/drop channels are depicted by transmitters (TX) and receivers (RX) below the node. Any fiber cut or component failure in a link leading toward the node has a potential of interrupting a large fraction of channels in a link leaving the node. At least the locally added channels will still be present in this case and have to be protected against power transients. The situation is most critical in a node with degree 2, i.e., only one incoming link and one outgoing. Any failure in the incoming link will usually interrupt a large number of channels passing through the node to the outgoing link, usually leaving only a small number of locally added channels. In summary, meshed transparent photonic networks are characterized by a multitude of failure scenarios, where several channels can be deactivated on
161
162
CHAPTER 6 Mastering power transients
RX
TX
RX
TX
OXC
FIGURE 6.4 Fiber cut in front of an OXC node
a sub-millisecond timescale and surviving channels have to be protected against transients. Resilience of the network requires suppression of power transients in remaining channels to avoid failure propagation. Today, fast channel deactivations due to component failures or fiber cuts are the most critical sources of power transients in photonic networks. The potential of the OXC nodes for wavelength switching and rerouting is so far not used for dynamic adaptation to changing traffic demandsdat least not on time scales smaller than a day. The main benefit from flexible and dynamic OXCs used today is easier network operation and maintenance. In future networks, the capabilities of OXCs may be leveraged also for dynamically adapting the network to changing traffic patterns. A control plane approach can enable switching of connections based on customer demand. This potentially increases the frequency of channel activations and deactivations in a given link. Dynamic switching of channels can be deployed only if the network is protected against channel power transients. In consequence, the performance of the physical layer has an important impact on possible types of services in future transparent photonic networks. Optical power transient suppression capabilities strongly influence the transparent reach, capacity, reliability, and flexibility potentials of these networks.
6.1.2 Remaining effects Electronic gain control has been deployed successfully to suppress channel power transients in static point-to-point links. It is considered state of the art and will not be discussed in further detail in this chapter. Instead, the following section will focus on
6.1 Introduction to power transients in optical networks
remaining effects. Even a perfect implementation of fast electronic gain control cannot cover all significant origins of power transients found in current transparent networks. Therefore, additional measures have to be taken to provide adequate transient suppression. Before these measures can be defined, it is important to understand why electronic gain control does not suffice any longer.
6.1.2.1 Gain control error due to gain spectrum ripple One reason constant gain control cannot completely suppress channel power transients is the non-flat gain spectra of optical amplifiers. Ideally, all channels going through a WDM amplifier should experience the same gain. The non-flat gain spectrum of erbium-doped fibers is usually flattened by optical filtering. As the required filter function is rather complex, manufacturable devices can only approximate the ideal filter function for a perfectly flat gain. In addition, the shape of the active fiber gain spectrum and the shape of the filter loss spectrum may exhibit different temperature dependences. As a consequence, the difference of the gain spectrum in dB and the filter loss spectrum in dB is not a flat straight line, resulting in a ripple of the amplifier gain spectrum. For well-designed amplifiers, gain variations of approximately 1 dB have to be expected across the C-band for an average gain of 30 dB. The monitor diodes of the electronic gain control measure total input power and total output power of the amplifier. Consequently, the gain control keeps the average gain of all active channels constant. Due to the ripple of the gain spectrum, individual channels experience different gain values. If channels with different gain values are activated or deactivated, the average gain will also change. In this case, the gain control will alter the output power of the channels after the switching event to restore the set average gain level to the target setpoint. This undesirable intervention of the gain control results in power fluctuations of initially present channels. Figure 6.5 tries to illustrate an example. The graph shows the power spectrum at the amplifier output for the case of four active channels. All channels are launched
P
λ
FIGURE 6.5 Gain control error due to a non-flat gain spectrum
163
164
CHAPTER 6 Mastering power transients
into the amplifier with equal power. Due to the non-flat gain spectrum, the two channels depicted with solid lines experience more gain and arrive at the amplifier output with more power than the two channels depicted with dashed lines. The horizontal dotted line shows the level of the average channel power, which is used by the gain controller to determine the average gain. Deactivation of the two channels depicted by dashed lines increases the average channel power at the amplifier output. The gain control will notice the corresponding change of the average gain and will reduce the pump power to restore the desired gain. The gain reduction decreases the average channel power and the power of the two surviving channels at the amplifier output. The power change due to the gain control error resulting from a non-flat gain spectrum can be as high as the gain ripple or even exceed it if the number of channels added or dropped is large and the number of unchanged channels is small. If several amplifiers with similar gain ripple spectra are cascaded, the power changes of channels in the worst case can accumulate constructively. In consequence, power fluctuations caused by gain control errors due to non-flat gain spectra can be as high as the gain ripple of an individual amplifier times the number of cascaded amplifiers. For a gain ripple of 1 dB, this leads to a power change of 10 dB for a cascade of 10 amplifiers. Such a strong power change usually cannot be tolerated by the system, and error-free operation of the affected channels is not maintained.
6.1.2.2 Gain variations due to spectral hole burning Even with a perfect gain equalization filter, changes of the gain spectrum can result in alterations of individual channel powers. Such changes of the gain spectrum can result from spectral hole burning (SHB) [82e85]. The laser line of the erbium ions in EDFAs is broadened mostly homogenously. For homogeneous broadening, the shape of the gain spectrum remains constant for constant average inversion regardless of the channel load. As different ion groups inside the active fiber exhibit slightly different spectral responses, some inhomogeneous broadening can be observed. In consequence, the shape of the gain spectrum can change depending on channel load. Figure 6.6 tries to illustrate the impact of spectral hole burning on channel powers. Both diagrams show power spectra at the output of a given EDFA. In the upper part, all channels are present. The lower part contains the spectrum after deactivation of a few channels in the short wavelength region. The output power spectrum with all channels includes some spectral hole burning, as the amplifier is operated in saturation. If channels are dropped, the ion groups with a stronger contribution to the gain in the respective wavelength range will experience less saturation. Consequently, the gain increases in this wavelength range. The remaining channels in this region experience more gain and achieve higher powers at the amplifier output. The power growth of the remaining channels due to SHB will usually not exceed a few tenth of a dB in one EDFA, with a stronger increase on the short wavelength side than on the long wavelength side due to the wavelength dependence of SHB.
6.1 Introduction to power transients in optical networks
P
λ
P
λ
FIGURE 6.6 Channel power changes due to spectral hole burning
However, the wavelength dependence of SHB is usually more deterministic than the gain ripple due to filter imperfections and will usually be very similar in all amplifiers of a cascade. Channels experiencing more gain in one amplifier after deactivation of neighboring channels will be exposed to the same effect in the following amplifiers. In a long cascade, the contributions of the individual EDFAs may add up to a few dB. As a consequence, channel power variations due to SHB (after a change in the channel load) have to be considered at least in long-haul systems.
6.1.2.3 Stimulated Raman scattering in the transmission fiber Another important source of channel power transients in long haul WDM transmission systems has nothing to do with erbium doped amplifiers. It results from a nonlinear effect in the transmission fiberdstimulated Raman scattering (SRS) [86e88]. An illustration of the concept is shown in Figure 6.7. A group of WDM channels with equal powers is launched into a transmission fiber section. Due to the nonlinear effect of SRS, they exchange energy while propagating through the fiber. Energy from channels on the short wavelength side is transferred to channels on the long wavelength side. This energy transfer results in a tilt of the channel power distribution at the output of the fiber. The strength of the impact of SRS and the magnitude of the induced tilt depend on a number of parameters. Due to the nonlinear nature of SRS, a higher launch power of the channels results in a larger tilt. The center wavelengths of the individual
165
166
CHAPTER 6 Mastering power transients
gR
λ
P
P
Transmission Fiber λ WDM signal
λ WDM signal
FIGURE 6.7 Tilt of the channel power distribution caused by stimulated Raman scattering
channels have an impact, too. The plot above the fiber section in Figure 6.7 depicts the shape of the spectrum of the Raman gain coefficient gR. The Raman gain increases up to a spacing of approximately 114 nm between the pump and the amplified channel for a pump wavelength of 1550 nm. Channels with a wider spacing inside this interval will experience a stronger interaction due to SRS. As a good approximation, the tilt in dB, i.e., the difference of the power levels of the shortest and the longest wavelength channels in dB, is proportional to the total launch power of the channels in mW and the frequency range in Hz corresponding to the wavelength range populated with channels. SRS is a very fast nonlinear effect, at least much faster than the timescales of channel power transients discussed in this chapter. As a consequence, the tilt changes instantaneously with fluctuations of the total power or changes of the width of the wavelength range populated with channels. Activation or deactivation of channels changes the total power and potentially the width of the wavelength range. After the switching event, the surviving channels will experience a different amount of gain or loss induced by SRS, resulting in a different power level at the fiber output. Figure 6.8 shows an example for channel power changes due to SRS. The diagram contains power spectra at the output of a standard single mode transmission fiber (SSMF) section with a length of 100 km. Before the switching event, 80 channels are launched into the input end of the fiber with a total power of 20 dBm. The individual launch power levels were chosen to achieve equal channel powers at the fiber output, depicted by the dark gray diamonds in the plot. Deactivation of half of the channels on the short wavelength side results in the output power spectrum depicted by the lighter gray squares. The remaining 40 channels contribute only half of the initial launch power and cover a much smaller wavelength range. As a consequence, the energy exchange between channels due to SRS reduces significantly compared with full channel load. The SRS-induced tilt of the channel power distribution was present for the configuration with 80 channels but not visible in the plot due to the launch power distribution at the fiber input,
6.1 Introduction to power transients in optical networks
-14.9 -15.0 80 ch.
Power in dBm
-15.1
40 ch. -15.2 -15.3 -15.4 -15.5 -15.6 -15.7 1525 1530
1535
1540
1545
1550
1555 1560
1565 1570
Wavelength in nm
FIGURE 6.8 Channel power distributions at the output of a transmission fiber
which was specially selected to achieve equal channel powers at the span output for operation with full channel load. The reduced impact of SRS causes a reduction of the gain experienced by the channels on the long wavelength side. The reduced SRS gain results in a decrease of the power level of the channel with the longest wavelength by more than 0.6 dB after deactivation of the channels on the short wavelength side. The power change of the shortest wavelength channel among the 40 channels, which is located close to the center of the band, is less than 0.3 dB. Even higher power changes can be observed for smaller numbers of surviving channels. Other transmission fiber types like non zero dispersion shifted fiber (NZDSF) will usually exhibit even stronger power changes for the same configuration due to the higher Raman gain coefficient. Dispersion compensating fibers (DCF) can also add a contribution to the tilt of the channel distribution in a span due to their very high Raman gain coefficients and small mode field areas. Compared with SSMF, the increase in Raman efficiency overcompensates the usually lower powers launched into these fibers. A reduction of a channel power of 0.6 dB can usually be tolerated by a WDM system. However, as in the case of spectral hole burning, the contributions from multiple elements in a cascade will superimpose constructively. Even if the link consists of fiber sections of different fiber types, SRS will always reduce the shortest wavelength channel power at the fiber output and increase the power of the channel with the longest. This leads to a power reduction of 6 dB in a link with 10 spans of SSMF for the configuration in the example. Such a strong reduction can usually not be tolerated by WDM system transponders. All three transient generating effects which remain if EDFAs are equipped with fast electronic gain control can individually result in power fluctuations exceeding
167
168
CHAPTER 6 Mastering power transients
the tolerance limits of typical WDM systems. The combined impact of these effects significantly impairs the performance and reliability of transparent photonic networks. Additional measures should be taken to further suppress channel power transients compared with the performance with fast electronic gain control alone.
6.2 OVERVIEW OF TRANSIENT SUPPRESSION CONCEPTS This section provides an overview of concepts for the suppression of channel power transients. Some of these concepts should be implemented in combination with fast electronic gain control and can be interpreted as an add-on [89e93]. Other concepts provide a complete solution for transient suppression and can replace constant gain control of EDFAs. An overview of all concepts is shown in Figure 6.9.
6.2.1 Linearized system The power transient suppression concepts can be divided into four groups. The first group contains concepts to linearize the system. Power transient problems arise because the power of a given channel depends on what happens in other channels. This coupling of channel powers results from nonlinear effects, either in the transmission fiber or optical amplifiers. The idea of a linearized system is to reduce the impact of nonlinear effects. In case of the first concept, a linear system, this is realized by using such low launch powers into the transmission fiber that channel interactions due to SRS can be neglected and by designing EDFAs which can be operated in the linear regime.
Linearized system
Replacement signal
Retilting element
Individual ch. power control
Linear system
One signal between OADMs
Controlled tilt filter
DEMUX plus VOA
Counterdir. Raman pumping
One signal per span
Controlled VOA plus EDFA dynamic gain tilt
DEMUX plus single channel EDFA
Bidirectional Raman pumping
Multiple signals gen. by controlled DFB lasers
Counterdir. Raman instead of EDFA pump upgrade
Multiple signals lasing EDFA ASE in front of filter ASE interleaved comb
single counterdir. Raman pumping of transmission fiber single counterdir. Raman pumping of DCF single codirectional Raman pump multi wavelength Raman pump
FIGURE 6.9 Overview of transient suppression concepts
Channel group equalizer filter
6.2 Overview of transient suppression concepts
The concept enables reduction of transients from any source until their impact can be completely neglected. In that sense, it provides a very powerful solution. However, the implementation does not come free. The decreased transmission fiber launch power significantly reduces the achievable transparent reach and span lengths, creating a need for additional amplifiers (and amplifier sites) for a given link length. EDFAs that can be operated in the linear regime with the necessary total output powers require expensive high power pump lasers and will probably consume more electrical power than EDFA designs used today. As can be seen from these comments, the potential benefits of a concept have to be weighted against difficulty of implementation and the impact on system cost. A detailed discussion of all concepts would require more space than this book chapter allows. In order to fit the comments to the available space, only the most important issues of the concepts will be mentioned in this section. One major drawback of the linear system approach is given by the reduced span length due to decreased transmission fiber launch powers. The second concept uses counterdirectional Raman pumping to achieve the same OSNR with reduced launch powers and enables keeping the span length constant. This reduces the cost for additional EDFAs and repeater sites for a given link length. On the other hand, adding Raman pumps is neither an inexpensive nor a low-power consumption solution. The third concept is to simplify the design of EDFAs for operation in the linear regime by adding bidirectional Raman pumping for further reduced EDFA output powers. With sufficient Raman pump power, EDFAs may be completely avoided. The last concept in this group does not go that far. In this case, codirectional Raman pumps are used only to avoid EDFA pump upgrades which are used in some modular systems to enable lower installation cost with a small number of channels.
6.2.2 Replacement signal In a system with coupling of channel powers due to nonlinear interaction, the change of the total power resulting from the activation or deactivation of channels leads to power changes of surviving channels. The idea of the concepts in the second group is to keep the total power constant using signals, which replace the contribution of deactivated channels. Such replacement signals will be activated only in case of unintentional loss of signal channels due to failures. Intentional channel activation and deactivation events are assumed to be carried out one at a time and sufficiently slow to enable electronic gain control to keep power fluctuations within acceptable limits As some transient generating effects do not only depend on total power but also on the channel wavelengths, ideal compensation would require a replacement signal for each lost channel. A lower cost solution is realized by the first concept. It uses a single replacement signal at a given wavelength to keep the total power constant. The approach will not be capable of completely suppressing power transients, but may offer a compromise between cost and sufficient reduction of power fluctuations. Also for cost reasons, the replacement signal is generated and added only once at the input of a link and removed in the node following the link.
169
170
CHAPTER 6 Mastering power transients
Due to wavelength dependences, the replacement signal may experience different gain and loss values than other channels in each span. Without precise knowledge of the individual differences per span, determining the right power of the replacement signal at the link input turns out to be quite difficult. An easier replacement signal power control is enabled by the second concept, which inserts a new signal at the input of each span. The required strength of the signal can be controlled by using information from the total power monitors at the EDFA ouputs of the electronic gain control. One replacement signal may not be sufficient to provide sufficient transient suppression. The following four concepts in the group use multiple replacement signals distributed across the channel wavelength band for better approximation of the initial spectral power distribution. They differ in the way the replacement signals are generated. The third concept in the replacement signal group uses controlled distributed feedback (DFB) lasers with sufficient output powers to generate the replacement signals. The next one avoids the cost for individual sources by combined generation of all signals in a lasing EDFA, i.e., an optical amplifier with laser oscillations at multiple different wavelengths. The last two concepts use amplified spontaneous emission (ASE) also generated by EDFAs as replacement signals.
6.2.3 Retilting element The concepts in the third group focus mainly on the compensation of a tilt of the channel power spectrum introduced by SRS. The first concept uses a controlled tilt filter to compensate for the tilt of the channel power distribution caused by SRS. A variable optical attenuator (VOA) inserted between two EDFA stages enables tilting of the gain spectrum by leveraging dynamic gain tilt (DGT). Such a tilt element is used in the second concept. The change of the average EDFA gain coming with different tilt settings has to be compensated by a second VOA. As discussed in the section on tilt of the channel power distribution due to SRS, the Raman gain increases with increasing difference between the pump and the signal wavelength. The third concept in the group uses this effect to implement a controlled tilt element by counterdirectional pumping of the transmission fiber. Tilt induced by SRS can be compensated by applying Raman gain, which also relies on SRS. Instead of pumping the transmission fiber, the fourth concept uses Raman gain in the DCF. The higher Raman coefficient of this fiber type enables achieving the same tilt values with lower pump powers. For easier implementation of the controller and faster response, the fiber is pumped codirectionally in case of the fifth concept. The sixth concept uses multiple Raman pumps with different wavelengths to enable individual control of the tilt and the average Raman gain. It offers the same two degrees of freedom as the concept with a controlled tilt filter in combination with a controlled VOA. In case of the Raman pump solution, control of the average gain and the tilt of the gain spectrum can be achieved by careful selection of the pump wavelengths and power distribution between pumps.
6.3 Compensation of power transients introduced by SRS
6.2.4 Individual channel power control Concepts for controlling the tilt of the channel power distribution enable compensation of the tilt introduced by SRS in the transmission fiber. However, the ability of tilt control to reduce more complex deformations of the channel power distribution as in case of gain control error due to gain ripple or SHB is rather limited. In the fourth group of transient suppression concepts, control of individual channel powers is used to compensate any deformation of the channel power distribution. The first concept in this category provides a straightforward implementation of individual channel power control. The channels are demultiplexed into separate paths to enable individual channel power control by VOAs. For cost efficiency, this feature together with the following multiplexing should be provided by a single integrated component. One major issue of the concept is the substantial insertion loss of the demultiplexer, VOA, and multiplexer. This insertion loss can be reduced by using controlled single channel EDFAs instead of VOAs, as proposed in the second concept. Strict implementation of individual channel power control requires access to each channel. As the wavelength dependence of all remaining effects introduces only gradual changes from one channel to its neighbor, small numbers of channels can be added to groups with the same processing of all channels inside a group. If the number of channels per group is kept small, the restriction in power distribution control capability has a small impact on transient suppression performance. The relaxed requirement to control the power of channel groups instead of individual channels enables less expensive implementations like the channel group equalizer filter used in the third concept.
6.3 COMPENSATION OF POWER TRANSIENTS INTRODUCED BY SRS Fast electronic gain control provides a decent suppression of channel power transients originating from EDFAs. The remaining effects from this transient source, gain control error due to gain ripple and gain shape variation due to SHB, can be addressed by appropriate EDFA design. Depending on the type of the gain flattening filter, stochastic variations of gain ripple shapes from one amplifier to the next in a cascade can be leveraged. With a small deterministic fraction of the gain ripple spectrum, power fluctuations due to gain control error will, to a large degree, average out from span to span. The impact from SHB can be reduced by changing the glass composition as well as the erbium doping concentration of the active fiber and by operating the amplifiers less deeply in saturation. One effect electronic gain control and more appropriate EDFA design definitively cannot address is the tilt of the channel power distribution due to SRS. If this effect causes a large contribution to overall channel power transients, other measures have to be taken. The following sections cover concepts to suppress channel power
171
172
CHAPTER 6 Mastering power transients
transients resulting from a tilt of the channel power distribution due to SRS. Results from numerical investigations are discussed as well as experimental verifications.
6.3.1 Numerical investigation of SRS tilt compensation A numerical simulation tool was used to analyze and compare the performance of different compensation concepts for channel power fluctuations resulting from SRS tilt [94]. Considerable reduction of the transient magnitude below acceptable limits can be demonstrated.
6.3.1.1 Description of the numerical model and simulations A major issue influencing the investigation of Raman transients are time scales. Simulation tools commonly used for analyzing WDM transmission systems usually consider two different time scales. Effects such as evolution of channel power levels and ASE along the link are usually modeled in the steady state while signal distortion by dispersion, nonlinear effects, etc., is modeled on a sub-nanosecond time scale. There are usually no solutions for analyzing effects such as Raman transients on a millisecond or microsecond time scale. Therefore, a dedicated simulation tool was designed and implemented, enabling the investigation of channel power envelope fluctuations on microsecond to millisecond time scales. The simulation tool is based on black box models to keep the computational effort and simulation times low. The tilt of the channel power evolution induced in the transmission fiber by SRS is calculated with the following expression, which was derived using the same assumptions and simplifications as described in [95]: DP ¼
10 gR 1 Dfch ½1 expðaS LÞ P0 ; DfR lnð10Þ Aeff aS
(6.1)
where DP denotes the power level difference of the shortest and longest wavelength channels in dB, gR the peak Raman coefficient in m/W for random polarizations, Aeff the effective mode field area in m2, aS the loss coefficient in 1/m, L the length of the fiber in m, Dfch the difference of the center frequencies of the shortest and longest wavelength channels in THz, DfR the frequency difference between the pump and the peak of the Raman gain spectrum, and P0 the total launch power in W (sum of the time averaged powers of the channels). Figure 6.10 shows a block diagram of the simulation setup. It corresponds to a link following an OADM or OXC and consisting of 22 spans. The signal sources generate 40 channels covering the wavelength range from 1529 nm to 1563 nm in the C-band with a channel spacing of 100 GHz. Individual channels or arbitrary channel groups can be switched on or off with selected transition times in the millisecond to microsecond range. The channel powers are boosted by linear amplifiers and launched into the transmission fiber with powers of 4 dBm per active channel, corresponding to a total launch power of 20 dBm for full channel load.
6.3 Compensation of power transients introduced by SRS
x 22
Sources
Linear amplifier
Transmission fiber
Fixed tilt filter
Linear amplifier
SRS G
G
FIGURE 6.10 Block diagram of the simulation setup
The 40 signals with 100 GHz spacing and launch powers of 4 dBm per signal were chosen to reduce the computational effort of the simulations. They represent channel pairs in a system with 80 channels spaced by 50 GHz with launch powers of 1 dBm per channelda more typical configuration for a terrestrial long-haul system. Due to the small wavelength difference of the channels in a pair, no noticeable difference will be observed with respect to SRS-induced tilt between switching of channel pairs in the real system and switching of signals in the simulation model. Six different transient compensation concepts were selected for the investigations. A graphical representation of the concepts is depicted in Figure 6.11. The first concept served as a reference-system without dynamic transient compensation. It consisted of static tilt filters compensating the tilt of each individual fiber section for full channel load. The second and third concept contained tilt filters with a log-linear wavelength dependence and adjustable slope. The filters were dynamically controlled in a closed loop using a signal provided by a tilt monitor. The filter in the second concept had a fixed rotation axis of the transmission spectrum in the center of the band, whereas a combination of the filter with a fast controlled variable optical attenuator enabled free selection of the rotation axis wavelength in case of the third concept. The idea behind the next two concepts was to split the total wavelength band into sub-bands and to adjust the total power of the individual subbands by one VOA for each sub-band. The VOAs were controlled individually in a closed loop with the aim of minimizing fluctuations of the surviving channels. The fourth concept used two sub-bands, whereas the channels were split into four sub-bands in case of the fifth concept. The sixth concept was also chosen as a reference. It can be considered an extension of the split and control approach, featuring demultiplexing down to individual channels and channel power adjustment for each individual channel by VOAs controlled in a closed loop.
6.3.1.2 Simulation results and discussion Several different topics were addressed during the investigation: the number and wavelengths of surviving channels, the location of compensators (after each span,
173
174
CHAPTER 6 Mastering power transients
(a)
(b)
(c)
Static tilt filter (reference)
Controlled tilt filter
Controlled tilt filter + VOA
(d)
Channel group (2) power controller
(e) Channel group (4) power controller
(f)
Individual channel power controller
FIGURE 6.11 Compensation concepts selected for the numerical investigation
after every third span, after every fifth span, etc.), the transition time of the power drop of switched channels, the controller speed of the compensator, the EDFA control concept and the impact of the transmission fiber type. Figure 6.12 provides an example for the time evolution of channel powers at the input of the link. The plot on the left hand side shows the power in mW of the individual channels versus time. Each channel is marked in a different shade of grey. Only two of them can be seen, because the majority of lines are hidden behind the two depicted ones. The spectra at different points in time before and after the first switching event as indicated by the arrows are shown on the right hand side of the figure. An example of a signal at the output of the link is shown in Figure 6.13. The change of the SRS-induced tilt due to the power decrease of the channels on the short wavelength side can be seen best in the diagrams on the right hand side of the figure.
6.3 Compensation of power transients introduced by SRS
-15
Channel spectrum at system input (t=0)
Power in dBm
-15.5 Channel power time evolution at system input
Output power in mW
0.025
0.02
-16
-16.5
-17 0.015 -17.5 1525 1530 1535 1540 1545 1550 1555 1560 1565
Wavelength in nm
0.01 -15 0.005
Channel spectrum at system input (t=0.2 t-stop)
-20
0
0.2
0.4
0.6
0.8
1
1.2
Time in ms
1.4
1.6
1.8
2
Power in dBm
-25 0
-30 -35 -40 -45 -50 -55 -60 1525 1530 1535 1540 1545 1550 1555 1560 1565
Wavelength in nm
FIGURE 6.12 Example for a signal at the input of the link: the time evolution is shown in the diagram on the left hand side; spectra are shown on the right hand side for points in time indicated by the arrows. The 20 channels on the short wavelength side are deactivated at t ¼ 0.2 ms and reactivated at t ¼ 1 ms.
In the plot on the left hand side showing the time evolution of the channels, the tilt change leads to a spreading of the lines with different shades of grey representing the individual channels. There is room only for presenting and discussing a small fraction of the results. Figure 6.14 shows an example for channel power time evolutions at the output of the link obtained with the six different control concepts. The results correspond to a configuration with deactivation of 20 channels on the short wavelength side at t ¼ 0.05 ms (the propagation time through the system is neglected) and a transition time of 0.1 ms. The channels were reactivated at t ¼ 0.55 ms. The link consisted of 22 spans of standard single mode fiber. Gain control of the EDFAs was assumed ideal, corresponding to linear operation. The control speed of the different control concepts was set to 0.1 ms. Gain tilt controllers were inserted on average after every fifth span (after spans 5, 10, 15, 20, and 22). The results corresponding to a static tilt filter depicted on the upper left hand side in part a) of the figure exhibit a strong tilt of the channel power distribution after the deactivation of channels. The static tilt filters were designed to compensate the SRSinduced tilt for full channel load. Hence, no spreading of the channel power lines can
175
CHAPTER 6 Mastering power transients
5 0
Power in dBm
-5
3
2.5
Output power in mW
-10 -15 -20 -25 -30
2 -35 1525
1530
1535
1540
1545
1550
1555
1560
1565
Wavelength in nm
1.5 5
1
Channel spectrum at system output (t=0.2 t-stop)
0 0.5
-5
0 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time in ms
Power in dBm
176
-10 -15 -20 -25 -30 -35 1525 1530 1535 1540 1545 1550 1555 1560 1565 Wavelength in nm
FIGURE 6.13 Example for a signal at the output of the link
be observed before the switching event. After the deactivation of channels, the amount of tilt due to SRS in the transmission fiber spans has changed. As the static tilt filter is not adapted to the new operating condition, it now compensates more than the actually occurring tilt. This leads to a remaining net tilt per span. The contributions from the individual spans accumulate, resulting in a power drop of the channel with the longest wavelength exceeding 6 dB at the output of the link. Errorfree operation of this channel and several shorter wavelength channels can probably not be maintained due to the strong power fluctuations. The controlled tilt filter concept uses a filter with an adjustable log-linear slope of the insertion loss spectrum to enable dynamic compensation of the SRS-induced tilt. Deriving a good signal for controlling this component in a feedback loop is not an easy task. The controller should not try to keep the power of channels constant that are deactivated. This requires identification of channels, which will survive a switching event. Good candidates are channels locally added in the node before the input of the link. If no traffic channels are added at a given location, special measurement signals may have to be inserted. The controller implemented in the simulation model monitors the shortest and longest wavelength channel of the surviving 20 channel group and tries to keep their power ratio constant. As can be deduced from the trace after deactivation of the 20 short wavelength channels, which is shown in part b) of Figure 6.14, the controller manages to
6.3 Compensation of power transients introduced by SRS
(a)
(d)
Static tilt filter Channel power time evolution at system output
3
Channel group (2) power control
4 3.5
Output power in mW
Output power in mW
2.5
2
1.5
1
0.5
3 2.5 2 1.5 1 0.5 0 0
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.1
0.2
0.3
Time in ms
(b)
(e)
Channel power time evolution at system output
0.7
0.8
0.9
1
Channel group (4) power control
3
Output power in mW
Output power in mW
0.6
Channel power time evolution at system output
3.5
2.5 2
1.5 1
0.5
2.5 2 1.5 1 0.5 0
0 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0
1
0.1
0.2
0.3
Time in ms
(c)
0.4
0.5
0.6
0.7
0.8
0.9
1
Time in ms
(f)
Controlled attenuator + tilt filter
Single channel power control
Channel power time evolution at system output
3
Channel power time evolution at system output
3
2.5
2.5
Output power in mW
Output power in mW
0.5
Time in ms
Controlled tilt filter
3
0.4
2 1.5 1 0.5
2 1.5 1.5 1 0.5
0 0
0.1
0.2
0.3
0.4
0.5
0.6
Time in ms
0.7
0.8
0.9
1
0 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time in ms
FIGURE 6.14 Comparison of channel power time evolutions at the output of the link obtained with different control concepts
177
178
CHAPTER 6 Mastering power transients
dynamically adjust to different amounts of SRS-induced tilt. The compensation results in a much smaller remaining tilt at the output of the link than in case of the static tilt filter. The small remaining tilt before the switching event can be attributed to a realistic dimensioning of the control loop with finite gain. More gain would reduce the static control error but in turn deteriorate the dynamic behavior. The large average power change of the surviving channels during the switching event results from a special characteristic of the SRS-induced tilt. The wavelength of the rotation axis of the log-linear tilt depends on the wavelengths of deactivated and surviving channels. As the tilt filter used for the compensation has a fixed rotation axis of the insertion loss spectrum, it can follow slope changes but not shifts of the rotation axis. Compensation of the shift requires a second degree of freedom. This additional flexibility is provided in the third compensation concept by combining the adjustable tilt filter with a wavelength independent VOA. Appropriate control of the two elements enables much better suppression of channel power fluctuations due to changes of the strength of SRS-induced tilt, as can be seen in Figure 6.14 c). In the column on the right hand side, the figure contains results from concepts trying to keep the power of channel groups constant. Controlling the power of a channel group approximates individual channel power control with less complexity and cost of implementation. If the number of channels inside the group is chosen to be sufficiently small, the differences of the SRS-induced gain or loss experienced by channels in the group as well as the resulting tilt can be neglected. For the results shown in part d) of Figure 6.14, a very coarse approximation to individual channel power control has been chosen. The total spectrum was split into only two groups, as indicated by the number in brackets. One group comprises the 20 channels on the short wavelength side, the other the 20 channels with the longer wavelengths. Due to the relatively large number of channels in the group, SRS induces amounts of channel power changes that are considerably different. The resulting tilt change after the deactivation of the short wavelength channels manifests in a spreading of the channel power time evolution lines. Better suppression of power transients can be obtained if the total channel number is split into four groups. The diagram depicted in part e) of Figure 6.14 still shows considerable channel power fluctuations. But it demonstrates the capability of the control concept to enable a compromise between the implementation effort and the achievable transient suppression. The best compensation of channel power fluctuations caused by SRS-induced tilt or any other mechanism should be expected from individual channel power control. Adjusting individual channel powers has the potential to stabilize the power of any surviving channel. However, the difficulty lies in the detection of time averaged power changes. Especially in case of amplitude shift keying (ASK) modulation formats, if the control time constant is chosen too small, the power stabilization starts to remove the modulation. If it is chosen to be too large, the controller will not be able to follow fast power changes induced by interactions between the channels.
6.3 Compensation of power transients introduced by SRS
A compromise has to be found between the detrimental effect of the remaining power fluctuations and the impact of the control on the modulation. Such an impact also has to be expected for phased shift keying (PSK) formats. The power envelope of propagating signals may differ from a constant value even in case of PSK due to chromatic dispersion (CD) or narrow bandwidth optical filtering. Transient suppression results achievable with individual channel power control are shown in the lower right corner of Figure 6.14 in part f). The controller time constant was chosen as 0.1 ms. This value coincides with the channel deactivation transition time. Due to the insufficient speed of the controller, undershoots and overshoots of the channels powers cannot be avoided. Much faster time constants would result in better transient suppression performance but probably cause bit errors for binary ASK signals with long sequence lengths. A comparison of results achieved with different compensation approaches reveals that the controlled tilt filter plus VOA concept seems best suited for the suppression of channel power transients caused by the SRS-induced tilt. Coping with these transients will be an important issue in future long-haul transparent photonic networks. The controlled tilt filter plus VOA concept offers a powerful approach to suppress these transients and its implementation is relatively simple and inexpensive compared with other approaches.
6.3.1.3 Transient propagation in transparent photonic networks An additional set of numerical simulations was carried out to investigate the propagation of SRS-induced transients in a transparent photonic network. If a fiber cut or component failure in a link carrying signals toward an OADM or OXC node endangers error-free operation of signals in a link directly following the node, this is hardly acceptable. Destabilization of the whole transparent network due to flooding with transients originating from a single failure definitely has to be avoided. The setup used to analyze the propagation of transients through a transparent network is depicted in Figure 6.15. It also uses 40 channels representing 80 channels in a real system to reduce computational effort. Many other system parameters such as launch powers, channel spacing, etc., were selected as in the simulations described in the previous section. In the following description, only the differences will be mentioned. A fully loaded link on the top left hand side carries all 40 channels in the C-band toward OADM node number 0 (OADM 0). In this node, half the channels are dropped and replaced by 20 locally generated channels in the long wavelength half of the spectrum (red band). The remaining 20 channels in the short wavelength half of the spectrum (blue band) are passing through the node and continue to propagate in the following link together with the added channels. The link following OADM 0 consists of 10 spans. Compensators for channel power transients due to SRS-induced tilt were inserted after each fifth span. At node OADM 1, the 20 channels in the blue band are dropped and replaced by 20 locally added channels. In this case, the red band channels originating from OADM 0 pass through the node and continue to propagate together with the locally added blue
179
180
CHAPTER 6 Mastering power transients
40 ch. C band
20 ch. red band
x5
tilt comp.
x5
tilt comp.
x5
tilt comp.
x5
tilt comp.
x5
tilt comp.
x5
tilt comp.
OADM 0 20 ch. blue band
OADM 1 20 ch. red band
OADM 2
20 ch. blue band
OADM 3
FIGURE 6.15 Setup for the investigation of the propagation of channel power transients in a transparent photonic network
band channels. The link following OADM 1 has the same layout as the one before it. OADM 2 again replaces the channels in the red band and so on. Due to this alternating wavelength band dropping of channels, each channel propagates through a maximum distance of 20 spansda realistic number for the current generation of long-haul systems. Without interactions between channels, the only implication of a fiber cut in the link before OADM 0 for the link following it would be the deactivation of the blue band channels. The red band channels originating from the node would propagate undisturbed toward OADM 1. As the blue band channels are replaced in this node, no changes of operating conditions occur in the network following OADM 1 after the fiber cut. When the blue band channels are deactivated, the energy transfer between channels induced by SRS results in a tilt change of the red band channels in the link following OADM 0. This change of the channel power distribution in the red band propagates through node OADM 1 and alters the strength of SRS in the following link. The changed strength of SRS in the link between OADM 1 and OADM 2 has an impact on the channel power distribution in the blue band. This causes changes of the blue band channel power distribution after OADM 2, resulting in a tilt change of the just added red band and so on. The interruption of channels before OADM 0 can result in ongoing transfer of power fluctuations between bands in the following links.
6.3 Compensation of power transients introduced by SRS
Simulations were carried out to investigate the propagation of SRS-induced channel power transients in a transparent photonic network without compensation. The results are depicted in Figure 6.16. In these simulations, realistic responses of EDFAs equipped with fast electronic gain control were included. The constant gain control manages to strongly reduce channel power fluctuations caused by changes of the total power, but it cannot compensate the SRS-induced tilt of the channel power distribution. The individual plots in the figure contain channel power time evolutions at different locations in the network. The diagram in the top left hand corner corresponds to the output of the link between OADM 0 and OADM 1, which is connected to the input of OADM 1. The fiber cut or component failure in the link before OADM 0 starts to decrease the power of the blue band channels at t ¼ 0.3 ms. Delay due to propagation of the channels has been neglected. As indicated by the lines with a steep slope, the power has dropped to negligible levels at t ¼ 0.5 ms. The impact of the deactivation of the blue band channels on the power of the red band channels can be deduced from the other lines in the plot. Just after the start of the power drop in the blue band, the channel powers in the red band start to spread due to the changing strength of SRS-induced tilt in the transmission fiber spans of the link. The following parallel downturn of the lines results from the interaction between SRS-induced tilt changes and EDFA transient response. At t ¼ 1.5 ms, the power levels in the red band have arrived at an intermediate steady state. Reactivation of the blue band channels causes further power fluctuations. The case of the reactivation of channels has been included in the simulations, because failure mechanisms other than fiber cuts may lead to intermittent interruption of channels. The diagram in part b) of Figure 6.16 shows channel power evolutions at the output of OADM 1. As the red band channels just pass through the node, all power fluctuations at the input of the node are transferred to the output. The horizontal lines in the diagram correspond to the blue band channels that are locally added in the node. These channels are launched with constant power levels into the input of the following link. The channel power evolutions at the output of the link, which corresponds to the input of OADM 2, can be seen in part c) of Figure 6.16. Channel power fluctuations in the red band have changed only slightly compared with the input of the link. However, a drastic change of the power evolutions of the blue band channels can be observed. The interplay between change of the SRSinduced tilt and EDFA transient response results in considerable power fluctuations of the blue band channels at the output of the link. As the red band channels are dropped in OADM 2, no further propagation of transients in these channels occurs. However, the channels in the blue band take their transients inherited from the other band with them through the node. The diagram in Figure 6.16 d) shows the situation at the output of OADM 2. It resembles the one at the output of OADM 1. Horizontal lines correspond to the channels locally added in the node, in this case in the red band. The lines with power fluctuations describe the transients in the blue band channels transferred from the input of the node to its output.
181
CHAPTER 6 Mastering power transients
(a)
(b)
Before OADM No. 1
3
2.5
2.5
Channel power in mW
Channel power in mW
3
2 1.5 1 0.5
After OADM No. 1
SP-Monitor: Channel power time evolution after OADM No. 1
SP-Monitor: Channel power time evolution before OADM No. 1
2 1.5 1 0.5 0
0 0
0.5
1
1.5
2
2.5
0
3
0.5
1
(c)
(d)
Before OADM No. 2
2
2.5
3
After OADM No. 2
SP-Monitor: Channel power time evolution before OADM No. 2
SP-Monitor: Channel power time evolution after OADM No. 2
3.5
3.5 3
Channel power in mW
3
Channel power in mW
1.5
Time in ms
Time in ms
2.5 2 1.5 1 0.5
2.5 2 1.5 1 0.5 0
0 0
0.5
1
1.5
2
2.5
0
3
0.5
1
(e)
1.5
2
2.5
3
Time in ms
Time in ms
(f)
Before OADM No. 3
After OADM No. 3
SP-Monitor: Channel power time evolution after OADM No. 3
SP-Monitor: Channel power time evolution before OADM No. 3
3
Channel power in mW
3
Channel power in mW
182
2.5 2 1.5 1 0.5
2.5 2 1.5 1 0.5 0
0 0
0.5
1
1.5
Time in ms
2
2.5
3
0
0.5
1
1.5
2
2.5
3
Time in ms
FIGURE 6.16 Time evolution of channel power fluctuations due to SRS at several locations in a transparent photonic network
6.3 Compensation of power transients introduced by SRS
The process with the transfer of transients from one band to the other and passing of newly infected channels through a node continues in the following parts of the network, resulting in still significant channel power fluctuations at the output of OADM 3. Fortunately, the strength of transients decreases from one node to the next. However, the weak damping of the fluctuations leads to a propagation of transients far into the network. The configuration chosen for the simulations corresponds to a linear section of a transparent photonic network without splitting of the paths of channels. In meshed networks, more branching of paths has to be expected. This will result in a more complex propagation of transients in multiple directions. Even power fluctuations running around loops can occur. Faster damping of fluctuations seems highly desirable in order to avoid destabilization of larger parts of a transparent network caused by a failure event in a single link. As the controlled tilt filter plus VOA approach provided good SRS-induced transient suppression in a single link, it should also help avoid spreading of transients in a transparent network. This assumption was tested using numerical simulations with the same setup but now including active tilt control elements. The results are shown in Figure 6.17. The order of the plots is the same as in Figure 6.16. The channel power evolution at the output of the first link, which corresponds to the input of OADM 1, can be found in part a) of Figure 6.17. Some power fluctuations in the red band channels caused by SRS-induced tilt changes following from the deactivation of the blue band channels can be observed, but they are much smaller than in the case without active tilt control. The introduction of a controlled tilt filter plus VOA compensator has reduced the transient magnitude at the output of OADM 1 below the level occurring at the output of OADM 3 in the case without active tilt compensation. Much faster damping of fluctuations can be achieved by the deployed compensation approach. This demonstrates the good potential of the controlled tilt filter plus VOA concept to suppress channel power transients in transparent photonic networks resulting from SRS-induced tilt.
6.3.2 Experimental investigation of SRS tilt compensation In the previous section, the performance of different approaches to compensating channel power transients resulting from the dynamic tilt of the channel power distribution introduced by SRS was investigated using numerical simulation tools. The numerical study was helpful to compare the characteristics of the different concepts and to identify the best performing ones. The following section describes results from experiments to verify that the good performance predicted by the numerical simulations can also be achieved in practical implementations [96]. Figure 6.18 shows a block diagram of the experimental setup. It corresponds to an example for a transmission fiber span following an OADM or OXC with 50% express channels. The channel plan resembles the one used for the simulations in the sense that it contains only 40 channels for reduced effort but channel powers were doubled to represent an 80 channel system. Output signals from 20 unmodulated
183
CHAPTER 6 Mastering power transients
(a)
(b)
Before OADM No. 1
SP-Monitor: Channel power time evolution before OADM No. 1
Channel power in mW
Channel power in mW
2.5 2 1.5 1 0.5
After OADM No. 1
SP-Monitor: Channel power time evolution after OADM No. 1
2.5 2 1.5 1 0.5 0
0 0
0.5
1
1.5
2
2.5
0
3
0.5
1
(c)
1.5
2
2.5
3
Time in ms
Time in ms
(d)
Before OADM No. 2
SP-Monitor: Channel power time evolution before OADM No. 2
After OADM No. 2
SP-Monitor: Channel power time evolution after OADM No. 2 2.5
Channel power in mW
Channel power in mW
2.5 2 1.5 1 0.5 0
2 1.5 1 0.5 0
0
0.5
1
1.5
2
2.5
3
0
0.5
1
Time in ms
(e)
1.5
2
2.5
3
Time in ms
(f)
Before OADM No. 3
After OADM No. 3
SP-Monitor: Channel power time evolution after OADM No. 3
SP-Monitor: Channel power time evolution before OADM No. 3
2.5
2.5
Channel power in mW
Channel power in mW
184
2 1.5 1 0.5 0
2 1.5 1 0.5 0
0
0.5
1
1.5
Time in ms
2
2.5
3
0
0.5
1
1.5
2
2.5
3
Time in ms
FIGURE 6.17 Suppression of channel power transient propagation by controlled tilt filters; the diagrams show the time evolution of channel powers at different locations in the network.
6.3 Compensation of power transients introduced by SRS
20 ch. blue band Fiber Filter
OSA 90:10 blue band Osci
20 ch. red band 1559.8 nm
1560.6 nm
1548.5 nm
FIGURE 6.18 Block diagram of the experimental setup
DFB lasers in the wavelength range from 1548.5 nm to 1563.8 nm with 100 GHz spacing were combined using a multiplexer filter and then amplified. These channels represent the red band channels locally added in the OADM. The express channels in the blue band were also generated by 20 unmodulated DFB lasers. They were placed in the wavelength region from 1529.1 nm to 1544.9 nm with 100 GHz spacing. A fast adjustable VOA was inserted after the multiplexer to emulate the fiber cut in front of the OADM that interrupts all express channels simultaneously. Amplification of the express channels was provided by a separate amplifier to avoid an impact of the EDFA transient response on the surviving channel power levels. The two wavelength bands were combined by a wavelength selective band splitter filter before launching them into the fiber span. The fiber type was DCF with a length of 16 km, which was selected instead of a real transmission fiber to obtain a stronger SRSinduced tilt owing to the higher Raman efficiency. The launch power per channel was þ5 dBm. This corresponds to a total launch power of 21 dBm. The control element of the Raman transient compensator was connected to the output of the fiber span. It consisted of a LiNbO3-based filter that provides an insertion loss spectrum with variable slope as well as adjustable wavelength independent attenuation. A tap coupler at the output of the filter is used to branch off a small amount of the total power from the signal path and directs it toward the monitoring function of the control loop. After demultiplexing, two signals corresponding to the shortest and the longest wavelength channels in the red band were converted into the electrical domain by photo diodes and fed into the controller. The output signals of the
185
CHAPTER 6 Mastering power transients
controller had to be boosted by electrical amplifiers with high voltage outputs before they were sent into the control element. The remaining elements in the setup were used to characterize the performance of the transient compensator. A band splitter filter coupled the switched blue band signals into a photo diode connected to an oscilloscope to monitor the falling slope during channel deactivation and to trigger the oscilloscope. Another photo diode connected to the oscilloscope enabled monitoring of the power of the channel with the second longest wavelength in the red band. An optical spectrum analyzer (OSA) connected to the other output of the band splitter filter helped monitor the distribution of channel powers in the red band. In a first step, the static behavior of the transient compensator was analyzed using the OSA. The power level differences of the channels in the red band for the operating conditions before and after deactivation of the blue band channels are depicted in Figure 6.19. The lighter gray diamonds depict the channel powers without compensation. The drive signals of the filter were kept constant to maintain a flat attenuation spectrum. Power level changes up to 1 dB could be observed for the channel with the longest wavelength. Activation of the filter control loop resulted in the spectrum described by the darker gray squares in the diagram. The power level variation of the channel with the longest wavelength could be reduced below 0.2 dB. The deviation of the spectra from a straight line was most likely caused by the polarization dependence of SRS. Polarization scrambling was intentionally not used, as it will not be present in a real system and could have masked some relevant effects. 0.4
Control on
0.2
Power level difference in dB
186
Control off
0.0 -0.2 -0.4 -0.6 -0.8 -1.0 -1.2 -1.4 1545
1550
1555
1560
1565
Wavelength in nm
FIGURE 6.19 Spectrum of the channel power differences before and after the switching event with and without active control
6.3 Compensation of power transients introduced by SRS
10 µs 1.1 control on control off
Voltage in V
1.0
0.9
0.8
0.7
0.6 -50
-40
-30
-20
-10
0
10
20
30
40
50
Time in µs
FIGURE 6.20 Time evolution of the power of the channel at the wavelength of 1559.8 nm with and without tilt compensation
The dynamic performance of the transient compensator and the impact of controller speed were analyzed using the oscilloscope. Figure 6.20 shows the time evolution of the power of the channel with the second longest wavelength before, during, and after deactivation of the blue band channels. The transition time of the switching event was chosen as 10 ms in this case, which was the highest speed that could be obtained using the fast VOA. Similar transition times can be observed in real networks, if a cable is pulled out of the ground and the fiber breaks at the location of a fusion splice. Without compensation, deactivation of the blue band channels results in a considerable reduction of the channel powers on the long wavelength side in the red band. The transition time resembles one of the deactivated channels, demonstrating that the change of SRS-induced tilt can follow even very fast power level changes in case of codirectional propagation of channels. The power fluctuation amplitude of the channel with the second longest wavelength could be reduced considerably by activating the tilt filter control loop. As in the static case, the power level change could be reduced below 0.2 dB. This demonstrates the capability of the control loop to follow transients that can be expected in a real network. The importance of a sufficiently high controller speed can be concluded from Figure 6.21. It depicts results from the same measurement as described in the previous figure with the only difference that the controller speed was decreased intentionally from the original approximately 2 ms to approximately 40 ms. With this controller speed below the channel deactivation transition time of 10 ms, significant undershoots of the channel power can be observed. As control theory predicts, the
187
CHAPTER 6 Mastering power transients
10 µs, slow controller 1.1
1.0
Voltage in V
188
0.9
0.8
control on
0.7
control off 0.6 -50
-40
-30
-20
-10
0
10
20
30
40
50
Time in µs
FIGURE 6.21 Time evolution of the power of the channel at 1559.8 nm with insufficient controller speed
speed of the compensator has to be selected to be higher than channel power transition times to achieve good suppression of channel power fluctuations. The experiment was done with a single fiber span to demonstrate the performance of the compensation concept. In multispan transmission systems, the SRSinduced tilt in dB increases linearly with the number of spans. Depending on the contribution per span and the tolerable OSNR penalties due to undershooting channel powers or nonlinear penalties due to overshooting channel powers, the compensation of transients should be performed either after each span or after a small number of spans. This follows from the results of the numerical simulations described in the preceding section.
6.4 COMPENSATION OF TRANSIENTS FROM ALL SOURCES The numerical and experimental investigations have shown that decent suppression of channel power transients resulting from changing SRS-induced tilt of the channel power distribution can be achieved using compensators with controlled tilt filters and VOAs. This compensation approach addresses power fluctuations caused by a nonlinear effect in the transmission fiber. It should be used in combination with fast electronic gain control which primarily addresses power level fluctuations generated inside the EDFA.
6.4.1 Individual channel power control If transients from EDFAs are suppressed by constant gain control and power fluctuations from changes of SRS-induced tilt in the transmission fiber by the controlled
6.4 Compensation of transients from all sources
tilt filter plus VOA concept, SHB will contribute the dominating part of the remaining channel power fluctuations in many cases. Due to the complexity of the gain shape changes caused by this effect, compensation by inverse filter approaches turns out to be very challenging. The best suited channel power distribution equalizing concept in this case seems to be individual channel power control. However, as discussed before, the excellent channel power stabilization potential of this approach is offset by several major drawbacks. One of them consists of the large complexity of implementation. Up to 80 or even more channels have to be processed simultaneously, creating challenges concerning the realization of optical components as well as power monitoring and electrical signal processing functions. In order to keep requirements for the channel power controller more moderate, fast electronic gain control of EDFAs should not be omitted but rather complemented with individual channel power control. Due to cost considerations, replacing the need for the controlled tilt filter plus VOA compensator would be highly desirable, when individual channel power stabilization is deployed. Without the tilt filter, the channel power control has to cope with the potentially very fast power fluctuations resulting from SRS-induced tilt changes. This probably exceeds achievable controller speeds. Besides implementation issues of the simultaneous processing of 80 channels, the already mentioned impact of fast power stabilization on the signal modulation limits the realizable control speed. The maximum tolerable control speed from a signal modulation perspective will probably be smaller than the minimum speed required for sufficient suppression of the transients generated by fast changes of the SRS-induced tilt. Last but not least, the signal bandwidth reduction introduced by channel demultiplexing and multiplexing filters constitutes a major disadvantage of the individual channel power control concept. The current trend concerning system capacity enhancement relies on more bandwidth efficiency for increased capacity per fiber. Reducing the available bandwidth by wider gaps between channels due to filtering required for individual channel power control severely complicates the currently targeted realization of bitrates of 100 Gbit/s per channel with channel spacing of 50 GHz. In summary, individual channel power control has a high potential for the suppression of power transients caused by SHB-induced gain shape changes and even control errors due to gain ripple. These advantages are offset by numerous drawbacks. Some of them may be alleviated by clever design, for example if the fast channel power stabilization can be implemented inside OADM or OXC nodes where access to individual channels has to be possible anyway.
6.4.2 Fill signals Due to the major drawbacks, the individual channel power control concept does not provide the ultimate solution to suppress transients from all sources occurring in a WDM transmission system. Letting channel interactions resulting from nonlinear effects alter the power envelope of channels first and trying to remove the
189
190
CHAPTER 6 Mastering power transients
accumulated power changes later on by controlled gain or loss elements may not be the best approach anyway. The detrimental effect of OSNR reduction as well as nonlinear effects will be more severe the longer they are allowed to accumulate before compensation is done. A better solution could be provided by keeping the impact of nonlinear effects on the power of surviving channels constant to avoid any channel power changes in the first place. This can be achieved by fill signals. They can either replace the power of deactivated channels or be removed for the activation of new signal channels. Nearly ideal constant strength of nonlinear effects and static operation of a link is achieved if fill and signal channels are seamlessly exchanged, resulting in a constant power envelope spectrum across the whole wavelength band used by a system. Such an approach comes with large first installation cost, as all the channels have to be present on day one. The optical amplifiers have to be equipped to process the full channel load right after the activation of the link, even if only a small number of them are signal channels. As a sudden and unplanned activation of many channels usually does not occur in a link, starting link operation with full channel load is not necessary. The power in the fill signals can be kept much smaller than the power in the signal channels in the steady state. This significantly reduces the requirements on the available total fill signal power if only a few signal channels are present after installation of the link. The requirements on the number of fill signal wavelengths can be relaxed, too. In the ideal case, each operating signal channel is complemented by a fill signal with the same wavelength in standby mode, ready to replace the modulated signal and keep the time averaged channel power constant, if the power of the data carrying signal is reduced for any reason. Due to the rather weak wavelength dependence of the nonlinear effects relevant for channel power transients on scales corresponding to a channel spacing, the strength of these nonlinear effects will not change significantly when a small group of adjacent signal channels is replaced by a single fill signal. The fill signal has to provide the same total power as the group at a wavelength located near the center of the group. One major advantage of the fill signal approach lies in its scalability, which takes advantage of these relaxed requirements. In a short reach link with small potential for transient generation, a few fill signals at a selected number of wavelengths will be enough to sufficiently suppress signal channel power level fluctuations. Longer links can be equipped with more fill signals at additional wavelengths to keep the magnitude of channel power fluctuations below the same limit despite the larger impact of nonlinear effects. A relatively small number of three fill signals located at appropriate wavelengths alone can provide decent transient suppression in an 80 channel full C-band link with a length up to 1000 km. The replacement of dropped signal channels by appropriately increasing the power of fill signals keeps the total power constant, alleviating the need for fast electronic gain control. If the distribution of power across the band is kept approximately constant, too, the fill signals can avoid changes of the SRS-induced tilt. Only a strong impact of gain shape changes due to SHB may require a need for a larger number of fill signals. A placement of more
6.4 Compensation of transients from all sources
dense signals could be required especially in the short wavelength range around 1535 nm of the EDFA gain spectrum. In this region, holes are deeper and exhibit a narrower width than on the long wavelength side. Finding the right wavelengths for fill signals is not a trivial task. Placing them in gaps between signal channels may be possible in the center of the band if some slots on the ITU grid are left unused for easier multiplexer and demultiplexer filter design. Other potentially free wavelength regions are next to the WDM channel band edges. However, EDFA gain usually drops fast outside the range allocated by signal channels. Fill signals in these locations may not experience sufficient gain to maintain required power levels along a cascade of multiple spans. Inside the WDM channel bands, there are usually no unused wavelength regions left where fill signals could be placed. This applies especially to high spectral density systems with bitrates of 40 Gbit/s per channel and above and channel spacing of 50 GHz. Fill signals replacing a group of channels that are located inside the channel wavelength band can be activated only if the signal channel at the wavelength of a fill signal is lost anyway. Otherwise, the strong fill signal will interfere with the signal channel using the same wavelength and prohibit error-free operation. As the fill signal has to be activated as soon as any channel in the group loses power, its wavelength cannot be used for a channel that has to be protected against transients. Consequently, implementation of a fill signal approach for transient suppression usually reduces the number of possible signal channels in a given band. Furthermore, care has to be taken that the high power of the fill signals at wavelengths close to signal channels does not induce detrimental nonlinear effects such as four wave mixing (FWM). The fill signal itself has to be protected against stimulated Brillouin scattering (SBS). Otherwise, power fluctuations of the fill signal induced by SBS can be transferred to signal channels. This would convert the measure implemented for transient suppression into a source of channel power transients.
6.4.3 Linearized system If the fill signal approach does not provide sufficient transient suppression or cannot be implemented due to other reasons, the only remaining concept with a good capability to address all sources of transients is given by the linearized system. Without deployment of distributed Raman amplification, reducing the launch power into the transmission fiber will decrease the OSNR and result in an unacceptable reduction of span lengths and system reach. Adding the required Raman pumps comes with a significant cost increase. However, this measure may be necessary due to other reasons anyway. Higher level modulation formats currently considered for increased spectral efficiency or realization of bitrates of 100 Gbit/s per channel tend to be less tolerant than binary formats towards nonlinear effects. This applies to differential quadrature phase shift keying in combination with polarization multiplexing (PM-DQPSK) as well as orthogonal frequency division multiplexing (OFDM) and quadrature amplitude modulation (QAM). Fiber launch power levels for these formats have to be reduced below currently used levels to limit the impact of nonlinear effects.
191
192
CHAPTER 6 Mastering power transients
Achieving the same span lengths above 80 km and transparent reaches beyond 1000 km known from current long-haul system generations may require deployment of distributed Raman amplification in the transmission fiber. This measure could help with the suppression of channel power transients. However, experts will strive to enable higher launch powers even with multiple bits per symbol modulation formats, for example by deploying equalization or compensation of the impact of nonlinear effects on the signal modulation. This brings back the channel power transient issues and the need to cope with them.
6.5 SUMMARY AND CONCLUSIONS Channel power transients constitute a major limitation for the reach and reliability of long-haul transparent photonic networks. They have to be considered in the design of current and future optical data transport networks. Fast electronic gain control provides a powerful approach to improve the transient response of EDFAs. However, remaining effects such as control error due to gain spectrum ripple and gain spectrum shape alterations caused by SHB can lead to unacceptably strong power fluctuations in long-reach systems with many spans. Moreover, an effect not originating from EDFAs but the energy transfer between channels in the transmission fiber due to SRS that cannot be compensated by constant gain control contributes its own part to the magnitude of channel power fluctuations. A controlled tilt filter in combination with an adjustable VOA provides a powerful approach to compensate channel power fluctuations resulting from the changes in SRS-induced tilt. This was demonstrated in studies with numerical simulations tools as well as proof of concept experiments. Implementation of this approach in combination with fast electronic gain control significantly improves transient suppression. However, the remaining effects such as gain shape changes due to SHB and gain control error caused by gain ripple can still result in unacceptably strong channel power fluctuations. Further concepts like individual channel power control, fill signals, and linearization of the system have been proposed. They potentially enable strong suppression of transients resulting from any source. However, drawbacks and implementation issues keep all the described concepts from serving as the ultimate solution. Suppression of power transients cannot be handled as an isolated task. It has to be tailored to the characteristics of the transmission system or network. Implementation of channel power transient suppression is necessary to guarantee the stability and resilience of current transparent photonic networks in case of failure scenarios interrupting multiple channels. Consequences of fiber cuts or component failures have to be restricted to the channels transported in the affected link. For stable network operation, transients have to be kept from propagating beyond the following node and spreading further. Transient suppression requirements will probably be aggravated in future photonic networks. Flexible dynamic adaptation of the network to changing traffic
References
patterns or establishing of connections triggered by customer demand will result in frequent activation and deactivation of channels, in addition to changes in channel loading resulting from failure events. Currently known approaches may be sufficient to address these new requirements. However, further research is highly desirable to determine which concept is best suited for a given configuration and to continue the search for more powerful channel power transient suppression concepts.
ACRONYMS ASE ASK BER CD DCF DFB DGT DQPSK EDFA FEC FWM NZDSF OADM OEO OFDM OSA OSNR OXC PM QAM PSK RX SBS SDH SHB SONET SRS SSMF TX VOA WDM
Amplified spontaneous emission Amplitude shift keying Bit error ratio Chromatic dispersion Dispersion compensating fibers Distributed feedback Dynamic gain tilt Differential quadrature phase shift keying Erbium-doped fiber amplifiers Forward error correction Four wave mixing Non zero dispersion shifted fiber Optical add/drop multiplexer Optincal-to-electrical-to-optical conversion Orthogonal frequency division multiplexing Optical spectrum analyzer Optical signal to noise ratio Optical crossconnects Polarization multiplexing Quadrature amplitude modulation Phased shift keying Receivers Stimulated Brillouin scattering Synchronous digital hierarchy Spectral hole burning Synchronous optical network Stimulated Raman scattering Standard single mode transmission fiber Transmitters Variable optical attenuator Wavelength division multiplexing
193
194
CHAPTER 6 Mastering power transients
References [1] A. Morea, J. Poirrier. A critical analysis of the possible cost savings of translucent networks. Proceedings of the 5th International Workshop on Design of Reliable Communication Networks (DRCN 2005), October 16e19, 2005. [2] A.A.M. Saleh, J.M. Simmons, Evolution Toward the Next-Generation Core Optical Network, IEEE J. Lightwave Technol. 24 (9) (2006) 3303e3321. [3] R.S. Tucker, R. Parthiban, J. Baliga, K. Hinton, R.W.A. Ayre, W.V. Sorin, Evolution of WDM Optical IP Networks: A Cost and Energy Perspective, IEEE J. Lightwave Technol. 27 (3) (2009) 243e252. [4] C.R. Giles, E. Desurvire, J.R. Simpson, Transient gain and cross talk in erbium-doped fiber amplifiers, Optics Letters 14 (16) (1989) 880e882. [5] E. Desurvire, Analysis of transient gain saturation and recovery in erbium-doped fiber amplifiers, IEEE Photon. Technol. Lett. 1 (8) (1989) 196e199. [6] E. Desurvire, C.R. Giles, J.R. Simpson, Gain dynamics of erbium-doped fiber amplifiers, SPIE 1171, Fiber Laser Sources and Amplifiers (1989) (September 6e8) 103e117. [7] E. Desurvire, C.R. Giles, J.R. Simpson, Gain saturation effects in high-speed, multichannel erbium-doped fiber amplifiers at l ¼ 1.53 mm, IEEE J. Lightwave Technol. 7 (12) (1989) 2095e2104. [8] A.D. Ellis, R.M. Percival, A. Lord, W.A. Stallard, Automatic gain control in cascaded erbium-doped fiber amplifier systems, Electron. Lett. 27 (3) (1991) 193e195. [9] M. Zirngibl, Gain control in erbium-doped fibre amplifiers by an all-optical feedback loop, Electron. Lett. 27 (7) (1991) 560e561. [10] H. Okamura, Automatic optical-loss compensation with Er-doped fibre amplifier, Electron. Lett. 27 (23) (1991) 2155e2156. [11] E. Desurvire, M. Zirngibl, H.M. Presby, D. DiGiovanni, Dynamic Gain Compensation in Saturated Erbium-Doped Fiber Amplifiers, IEEE Photon. Technol. Lett. 3 (5) (1991) 453e455. [12] H. Okamura, Automatic optical loss compensation with erbium-doped fiber amplifier, IEEE J. Lightwave Technol. 10 (8) (1992) 1110e1116. [13] E. Delevaque, T. Georges, J.F. Bayon, M. Monerie, P. Niay, P. Bernage, Gain control in erbium-doped fibre amplifiers by lasing at 1480 nm with photoinduced Bragg gratings written on fibre ends, Electron. Lett. 29 (12) (1993) 1112e1114. [14] K. Motoshima, L.M. Leba, D.N. Chen, M.M. Downs, T. Li, E. Desurvire, Dynamic Compensation of Transient Gain Saturation in Erbium-Doped Fiber Amplifiers by Pump Feedback Control, IEEE Photon. Technol. Lett. 5 (12) (1993) 1423e1426. [15] K. Motoshima, K. Shimizu, K. Takano, T. Mizuochi, T. Kitayama. EDFA with dynamic gain compensation for multiwavelength transmission systems. Conference on Optical Fiber Communication (OFC 1994), San Jose, CA, USA, February 20e25, paper ThC4, pp. 191e192. [16] J.F. Massicott, S.D. Willson, R. Wyatt, J.R. Armitage, R. Kashyap, D. Williams, et al., 1480nm pumped erbium doped fibre amplifier with all optical automatic gain control, Electron. Lett. 30 (12) (1994) 962e964. [17] K.Y. Ko, M.S. Demokan, H.Y. Tam, Transient analysis of erbium-doped fiber amplifiers. IEEE Photon. Technol. Lett. 6 (12) (1994) 1436e1438.
References
[18] M. Fake, T.J. Simmons, J. Massicott, R. Wyatt. Optically stabilized EDFA for in-band WDM systems. Conference on Optical Fiber Communication (OFC 1995), San Diego, CA, USA, 1995, February 26eMarch 3, paper TuP3, pp. 79e80. [19] A. Yu, M.J. O’Mahony, Properties of gain controlled erbium doped fibre amplifiers by lasing, Electron. Lett. 31 (16) (1995) 1348e1349. [20] F. Shehadeh, R.S. Vodhanel, C. Gibbons, M. Ali. Comparison of gain control techniques to stabilize EDFAs for WDM networks. Conference on Optical Fiber Communication (OFC 1996), San Jose, CA, USA, 1996, February 25eMarch 1, paper WM8, pp. 190e191. [21] J.L. Zyskind, Y. Sun, A.K. Srivastava, J.W. Sulhoff, A.J. Lucero, C. Wolf et al. Fast power transients in optically amplified multiwavelength optical networks. Conference on Optical Fiber Communication (OFC 1996), San Jose, CA, USA, 1996, February 25eMarch 1, paper PD31, pp. 451e454. [22] J. Massicott, C. Lebre, R. Wyatt, R. Kashyap, D. Williams. Asymmetric Control Laser Cavity Design for Low Noise Operation of an All-Optical Gain Controlled ErbiumDoped Fibre Amplifier. Conference on Optical Amplifiers and their Applications, Monterey, CA, USA, 1996, July 11e13, paper FB2, pp. 77e80. [23] A.K. Srivastava, Y. Sun, J.L. Zyskind, J.W. Sulhoff, C. Wolf, R.W. Tkach. Fast gain control in an erbium-doped fiber amplifier. Conference on Optical Amplifiers and their Applications, Monterey, CA, USA, 1996, July 11e13, paper PDP 4, pp. 24e27. [24] J.L. Zyskind, A.K. Srivastava, Y. Sun, J.C. Ellson, G.W. Newsome, R.W. Tkach, et al. Fast link control protection for surviving channels in multiwavelength optical networks. 22nd European Conference on Optical Communication (ECOC 1996), Oslo, Norway, 1996, September 15e19, 5, 49e52. [25] J. Chung, S.Y. Kim, C.J. Chae, All-optical gain-clamped EDFAs with different feedback wavelengths for use in multiwavelength optical networks, Electron. Lett. 32 (23) (1996) 2159e2161. [26] Y. Sun, A.K. Srivastava, J.L. Zyskind, J.W. Sulhoff, J. Wolf, R.W. Tkach, Fast power transients in WDM optical networks with cascaded EDFA’s, Electron. Lett. 33 (4) (1997) 313e314. [27] C. Konishi, T. Yoshida, S. Hamada, K. Asahi, S. Fujita. Dynamic gain-controlled erbium-doped fiber amplifier repeater for WDM network. Conference on Optical Fiber Communication (OFC 1997), Dallas, TX, USA, 1997, February 16e21, 6, paper TuE1, pp. 18e19. [28] J. Chung, S.Y. Kim, C.J. Chae. Performances of all optical gain-clamped EDFAs with different feedback wavelengths for use in multiwavelength optical networks. Conference on Optical Fiber Communication (OFC 1997), Dallas, TX, USA, 1997, Febrary 16e21, 6, paper TuE5, pp. 22e24. [29] J.L. Jackel, D. Richards. All-optical stabilization of cascaded multichannel erbiumdoped fiber amplifiers with changing numbers of channels. Conference on Optical Fiber Communication (OFC 1997), Dallas, TX, USA, 1997, February 16e21, 6, paper TuP4, pp. 84e85. [30] G. Luo, J.L. Zyskind, Y. Sun, A.K. Srivastava, J.W. Sulhoff, M.A. Ali. Relaxation oscillations and spectral hole burning in laser automatic gain control of EDFAs. Conference on Optical Fiber Communication (OFC 1997), Dallas, TX, USA, 1997, February 16e21, 6, paper WF4, pp. 130e131.
195
196
CHAPTER 6 Mastering power transients
[31] H. Dai, J.-Y. Pan, C. Lin. All-optical gain control of in-line EDFA for hybrid AM/ digital WDM systems. Conference on Optical Fiber Communication (OFC 1997), Dallas, TX, USA, 1997, February 16e21, 6, paper WF6, pp. 133e134. [32] N.E. Jolley, F. Davis, J. Mun. Out-of-band electronic gain clamping for a variable gain and output power EDFA with low dynamic gain tilt. Conference on Optical Fiber Communication (OFC 1997), Dallas, TX, USA, 1997, February 16e21, 6, paper WF7, pp. 134e135. [33] A.K. Srivastava, Y. Sun, J.L. Zyskind, J.W. Sulhoff, EDFA Transient Response to Channel Loss in WDM Transmission System. IEEE Photon. Technol. Lett. 9 (3) (1997) 386e388. [34] H. Dai, J.-Y. Pan, C. Lin, All-optical gain control of in-line erbium-doped fiber amplifiers for hybrid analog/digital WDM systems. IEEE Photon. Technol. Lett. 9 (6) (1997) 737e739. [35] Y. Sun, A.K. Srivastava. Dynamic Effects in Optically Amplified Networks. Conference on Optical Amplifiers and their Applications (OAA 1997), Victoria, BC, Canada, 1997, July 21e23, paper MC4, pp. 44e47. [36] S.Y. Kim, J. Chung. Effects of Relaxation Oscillations on Transmission Performances in the All-Optical Link-Controlled EDFA Cascade. Conference on Optical Amplifiers and their Applications (OAA 1997), Victoria, BC, Canada, 1997, July 21e23, paper MC5, pp. 48e51. [37] D.H. Richards, J.L. Jackel, M.A. Ali, A theoretical investigation of dynamic all-optical automatic gain control in multichannel EDFA’s and EDFA cascades, IEEE J. Sel Topics Quantum Electronics 3 (4) (1997) 1027e1036. [38] S.Y. Kim, J. Chung, B. Lee, Dynamic performance of the all-optical gain controlled EDFA cascade in multiwavelength optical networks, Electron. Lett. 33 (17) (1997) 1475e1477. [39] D.H. Richards, J.L. Jackel, M.A. Ali, EDFA chain control: a comparison of two alloptical approaches, IEEE LEOS Summer Topical Meeting, Montreal, Quebec, Canada, 1997, August 11e15, paper FB2pp. 70e71. [40] A.K. Srivastava, Y. Sun, Optical amplifiers for optical networks, IEEE LEOS Summer Topical Meeting, Montreal, Quebec, Canada, 1997. August 11e15, paper FB4 pp. 74e75. [41] D.H. Richards, J.L. Jackel, M.A. Ali, A theoretical investigation of dynamic all-optical automatic gain control in multichannel EDFA’s and EDFA cascades, IEEE J. Selected Topics Quantum Electronics 3 (4) (1997) 1027e1036. [42] D.H. Richards, M.A. Ali, J.L. Jackel. A theoretical investigation of dynamic automatic gain control in multi-channel EDFA cascades. 23rd European Conference on Optical Communication (ECOC 1997), Edinburgh, UK, 1997, September 22e25, 3, pp. 47e50. [43] J. Chung, S.Y. Kim. Dynamic Performance of the All-Optical Gain-Controlled EDFA Cascade in Multi-Wavelength Add/Drop Networks. 23rd European Conference on Optical Communication (ECOC 1997), Edinburgh, UK, 1997, September 22e25, 2, pp. 139e142. [44] T. Tokura, J. Nakagawa, K. Motoshima, T. Kitayama. Quantitative analysis of optical surge propagation on transmission systems. 23rd European Conference on Optical Communication (ECOC 1997), Edinburgh, UK, 1997, September 22e25, 3, pp. 263e266
References
[45] J.C. van der Plaats, F.W. Willems, A.M.J. Koonen. Dynamic pump-loss controlled gain-locking system for erbium-doped fiber amplifiers in multi-wavelength networks. 23rd European Conference on Optical Communication (ECOC 1997), Edinburgh, UK, 1997, September 22e25, 3, pp. 127e130. [46] S.Y. Park, H.K. Kim, G.Y. Lyu, H.J. Lee, J.H. Lee, S.-Y. Shin. Accurate control of output power level in gain-flattened EDFA with low noise figure. 23rd European Conference on Optical Communication (ECOC 1997), Edinburgh, UK, 1997, September 22e25, 3, pp. 43e46. [47] G. Luo, J.L. Zyskind, Y. Sun, A.K. Srivastava, J.W. Sulhoff, C. Wolf, et al., Performance degradation of all-optical gain-clamped EDFAs due to relaxation-oscillations and spectral-hole burning in amplified WDM networks, IEEE Photon. Technol. Lett. 9 (10) (1997) 1346e1348. [48] A.K. Srivastava, J.L. Zyskind, Y. Sun, J. Ellson, G. Newsome, R.W. Tkach, et al., FastLink Control Protection of Surviving Channels in Multiwavelength Optical Networks, IEEE Photon. Technol. Lett. 9 (12) (1997) 1667e1669. [49] D.H. Richards, J.L. Jackel, M.A. Ali, Multichannel EDFA chain control: a comparison of two all-optical approaches, IEEE Photon. Technol. Lett. 10 (1) (1998) 156e158. [50] S.J.B. Yoo, J.C. Young, W. Xin, G. Ellinas, M. Rauch, H. Leblanc, et al. Suppression of power fluctuations due to interaction between cascaded automatic channel power equalizers, and fiber amplifiers in a 1018-km fiber link in a reconfigurable multiwavelength network. Conference on Optical Fiber Communication (OFC 1998), San Jose, CA, USA,, 1998, February 22e27, paper WJ2, pp. 162e163. [51] A.K. Srivastava, Y. Sun, J.L. Zyskind, J.W. Sulhoff, T.A. Strasser, C. Wolf, J.R. Pedrazzani. Signal power transients in optically amplified WDM ring networks. Conference on Optical Fiber Communication (OFC 1998), San Jose, CA, USA, 1998, February 22e27, paper WJ3, pp. 164e165. [52] N. Takachio, H. Suzuki, M. Koga, O. Ishida. WDM linear repeater gain control scheme by automatic maximum power channel selection for photonic transport network. Conference on Optical Fiber Communication (OFC 1998), San Jose, CA, USA, 1998, February 22e27, paper WJ4, pp. 165e166. [53] F. Bruyere, A. Bisson, L. Noirie, J.-Y. Emery, A. Jourdan. Gain stabilization of EDFA cascade using clamped-gain SOA. Conference on Optical Fiber Communication (OFC 1998), San Jose, CA, USA, 1998, February 22e27, paper WJ5, pp. 166e167. [54] S.H. Lee, S.H. Kim. All-optical gain-clamped erbium-doped fiber amplifier for wavelength-division multiplexed networking using stimulated Brillouin scattering. Conference on Optical Fiber Communication (OFC 1998), San Jose, CA, USA, 1998, February 22e27, paper WM52, pp. 241e242. [55] F.W. Willems, J.C. van der Plaats, M., Karasek, K.L. Walker. Inherently small multichannel addition/removal response in cascaded EDFAs. Conference on Optical Fiber Communication (OFC 1998), San Jose, CA, USA, 1998, February 22e27, paper ThO1, pp. 316e318. [56] J. Nagel. The dynamic behavior of amplified systems. Conference on Optical Fiber Communication (OFC 1998), San Jose, CA, USA, 1998, February 22e27, paper ThO3, pp. 319e320. [57] Y. Takushima, K. Kikuchi, Gain stabilisation of all-optical gain-clamped amplifier by using Faraday rotator mirrors, Electronics Letters 34 (5) (1998) 458e460.
197
198
CHAPTER 6 Mastering power transients
[58] S.Y. Park, H.K. Kim, D.H. Lee, S.-Y. Shin, Feasibility demonstration of 10 Gbit/s/ channel WDM network using dynamic gain-controlled EDFAs, Electronics Letters 34 (5) (1998) 482e484. [59] G. Luo, J.L. Zyskind, J.A. Nagel, M.A. Ali, Experimental and theoretical analysis of relaxation-oscillations and spectral hole burning effects in all-optical gainclamped EDFA’s for WDM networks, IEEE J. Lightwave Technol. 16 (4) (1998) 527e533. [60] H. Suzuki, N. Takachio, O. Ishida, M. Koga, Dynamic gain control by maximum signal power channel in optical linear repeaters for WDM photonic transport networks, IEEE J. Lightwave Technol. 16 (5) (1998) 734e736. [61] A. Mecozzi, D. Marcenac, Theory of optical amplifier chains, IEEE J. Lightwave Technol. 16 (5) (1998) 745e756. [62] A. Bononi, L.A. Rusch, Doped-Fiber Amplifier Dynamics: A System Perspective, IEEE J. Lightwave Technol. 16 (5) (1998) 945e956. [63] S.Y. Ko, M.W. Kim, D.H. Kim, S.H. Kim, J.C. Jo, J.H. Park, Gain control in erbiumdoped fibre amplifiers by tuning centre wavelength of a fibre Bragg grating constituting resonant cavity, Electronics Letters 34 (10) (1998) 990e991. [64] S.Y. Park, H.K. Kim, G.Y. Lyu, S.M. Kang, S.-Y. Shin, Dynamic Gain and Output Power Control in a Gain-Flattened Erbium-Doped Fiber Amplifier, IEEE Photon. Technol. Lett. 10 (6) (1998) 787e789. [65] S.Y. Park, H.K. Kim, S.M. Kang, G.Y. Lyu, H.J. Lee, J.H. Lee, et al., A gain-flattened two-stage EDFA for WDM optical networks with a fast link control channel, Optics Comm 153 (1998) 23e26. [66] J. Bryce, G. Yoffe, Y. Zhao, R. Minasian, Tunable, gain-clamped EDFA incorporating chirped fibre Bragg grating, Electronics Letters 34 (17) (1998) 1680e1681. [67] M. Karasek, J.C. van der Plaats, Modelling of a pump-power-loss-controlled gainlocking system for EDFA application in WDM transmission systems, IEE Proc. Optoelectron 145 (4) (1998) 205e210. [68] J. Lee, G.H. Song, K. Oh, Coaxial-core erbium-doped fibre amplifiers for self-regulation of gain spectrum, Electronics Letters 34 (19) (1998) 1852e1854. [69] L. Tancevski, L.A. Rusch, A. Bononi, Gain control in EDFA’s by pump compensation. IEEE Photon. Technol. Lett. 10 (9) (1998) 1313e1315. [70] S.H. Lee, S.H. Kim, All optical gain-clamping in erbium-doped fiber amplifier using stimulated Brillouin scattering, IEEE Photon. Technol. Lett. 10 (9) (1998) 1316e1318. [71] M. Karasek, J.A. Valles, Analysis of channel addition/removal response in all-optical gain-controlled cascade of erbium-doped fiber amplifiers, IEEE J. Lightwave Technol. 16 (10) (1998) 1795e1803. [72] M. Karasek, J.C. van der Plaats, Analysis of dynamic pump-loss controlled gainlocking system for erbium-doped fiber amplifiers. IEEE Photon, Technol. Lett. 10 (11) (1998) 1171e1173. [73] S.J.B. Yoo, W. Xin, L.D. Garratt, J.C. Young, G. Ellinas, J.C. Chiao, et al., Observation of prolonged power transients in a reconfigurable multiwavelength network and their suppression by gain-clamping of optical amplifiers, IEEE Photon. Technol. Lett. 10 (11) (1998) 1659e1661. [74] L. Tancevski, A. Bononi, L.A. Rusch, Output power and SNR swings in cascades of EDFAs for circuit- and packet-switched optical networks, IEEE J. Lightwave Technol. 17 (5) (1998) 733e742.
References
[75] K. Inoue, Gain-clamped fiber amplifier with a loop mirror configuration, IEEE Photon. Technol. Lett. 11 (5) (1999) 533e535. [76] M.I. Hayee, A.E. Willner, Transmission penalties due to EDFA gain transients in adddrop multiplexed WDM networks, IEEE Photon. Technol. Lett. 11 (7) (1999) 889e891. [77] A. Bononi, L. Barbieri, Design of gain-clamped doped-fiber amplifiers for optimal dynamic performance, IEEE J. Lightwave Technol. 17 (7) (1999) 1229e1239. [78] K.-W. Na, J.-T. Choi, W.-J. Lee, S.-H. Park, W.-W. Yoon, K.-K. Lee, A cost-effective gain control using pump modulation for erbium-doped fiber amplifiers, IEEE Photon. Technol. Lett. 12 (4) (2000) 383e385. [79] T.C. Teyo, N.S.M. Shah, M.K. Leong, P. Poopalan, H. Ahmad, Comparison between regenerative-feedback and cofeedback gain-clamped EDFA, IEEE Photon. Technol. Lett. 14 (9) (2002) 1255e1257. [80] S. Pachnicke, M. Obholz, E. Voges, P.M. Krummrich, E. Gottwald. Electronic EDFA gain control for the suppression of transient gain dynamics in long-haul transmission systems. Conference on Optical Fiber Communication (OFC 2007), Anaheim, CA, USA, 2007, March 25e29, paper JWA15. [81] S. Pachnicke, P.M. Krummrich, E. Voges, E. Gottwald, Transient gain dynamics in long-haul transmission systems with electronic EDFA gain control, Journal of Optical Networking 6 (9) (2007) 1129e1137. [82] E. Desurvire, J.L. Zyskind, J.R. Simpson, Spectral gain hole-burning at 1.53 mm in erbium-doped fiber amplifiers, IEEE Photon. Technol. Lett. 2 (4) (1990) 246e248. [83] E. Desurvire, J.W. Sulhoff, J.L. Zyskind, J.R. Simpson, Study of Spectral Dependence of Gain Saturation and Effect of Inhomogeneous Broadening in ErbiumDoped Aluminosilicate Fiber Amplifiers, IEEE Photon. Technol. Lett. 2 (9) (1990) 653e655. [84] A.K. Srivastava, J.L. Zyskind, J.W. Sulhoff, J.D. Evankow, M.A. Mills. Room temperature spectral hole-burning in erbium-doped fiber amplifiers. Conference on Optical Fiber Communication (OFC 1996), San Jose, CA, USA, 1996, February 25eMarch 1, paper TuG7, pp. 33e34. [85] J.W. Sulhoff, A.K. Srivastava, C. Wolf, Y. Sun, J.L. Zyskind, Spectral-Hole Burning in Erbium-Doped Silica and Fluoride Fibers, IEEE Photon. Technol. Lett. 9 (12) (1997) 1578e1579. [86] P.M. Krummrich, E. Gottwald, A. Mayer, R. Neuhauser, G. Fischer. Channel power transients in photonic networks caused by stimulated Raman scattering. Conference on Optical Amplifiers and their Applications (OAA 2000), Quebec, Canada, 2000, July 9e12, OtuC6, pp. 143e145. [87] P.M. Krummrich, Raman impairments in WDM systems, in: M.N. Islam (Ed.), Raman amplifiers for telecommunications 2, Springer, New York, 2003. Chapter 16. [88] Fu¨rst, C., Hartung, R., Elbers, J.-P., Glingener, C., 2003. Impact of Spectral Hole Burning and Raman Effect in Transparent Optical Networks. Proceedings 29th European Conference on Optical Communication (ECOC 2003), Rimini, Italy, September 21e25, paper Tu4.2.5. [89] S. Pachnicke, E. Gottwald, P. Krummrich, E. Voges. Combined impact of Raman and EDFA transients on long haul transmission system performance. 33rd European Conference on Optical Communication (ECOC 2007), Berlin, Germany, 2007, September 16e20, paper P074.
199
200
CHAPTER 6 Mastering power transients
[90] S.Y. Park, H.K. Kim, S.M. Kang, G.Y. Lyu, H.J. Lee, J.H. Lee, S.-J. Shin, A gainflattened two-stage EDFA for WDM optical networks with a fast link control channel, Optics Communications 153 (1-3) (1998) 23e26. [91] K.-M. Feng, J.-X. Cai, X.P. Chen, A.E. Willner, D.A. Smith. Experimental demonstration of dynamic high-speed equalization of three channels using wavelength demultiplexer and acoustooptic modulators. Conference on Optical Fiber Communication (OFC 1997), Dallas, TX, USA, 1997, February 16e21, 6, paper ThS4, pp. 334e335. [92] J.-X. Cai, K.-M. Feng, A.E. Willner. Simultaneous compensation of fast add/drop power-transients and equalization of inter-channel power differentials for robust WDM systems with EDFAs. Conference on Optical Amplifiers and their Applications (OAA 1997), Victoria, BC, Canada, 1997, July 21e23, paper MC6, pp. 52e55. [93] J.-X. Cai, K.-M. Feng, X. Chen, A.E. Willner, D.A. Smith, C.-H. Lee, Y.-J. Chen, Experimental demonstration of dynamic high-speed equalization of three WDM channels using acoustooptic modulators and a wavelength demultiplexer, IEEE Photon. Technol. Lett. 9 (5) (1997) 678e680. [94] P.M. Krummrich, R.E. Neuhauser, H.-J. Schmidtke, H. Zech, M. Birk. Compensation of Raman transients in optical networks. Conference on Optical Fiber Communication (OFC 2004), Los Angeles, CA, USA, 2004, February 22e27, paper MF82. [95] M. Zirngibl, Analytical model of Raman gain effects in massive wavelength division multiplexed transmission systems, Electronics Letters 34 (8) (1998) 789e790. [96] P.M. Krummrich, M. Birk, Experimental Investigation of Compensation of Raman Induced Power Transients from WDM Channel Interactions, IEEE Photonics Technology Letters 17 (5) (2005) 1094e1096.
CHAPTER
Spectral Power Fluctuations in DWDM Networks Caused by Spectral-Hole Burning and Stimulated Raman Scattering
7
Jo¨rg-Peter Elbers, Cornelius Fu¨rst ADVA AG Optical Networking
CHAPTER OUTLINE HEAD 7.1. Introduction ................................................................................................ 7.2. Description of physical effects ..................................................................... 7.3. Experimental investigation of SHB and SRS impact........................................ 7.3.1. Experimental setup .................................................................... 7.3.2. Experimental results .................................................................. 7.4. Black-box modeling of the amplifier spectral gain, SHB and SRS impact......... 7.4.1. Scenario 1 - Surviving channels in the red band ........................... 7.4.2. Scenario 2 - Surviving channels in the blue band ......................... 7.4.3. Scenario 3 - Surviving channels in the center ............................... 7.4.4. Scenario 4 - Surviving channels evenly distributed........................ 7.5. Parameterization of wavelength dependent power excursions ........................ 7.6. Mitigation of spectral power excursions ....................................................... 7.6.1. Fast optical control of channel power levels ................................. 7.6.2. Substitution of missing channels................................................. 7.6.3. Wavelength-dependent reach and light path assignment................ 7.7. Summary and conclusion ............................................................................. Acknowledgements ............................................................................................. Acronyms ........................................................................................................... References .........................................................................................................
201 203 206 206 207 209 211 212 212 212 212 215 215 215 216 218 218 218 219
7.1 INTRODUCTION Advances in terrestrial fiber transmission and the availability of multi-degree reconfigurable optical add/drop multiplexers MD-ROADMs facilitate the commercial deployment of transparent optical dense wavelength-division multiplexing DWDM networks. Typically operating with 80 10 or 40 Gb/s channels at 50 GHz Optically Amplified WDM Networks. DOI: Copyright Ó 2011 Elsevier Inc. All rights reserved.
201
202
CHAPTER 7 Spectral power fluctuations in DWDM networks
channel spacing, these networks allow un-regenerated optical path lengths in excess of 2000 km. Transparent optical light paths traverse multiple ROADMD nodes. The nodes are connected by fiber links comprising several fiber spans (with 80 km a typical length) periodically amplified by erbium-doped fiber amplifiers (EDFAs). The economic benefit of transparent DWDM networks is evident [1]: because many network nodes have to process a large amount of transit traffic, a significant reduction in optical/electrical/optical (conversion) (O/E/O) interfaces can be achieved by a transparent optical bypass of the through traffic. With O/E/O interfaces dominating the network cost already at medium load conditions, an elimination of unnecessary O/E/O interfaces directly translates into substantial cost savings. The transparent networking approach, however, gives rise to new challenges on the physical layer. While optical networks traditionally consist of point-to-point links terminated by O/E/O interfaces or transponders, the elimination of these O/E/O conversion points introduces an optical coupling between internodal links in the transparent network: due to optical amplifier characteristics and fiber nonlinearities, sudden changes in the channel count in one link can influence the channel power levels in other network segments. Appropriate scheduling can prevent rapid and at the same time large load changes in case of an intentional reconfiguration or commissioning/decommissioning of channels. Critical, though, are networks failures that lead to an abrupt loss (drop event) or insertion (add event) of channels: surviving channels may experience large power transients which can exceed the receiver dynamic range, substantially lower the optical signal-to-noise ratio (OSNR), and/or increase nonlinear signal degradations. Bit error ratio (BER) bursts or channel outages are the likely result. Figure 7.1 exemplifies a possible network scenario. In a meshed network topology, 32 wavelengths are suddenly dropped due to a network failure. The surviving eight channels travel on a different optical end-to-end path but share links and nodes with the dropped channels and may consequently experience power 32 λ dashed channels see power changes
no O/E/O conversion
8λ
FIGURE 7.1 Power transients in transparent optical networks. The links connecting the ROADM nodes (shown) may include multiple optically amplified spans (not shown).
7.2 Description of physical effects
Table 7.1 Origins of Fast Power Transients Mechanism
Switching Time
Comments
Fiber cut Unplug of connector Break of fusion splice Optical switching Card failure
>1 ms >100 ms 1.5 dB higher than the power at optimum performance for the 17 GHz system due to reduced cross-channel penalties in the 33 GHz system.
FIGURE 9.17 Measured performance in ROPA/EDFA chain versus pre-emphasis for 33 and 17 GHz channel spacing
9.3 Comparison of Raman- and ROPA-assisted systems
It bears pointing out that the 17 GHz system operates optimally at a higher total signal power out of the repeater than a 33 GHz system (lower per-channel power, but many more channels). Repeater signal power in the ROPA/EDFA experimental test bed was originally optimized to transmit 200x10 Gb/s channels with 17 GHz channel spacing where the Q-Factor for all channels were measured across the full bandwidth of the system. In contrast, the hybrid DRA/EDFA test bed was optimized for 96x10Gb/s channels with 33 GHz channel spacing. So in order to realize a fair performance comparison between the ROPA/EDFA test bed and the DRA/EDFA test bed at 33 GHz, signal powers for the former would first need to be de-emphasized. This was achieved by using what has now become a standard experimental technique, the tunable comb method with 64 loading tones [15]. In this method, unmodulated loading tones maintain the proper level of inversion and power spectral density across the signal band. Then, to measure the performance of a channel, a few loading tones are replaced by six modulated signals, and the performance of the center channel is measured. In general, this group of channels is sequentially moved across the signal band so that performance at each signal channel wavelength may be evaluated. In this experiment, as the group of six signal channels was tuned to different parts of the band, they were propagated with signal power approximately 2 dB lower than the repeater design target in order to operate at the optimum channel power for the chosen channel spacing. The performance of the channels in the middle of the group was then measured as described above. Figure 9.18 shows this direct experimental performance comparison between the ROPA/EDFA scheme and the DRA/EDFA scheme at 33 GHz channel spacing at transmission distances of 9342 km and 8900 km respectively. When
FIGURE 9.18 System performance for the DRA/EDFA test bed at optimum power level and the ROPA/EDFA test bed with signal power adjusted for optimum performance for the 33 GHz channel spacing
269
270
CHAPTER 9 Advanced schemes in long-Haul undersea systems
signal power pre-emphasis is properly set so that the two systems are operating under equivalent conditions, the measured Q-factor performance of the ROPA/EDFA exceeds by 0.2 dB to 0.8 dB the performance of the DRA/EDFA. Further, correcting for the disparity in system lengths, this translates to an advantage of 0.4 dB to 1.0 dB for the ROPA system. Recall (Figure 9.15) that the OSNR advantage of hybrid ROPA over hybrid Raman is approximately 1 dB. Thus we see that the Q performance advantage largely derives from the relative advantage in noise properties. Two earlier papers of note established the first records of transmission over transPacific distances with repeater spans of 100 km [16] and 150 km [17] using Raman assisted EDFAs. The first experiment successfully demonstrated 64 11.4 Gb/s 9000 km transmission with 100 km repeater span by using Raman assisted EDFA for the first time. A unique aspect of this design is that the DRA and the EDFA segments shared a common pump source (Figure 9.19). Two 1460 nm pump diodes were combined by a PBS coupler and directed to the EDF. The residual pump was fed to the transmission line to power the DRA section. Error-free transmission was enabled in that case by using 14% redundancy super FEC in conjunction with the low noise hybrid amplifiers. That result first demonstrated the feasibility of trans-Pacific distance transmission with 100 km repeater spans using Raman assisted EDFAs. In the second paper by the same group [17], by using a narrower portion of the available bandwidth, 32 12.4 Gb/s, 9000 km transmission with 150 km repeater span was demonstrated for the first time. The received optical spectrum (for continuous wave [CW] tones) is presented in Figure 9.20. Both experiments had channel spacing of 37.5 GHz. The small variation in power across the transmission bandwidth allowed for a received OSNR variation of only a few tenths of a dB by the use of simple signal power pre-emphasis at the transmitter, with an average OSNR of 9.5 dB. The received Q-factor was similarly smooth across the bandwidth, with an average Q of 9.5 dB. Those results showed the feasibility of 2.8 dB FEC margin transmission after trans-Pacific distance with 150 km repeater span using Raman assisted EDFA and carrier suppressed return to zero (CSRZ)edigital phase-shift keying (DPSK) modulation format.
FIGURE 9.19 A schematic diagram of the hybrid DRA-EDFA which enabled transpacific transmission with 100 km repeater spacing [16]
9.4 Advanced modulation formats and high capacity
FIGURE 9.20 Optical spectrum for the 150 km span system after 9000 km transmission. Signal power preemphasis at the transmitter provided flat OSNR at the receiver.
9.4 ADVANCED MODULATION FORMATS AND HIGH CAPACITY IN HYBRID ROPA-EDFA SYSTEMS Given the present need for ever-increasing system capacity, it is important to demonstrate the compatibility of hybrid amplifiers with the enabling technologies and techniques that are being developed and deployed, such as the various implementations of digital phase shift key modulation format (DBPSK and digital quadrature phase-shift keying [DQPSK], for example), coherent transmission and detection, higher data rates, and tighter channel spacing. While enabling greater fiber capacity, these technologies in many cases will be more demanding in terms of noise performance because of higher data rates, more complex signal constellations, or greater susceptibility to nonlinear impairments related to closer channel spacing. The improved noise performance of hybrid amplifiers can thus be important for such systems. A number of transmission results with high spectral efficiency 0.6 Bit/Hz/s have already been obtained with ROPA/EDFA scheme using its superior OSNR performance. Foursa et al. [12] demonstrated transmission of 2 Tb/s (200 10 Gb/s) of data over 7300 km with a Q factor that varied from 10.5 to 12.3 dB (Figure 9.21). The 10 Gb/s RZ-DBPSK channels had 16.6 GHz separation and orthogonal launch of neighboring channels, and prefiltering techniques were used to boost performance. In part, the excellent Q performance was due to the noise advantage of the ROPA/EDFA system design over the DRA/EDFA design (see Figure 9.15). A fair comparison of two laboratory test beds, scaling properly for distance and channel power, demonstrates that the ROPA/EDFA noise advantage ranges from 1 to 3 dB across the signal band (Figure 9.22). Similar to the trend in OSNR, the Q-factor increased with wavelength and varied from 10.5 dB in the short wavelength region up to 12.3 dB in the long wavelength region. Over that same loop test bed in a 40 Gb/s experiment, 50 CSRZ-DBPSK channels (2 Tb/s capacity) were transmitted over 3100 km with Q factors all > 4 dB above the FEC threshold of 9.1 dB. As in the 10 Gb/s experiment, measured Q performance varied across the signal band from
271
CHAPTER 9 Advanced schemes in long-Haul undersea systems
14 13
ROPA/EDFA
OSNR [dB]
Raman/EDFA 12 11 10 9 8 1535
1540
1545
1550
1555
1560
1565
Wavelength [nm] FIGURE 9.21 Performance of 200x10G RZ-DBPSK channels after 7300 km distance using the ROPA/ EDFA span design 13 12
Q-factor [dB]
272
11 10 9 8 7 6 1535
FEC Threshold
1540
1545
1550
1555
1560
1565
Wavelength [nm] FIGURE 9.22 OSNR in 0.1 nm resolution bandwidth after 9 Mm for hybrid DRA/EDFA and ROPA/EDFA systems
12.9 dB to 14.2 dB (Figure 9.23). The repeater output power was optimized for 40 Gb/s transmission at the chosen distance. An added challenge to long-haul optical paths using 40 Gb/s channels is successful transmission at high spectral efficiency (>0.5 Bit/Hz/s). A recent experiment over a recirculating loop test bed (ROPA/EDFA with 150 km spans)
9.4 Advanced modulation formats and high capacity
15
Q-factor [dB]
14 13 12 11 10
FEC Threshold
9 8 1535
1540
1545
1550
1555
1560
1565
Wavelength [nm] FIGURE 9.23 Q-factor measured for 50 channels at a signal line rate of 42.8 Gb/s after 3100 km transmission distance using RZ-DBPSK
transmitted 2 Tb/s over 5,200 km at 60% spectral efficiency (0.6 Bit/Hz/s). That record was enabled through the use of the polarization multiplexed signals with RZ-DPSK modulation format and an automatic polarization tracking scheme [18].Polarization multiplexed [PM]eRZeDPSK modulation format increases nonlinear tolerance relative to 40 Gb/s RZ-DBPSK signals. Figure 9.24 summarizes the Q performance across the signal band after 5200 km. The automatic polarization
15 Ch26
Q-factor [dB]
14 13 12
3.5 dB
11 Concatenated RS FEC Threshold
10 9 1536
1541
1546
1551
1556
1561
1566
Wavelength [nm]
FIGURE 9.24 Q-factor for PM RZ-DBPSK after 5200 km. Each polarization was measured for 2 minutes, and the average reported here.
273
274
CHAPTER 9 Advanced schemes in long-Haul undersea systems
tracking receiver demonstrated long-term stability even in a loop test bed which has a state of polarization which is discontinuous by its nature. Figure 9.24 also shows the FEC threshold for a standard concatenated Reed-Solomon code with a 9.1 dB threshold. In this case, the worst channel in the system has a 3.5 dB average FEC margin.
9.5 CONCLUSION We have presented in this chapter a survey of the recent history of the development of the DRA and ROPA technologies and their implementation in some of the significant laboratory demonstrations in the literature. The hybrid DRA-EDFA provides the advantages of the pump power efficiency and well developed technology of the EDFA combined with the low noise figure and broadband capability of the Raman gain. Furthermore, the Raman section of the hybrid DRA-EDFA can be configured to either provide extra gain in the conventional portion of the signal band or to provide extra signal bandwidth, all determined by selection of the wavelength of the Raman pump laser. Using a simple combination of a singlewavelength pumped DRA and C-band EDFA we have demonstrated 80 nm of continuous bandwidth. This compares favorably to the 18 nm to 35 nm of conventional EDFA alone. Transmission experiments have confirmed high performance of the hybrid DRA-EDFA, demonstrating 11,000-km distances. This good transmission performance has been realized with both RZeon-off keying (OOK) and RZ-DBPSK formats. In the case of DBPSK modulation format transmission experiment, an average of 2.6 dB performance above the FEC threshold was achieved at a transmission distance of 11,000 km with 40% spectral efficiency. A detailed analysis of both the fundamental properties and the signal power and noise performance enhancement of the RA and ROPA technologies when deployed in hybrid design with EDFAs highlights the potential contribution of each to improvement in system performance. Further, the specific examples we have surveyed provided a fair comparison between the designs and lay the groundwork for design decisions between the two amplifier types. In the reviews of our own experimental and modeling work we implemented reasonably simple amplifier architectures for both Distributed Raman and ROPA assisted EDFA (DRA/EDFA and ROPA/EDFA). Consistently, the hybrid amplifier designs allowed for the large repeater spacing, while the simple pre-filter technique allowed increased 980 nm EDFA pump efficiency. Finally, a number of recent published studies which used hybrid amplification techniques have confirmed that the improved OSNR performance of hybrid amplification systems are clearly supportive of the advanced modulation formats, high data rates, and high spectral efficiencies that are moving rapidly into the optical telecommunications industry.
References
ACRONYMS ASE BER C-Band or L-Band CSRZ CW dB dBm DBPSK DPSK DFB DFF DQPSK DRA DRBS EDF EDFA FEC GFF Mm MPI NZDSF OOK OSNR PAI PM Q ROPA RA RZ SMF SRBS WDM
Amplified spontaneous emission Bit error ratio The conventional or the long-wavelength signal bands Carrier suppressed return to zero Continuous wave, i.e., no modulation on the channel Decibel Logarithm of power ratios normalized to a milliwatt Digital binary phase-shift keying Digital phase-shift keying Distributed feedback laser Dispersion-fattened fibers Differential quadrature phase-shift keying Distributed Raman amplifier (or amplification) Double Rayleigh back scatter Erbium-doped fiber Erbium-doped fiber amplifier (or amplification) Forward error correction Gain flattening filter One million meters, or one thousand kilometers (km) Multipath interference Non-zero dispersion shifted fiber On-off keying Optical signal-to-noise ratio Path average intensity Polarization multiplexed pffiffi Q is related to BER as Q ¼ 12ERFC BER 2 Remote optically pumped amplifier Raman amplifier (or amplification) Return to zero Single mode fiber Single Rayleigh back scatter Wavelength division multiplexing
References [1] P. Trischitta, W. Marra, Global Undersea Communications Networks, IEEE Communications Magazine 34 (2) (1996).
275
276
CHAPTER 9 Advanced schemes in long-Haul undersea systems
[2] N. Bergano, in: I.P. Kaminow, T. Li (Eds.), Undersea Amplified Lightwave Systems Design, Optical Fiber Telecommunications IIIA, Academic Press, San Diego, CA, 1997, pp. 302e335. [3] A. Pilipetskii, et al., Challenges of High Capacity Undersea Long-Haul Systems, in: Proc. Asia Communications and Photonics Conference, paper THC1, 2009. [4] S. Abbott, Optical Amplifiers in Undersea Systems, in: Proc. Optical Amplifiers and Their Applications, Paper OMC1, 2004. [5] H. Masuda, Hybrid EDFA/Raman Amplifiers, in: M.N. Islam (Ed.), Raman Amplifiers for Telecommunications 1, Springer-Verlag, New York, 2004, pp. 413e444. [6] T. Mizuochi, LEOS 2004 Annual Meeting 2 (2004) 467e468. [7] E.A. Golovchenko, et al., Pushing the Reach of Repeaterless Transmission Systems, in: Proc. Suboptic, paper We.207, 2007. [8] D. Foursa, et al., Ultra wide-band amplifiers for transoceanic length transmission, in: Proc. Optical Amplifiers and Their Applications, Paper OTuD1, 2003. [9] J. Bromage, et al., Multiple Path Interference and Its Impact on System Design, in: M.N. Islam (Ed.), Raman Amplifiers for Telecommunications 2, Springer, New York, 2004, pp. 491e568. [10] A. Lucero, et al., Long-Haul Raman-Assisted EDFA Systems with Ultra-Long Spans, in: Proc. Optical Fiber Communications Conference, Paper OFD2, 2006. [11] D. Foursa, et al., Transmission over 8,900 km with 150-km spans using a novel gain equalization scheme in Raman assisted EDF amplification, in: Proc. European Conference on Optical Communications, post-deadline paper Th4.1.7, 2006. [12] D. Foursa, et al., 2Tb/s (200x10Gb/s) data transmission over 7,300km using 150km spaced repeaters enabled by ROPA technology, in: Proc. Optical Fiber Communications Conference, post-deadline paper PDP-25, 2007. [13] J.-X. Cai, et al., Long-Haul 40 Gb/s RZ-DPSK Transmission over 4,450 km with 150-km Repeater Spacing using Raman Assisted EDFAs, in: Proc. Optical Fiber Communications Conference, paper OWM3, 2007. [14] A. Lucero, et al., Advanced repeater architectures with ultra-long spans for submarine systems, in: Proc. Optical Fiber Communications Conference, paper OTuE3, 2008. [15] J.-X. Cai, et al., A DWDM demonstration of 3.73 Tb/s over 11,000 km using 373 RZ-DPSK channels at 10 Gb/s, in: Proc. Optical Fiber Communications Conference, post-deadline paper PDP-22, 2003. [16] T. Inoue, et al., First Transpacific Distance Transmission Experiment Using Raman Assisted EDF Amplifier with 100 km Repeater Span, ECOC 2004, Th.3.5.2, Stockholm, Sweden, 2004. [17] T. Inoue, et al., 150 km Repeater Span Transmission Experiment over 9,000 km, ECOC 2004, PD4.1.3, Stockholm, Sweden, 2004. [18] J.-X. Cai, et al., 40 Gb/s Transmission Using Polarization Division Multiplexing (PDM) RZ-DBPSK with Automatic Polarization Tracking, in: Proc. Optical Fiber Communications Conference, post-deadline paper PDP-4, 2007.
CHAPTER
Challenges for Long-haul and Ultra-long-haul Dynamic Networks
10 Martin Birk, Kathy Tse AT&T Labs., USA
CHAPTER OUTLINE HEAD 10.1. Photonic network evolution ........................................................................ 10.1.1. Data rate evolution ................................................................. 10.1.2. WDM evolution ...................................................................... 10.1.3. Network architecture evolution ................................................ 10.1.4. Amplifier evolution ................................................................. 10.1.5. Status quo............................................................................. 10.2. Requirements for amplifiers in today’s photonic mesh networks ................... 10.2.1. Optical link control ................................................................ 10.2.2. Static amplifier requirements .................................................. 10.2.3. Dynamic amplifier requirements .............................................. 10.3. The future: requirements for a fully dynamic photonic mesh......................... 10.3.1. Provisioning through pre-cabling ............................................. 10.3.2. Simple photonic restoration techniques ................................... 10.3.3. The holy grail: photonic restoration .......................................... 10.4. Summary................................................................................................... Acronyms ........................................................................................................... References .........................................................................................................
277 278 278 279 283 285 285 285 287 287 288 291 292 293 294 294 294
This chapter is organized into three parts. Part one covers the historic evolution of photonic networksdsome background for understanding how the industry got to where it is today. The second section explains requirements for optical amplifiers used in today’s photonic mesh networks. The third section looks ahead into the future: how the networks may evolve and the requirements for optical amplifiers to support such future networks.
10.1 PHOTONIC NETWORK EVOLUTION Photonic networks will continue to evolve as technology advances and bandwidth demands drive needs for higher transmission capacity at lower costs. With this evolution comes a variety of challenges, both technical and operational. Every Optically Amplified WDM Networks. DOI: Copyright Ó 2011 Elsevier Inc. All rights reserved.
277
CHAPTER 10 Challenges for long-haul dynamic Networks
carrier has its own strategy to meet the demands at the best cost points, and choosing a path forward often depends on the marketplace, the technical alternatives, the willingness to be an early adopter and take risk, and the need for scale. From a technical perspective, amplifiers and their performance are becoming more and more important in meeting the increased demands at higher total system capacities.
10.1.1 Data rate evolution Data rates have increased over time and will continue to increase as the bandwidth demand increases. There is continuous pull between riding the cost curve at a lower bandwidth and jumping to a new, higher bandwidth to accommodate growth and enable new services. Usually, the lower bandwidth offers commodity pricing and high manufacturing volumes, whereas the higher bandwidth may be needed to fulfill the service needs. Today, traffic on carrier networks is still growing by 30% to 60% per year, driven by growth in internet protocol (IP) traffic [3]. This will lead to a network adoption of 100 G per wavelength by 2011.
10.1.2 WDM evolution As the channel data rate is increasing, having more and more wavelengths on a single fiber enables cost improvements and capacity growth (Figures 10.1 and 10.2). Wavelength division multiplexing (WDM) systems have grown from the original two-channel systems of the 1980s to 80-plus channels today. Often it is possible to accommodate even more channels through closer spacing or amplification over a wider spectral range, but with these come trade-offs in reach, cost, and Data rate evolution
Single channel data rate (Gbit/s)
278
100
10
1
0 .1
0 .0 1 1975
1980
1985
1990 1995 Year
FIGURE 10.1 Single channel commercial systems line rate over time
2000
2005
2010
10.1 Photonic network evolution
number of wavelength (any bitrate)
WDM evolution 100 90 80 70 60 50 40 30 20 10 0 1985
1990
1995
2000 Year
2005
2010
2015
FIGURE 10.2 Number of wavelengths over time for commercial systems (any bitrate)
operational issues that must be accounted for. Due to the nonlinear interaction of wavelengths on the optical fiber, closer spacing of wavelengths comes at a penalty to the overall reach, so each carrier has to decide what its typical reach needs to be before determining the ultimate fiber capacity and channel spacing.
10.1.3 Network architecture evolution Network architectures have evolved greatly in the 20-plus years that dense wavelength division multiplexing (DWDM) systems have been deployed. Early systems were point-to-point with terminals at each end. Transponders are used to condition the signal, apply the required overhead and use the appropriate wavelength for the DWDM transmission (Figure 10.3). The optical signal from the end user device (router, Ethernet switch, synchronous optical network [SONET] multiplexer) is received over a short reach client interface, the signal is framed, forward error correction information added and transmitted via a long-reach wavelength signal into the line. For a simple single-channel system, the transponder enables the information to be transmitted over a fiber for a moderate distance (w40 to 80 km) without the need for electrical regeneration (Figure 10.4). When longer distances were needed on these single-channel systems, transponders could be placed back to back to regenerate the signal. In the mid-1980s, these single channel systems were commonly used for enabling fiber optic transmission in carrier networks (Figure 10.5). As bandwidth needs grew above the single channel rates, there was a race between higher speed electronics and the potential for multichannel systems. Early WDM systems arose when the bandwidth need outstripped the electronic
279
280
CHAPTER 10 Challenges for long-haul dynamic Networks
definition
Short reach Optics
Framer/ FEC
Long reach Optics
Transponder
FIGURE 10.3 Definition of a transponder
Single wavelength, no amp Outside Plant Transponder
Transponder
Fiber FIGURE 10.4 Simple single channel transport system (1980s)
Single wavelength, no amp, network
Outside Plant n Transponder
Outside Plant Transponder
Transponder
Transponder
Fiber
Fiber Intra-office Fiber
Physical a building A
Physical building B
FIGURE 10.5 Network view of single channel transport system (1980s)
Physical building u C
10.1 Photonic network evolution
development and low channel count systems emerged. These systems were still limited by the distance between expensive regeneration points (Figure 10.6). The main enabler for the DWDM systems we have today was the emergence of the optical amplifier in the early 1990s. For the first time it was not necessary to regenerate the signal at such closely spaced intervals, and the entire payload could go five or more amplified spans. Early systems were limited by the quality of the components available, but systems with eight to 16 wavelengths of OC-48 were normal. Still, at this point it was necessary to fully demultiplex the multichannel signal to regenerate at an optical terminal. This required terminals at every point where add or drop was needed, even if some of the channels did not require signal regeneration at that point (Figure 10.7). To avoid costly and unnecessary regeneration of pass-through wavelengths, some systems allowed for optical pass-through from demultiplexer to multiplexer, via jumper cables. This required a great deal of manual effort, because each terminal site had to be hand-cabled to set up a wavelength at the through locations. In the early 2000s, commercial reconfigurable optical add/drop multiplexer (ROADM) systems started to emerge [1] (Figure 10.8). The early systems had 1x2 switches that allowed a wavelength to be passed through or dropped via a remote command at intermediate sites. For the first time it was possible to limit the “touch” required to set up a wavelength to only the endpoints or regenerator locations, and not every site that any wavelength on the system had to stop for add/drop. At the same time, tunable lasers were becoming more common, allowing a single
WDM Wavelength 1 . . . Wavelength n/2 . . .
Wavelength 1
W D M M U X
W D M
Outside Plant Fiber
D E M U X
Wavelength n
FIGURE 10.6 Wavelength division multiplexing transport system (late 1980s)
. . . Wavelength n/2 . . . Wavelength n
281
CHAPTER 10 Challenges for long-haul dynamic Networks
Amplified WDM Wavelength 1
Wavelength 1 . . . Wavelength n/2 . . .
W D M M U X
W D M
Outside Plant Fiber
Amplifier
Outside Plant D E M U X
Fiber
Wavelength n
. . . Wavelength n/2 . . . Wavelength n
FIGURE 10.7 Amplified WDM transport system (early 1990s)
Simple Optical Add and Drop Wavelength 1 . . .
Wavelength 1 . . .
OSP Fiber
OSP Fiber
. . . Wavelength n
Wavelength n
Wavelength n Wavelength n
282
Local Add/Drop
FIGURE 10.8 WDM transport system with simple MUX/DEMUX optical add/drop (late 1990s)
transponder to serve a spectrum of wavelengths. Technologies such as wavelength blockers and wavelength selective switches further drove down the cost and complexity of ROADM systems, making the ROADM the architecture of choice for modern DWDM networks (Figure 10.9). While 2-degree ROADMs enabled through wavelengths to be expressed through a drop site, many nodes had 3 or more degrees of traffic, i.e. three or more fiber pairs
10.1 Photonic network evolution
ROADM Wavelength 1
Wavelength 1 . . .
OSP
OSP ROADM
Fiber
Fiber
Wavelength x
Wavelength n Wavelength x
Wavelength n
. . .
Any wavelength, any number of Drop or Add
FIGURE 10.9 WDM transport system with ROADM (early 2000s)
meeting at the node. For early ROADM systems, the best that a carrier could do was to position the ROADM in the most advantageous direction, then for the other directions the traffic needed to be dropped and regenerated through a terminal. As ROADM technology evolved, architectures that allowed N > 2 degrees of optical express became commonplace [2]. Early systems were limited to 1x5 or 1x9 switches supporting the architectures, depending on the channel count and spacing. Today’s systems are evolving to support higher and higher degrees, allowing sites to have multiple rails of DWDM in each direction with optical pass-through (Figure 10.10). Given the advances in technologies, today full photonic mesh networks are being built. N-degree ROADMs at each node in the mesh allow any channel to be added, dropped, or expressed anywhere in the network (Figure 10.11). This is the kind of flexible network carriers want: one that does not require special planning or wavelength-specific equipment and that allows the network to grow as technology evolves. For example, as the transponder reach improves, it is now possible to express through sites that may have been drop sites for earlier transponders, continuously improving cost and avoiding network touch points.
10.1.4 Amplifier evolution From the early 1990s to about 1999, optical amplifiers improved their performance rapidly (usable bandwidth, flatness, output power). As the number of WDM channels increased significantly, system designers learned to take many new kinds of effects into account, such as stimulated Raman scattering (SRS). Around 1999, as 10 G direct detection started to move into networks, suddenly the 10 G chromatic dispersion limit required in-line chromatic dispersion compensation modules (DCM). To overcome the loss of these modules, amplifiers with mid-stage access
283
CHAPTER 10 Challenges for long-haul dynamic Networks
OSP Fiber
OSP N-degree ROADM
Fiber
Amplifier
Wavelength x
Amplifier
Fiber
OSP
Amplifier
Multi-degree ROADM
Wavelength x
284
Any wavelength, any number of Drop or Add
FIGURE 10.10 Multidegree ROADM node with add/drop on bottom (after 2005)
Photonic Mesh Network
N-degree ROADM
N-degree ROADM
N-degree ROADM
N-degree ROADM
FIGURE 10.11 Photonic mesh network (2006 and later)
N-degree ROADM
10.2 Requirements for amplifiers in today’s photonic mesh networks
were introduced. With the advent of 40 G coherent transmission in commercial networks in 2008, these DCM modules became unnecessary again for such systems, enabling the industry to go back to an amplifier design without mid-stage access for 40 G and 100 G coherent networks. Raman amplifiers have been deployed primarily to overcome long-reach spans or to give an added channel enhanced reach at an added cost. The additional operational complexities (e.g., fiber cleanliness and eye safety) and cost (Raman pump conversion efficiency is much smaller than EDFA pump conversion efficiency) need to be considered in any network design. Raman amplifiers may be a tool in the future to improve the optical signal-to-noise ratio (OSNR) of transmission at data rates beyond 100 G.
10.1.5 Status quo AT&T has deployed a full photonic mesh network in its backbone. The cartoon in Figure 10.12 shows how this network interconnects the key backbone routers with high-capacity 40 G wavelengths. By using a photonic mesh as described in the previous sections, AT&T has been able to grow the network from 10 G to 40 G and through a second generation of 40 G technology. At each step we have been able to get the best reach and cost without the need to make any changes in the photonic layer (amplifiers and ROADMs). There are no terminals in this network, and each node is capable of full add/drop/express [3]. For AT&T, the main driver to go from 10 G to 40 G was IP data growth. There were many other reasons, such as cost and scalability. Ultimately, these same drivers will push us to go to 100 G in the backbone.
10.2 REQUIREMENTS FOR AMPLIFIERS IN TODAY’S PHOTONIC MESH NETWORKS Design and deployment of a full photonic mesh network (or even islands of photonic mesh) involves careful design and control. Wavelengths can stop and start at any point in the network, and this can lead to unwanted interactions between channels and propagation of effects across the network in unintended ways. The system designer needs to account for these optical effects up front in any design, and the system needs to be modeled and tested to ensure proper operation in the field.
10.2.1 Optical link control Modern transmission systems have a very sophisticated optical link control compared with the systems of the 1980s and 1990s. This is driven, on the one hand, by the availability of lower cost memory and processors, and on the other hand by the need for higher flexibility and performance. The system usually consists of a fast (on the order of tens or hundreds of microseconds) constant gain optical amplifier control loop and possibly a gain tilt
285
286
CHAPTER 10 Challenges for long-haul dynamic Networks
Seattle
Chicago
Cleveland
Salt Lake City San Francisco
New York Kansas City
Denver
Los Angeles
Herndon
Atlanta Dallas
OC-768 capable DWDMs Backbone Router
Houston Miami
FIGURE 10.12 AT&T photonic mesh network (2006)
control loop for the individual amplifier [4] and a slower (on the order of seconds) control loop to keep the optical link in an optimum state (Figure 10.13). While the goal of the fast control loop is to keep the impact of any failures or fiber cuts on service to a minimum, the slow control loop continuously optimizes the link for best transmission performance, making sure the launch spectra and launch powers after every ROADM and amplifier are as close to the set points as possible. For long-haul transmission, these control loops will be a fundamental part governing the performance of the system. The exact requirements of these control loops will be part (among other things) of the margin allocation for the overall transmission performance. The system designer can trade off certain parameters in the margin allocation to relax the requirements on these control loops. For example, if the amplifier tilt control would require a tilt control device in every amplifier, but the designer wants to save system cost and place that expensive device in only every nth amplifier, a worst case gain tilt penalty can be introduced in the margin allocation. This gain tilt penalty will reduce the reach for certain channels, but lower the overall system cost. For ultra-long-haul (ULH) transmission systems fewer trade-offs will be possible due to the higher optical performance requirement. There are usually local control loops in every amplifier, as well as a separate control loop for every ROADM-to-ROADM link section. The local control loops attempt to keep the amplifier operating parameters such as gain, flatness, etc. in a certain set range, whereas the ROADM-to-ROADM control loop gives commands to the local control loops based on the increased network visibility. For example, if in one ROADM-to-ROADM section, any fiber span has an increased loss (e.g., a large bend loss due to a cable pinch), the local control loop in the following amplifier will
10.2 Requirements for amplifiers in today’s photonic mesh networks
ROADM to ROADM link control loops
N-degree ROADM
Amplifier
Local control loops
Amplifier
N-degree ROADM
Local control loops
FIGURE 10.13 Elements of optical link control (amps, ROADMs)
try to compensate for the added loss, while the ROADM-to-ROADM control loop may increase the launch power from the previous amplifier to help compensate for the larger loss. While these cases are not everyday occurrences in a network, having an intelligent optical infrastructure that keeps the network running as long as possible is a key requirement for carriers.
10.2.2 Static amplifier requirements Based on the description of today’s photonic mesh network, due to the number of concatenated amplifiers, static amplifier performance is key to the overall network performance. Low noise figure, flat gain and low gain ripple, as well as low polarization dependent gain are the most important parameters in a photonic mesh network [5]. In addition, manufacturing repeatability is also very important for being able to model a long chain of amplifiers correctly to design links over the network.
10.2.3 Dynamic amplifier requirements Dynamic amplifier requirements in a photonic mesh network are also important for protecting service from any dynamic events that can happen in a network. For example, if a fiber cut or an amplifier failure occurs in front of a ROADM, the other directions of the ROADM should not be impacted due to the drop in channels and therefore change in loading of the amplifier chains [7,8]. Today’s amplifiers usually contain a control loop to keep the optical gain of the amplifier constant. In a photonic mesh network, due to the impact of fast changes on spectral loading, a way to keep the amplifier flat with a fast gain tilt compensation is also critical [6]. Also, certain impairments can become intensified due to fast changes in the spectral loading of an amplifier (e.g., spectral hole burning) and lead to the need for additional stabilization of the amplifiers during dynamic events [9]. All these control loops need to be designed so that they are unconditionally stable in a network environment, as there are potential feedback paths in a photonic mesh network, where changes in one ROADM-to-ROADM section can propagate through perturbations caused by the changes in other sections.
287
288
CHAPTER 10 Challenges for long-haul dynamic Networks
10.3 THE FUTURE: REQUIREMENTS FOR A FULLY DYNAMIC PHOTONIC MESH All of the photonic networks that have been discussed above still have one limitation: the transponder has to be cabled to a particular channel port of the multiplexer or demultiplexer, which is then cabled to a particular direction at a multidegree node. Even though the laser on the transponder can be tuned to any of the International Telecommunication Union-grid wavelengths, it is connected to a port that is expecting a particular wavelength number. So for an electrically switched network, it is possible to precable the end user to the network and commit the bandwidth later once it is determined where the service needs to go (also reconfigure and move that bandwidth on the fly if changes are required). For photonic networks, however, the end user is committed to a particular path as soon as the transponder is connected to the system. To avoid this, carriers deploy transponders only when the service is ready to be committed; then provisioning requires a cabling operation at each end and at any regeneration points in the middle. Technologies have existed for a number of years that would allow the photonic network to behave more like an electrically switched one, making the on and off ramps colorless, able to add and drop any wavelength to any port (colorless), and the able to switch the transponder to any of the directions out of a node (non-directional/ steerable). This gives the carrier the ability to pre-cable the transponders to the router or Ethernet switch, then commit the bandwidth, as needed, through remote commands or an optical control plane. Unfortunately these technologies for colorless and non-directional transmission still remain expensive and immature for high-bandwidth and high-capacity applications. But suppliers are starting to offer options for reconfigurability, which are more attractive, and it is only a matter of time until these systems become commonplace in carrier networks. Figure 10.14 shows an early implementation of a two-degree ROADM with a fixed add/drop. In this figure, the wavelengths are split off and dropped, and transponders are connected to ports that are expecting the dropped signals. A wavelength blocker is used to selectively stop the dropped channels from being transmitted on the through path. On the add side, wavelengths are again cabled to ports on the channels that are being added, and those are combined with the through channels to be sent to the next downstream ROADM. Now consider the same architecture with tunable add/drops (Figure 10.15). The transponders can be cabled to any port on the multiplexer or demultiplexer, because it is now possible to drop and add any wavelength to any port. It is no longer required to connect a transponder to a particular port. So if channel 1 is required to be dropped at this site, it would be possible to tune the demultiplexer to receive channel 1 on port M. The blocker would prevent channel 1 from being passed through to the add section. Likewise, on the add side, channel 1 could again be added by tuning the laser to channel 1 into the tunable add multiplexer. This could all be done remotely,
10.3 The future: Requirements for a fully dynamic photonic mesh
Broadcast and select with fixed local add and drop
. . .
Wavelength 1
Wavelength x
Wavelength x
coupler
Wavelength Blocker
. . .
Wavelength 1
coupler
drop
add
FIGURE 10.14 Per-direction fixed optical add/drop
Broadcast and select with tunable local add and drop coupler
coupler
Wavelength Blocker
Tunable drop FIGURE 10.15 Per-direction tunable optical add/drop
Wavelength x
. . .
Wavelength m
tunable
Wavelength x
. . .
Wavelength m
tunable
Tunable add
289
CHAPTER 10 Challenges for long-haul dynamic Networks
N-degree ROADM with fixed nondirectional add/drop
Local add/drop 1
Wavelength x
. . .
Wavelength 1
Wavelength x
. . .
Wavelength 1
N-degree ROADM
Local add/drop 2
FIGURE 10.16 Multidegree ROADM with fixed optical add/drop
N-degree ROADM with tunable non-directional add/drop
N-degree ROADM
Tunable add/drop 1 FIGURE 10.17 Multidegree ROADM with tunable optical add/drop
Wavelength x
. . .
Wavelength m
tunable
Wavelength x
. . .
tunable
Wavelength m
290
Tunable add/drop 2
10.3 The future: Requirements for a fully dynamic photonic mesh
as long as the transponders were cabled to the tunable multiplexers/demultiplexers without the need for a truck roll. The next challenge comes because the channel add and drop is tied to a particular direction. For a two-degree node this is not a problem, but for a multidegree node this requires committing pools of transponders to each direction. Ideally, a carrier would like to pre-connect a transponder and then remotely steer it in any direction out of the node [10]. To do this, a switching mechanism is required to direct the output of the tunable transponder to the desired line direction of the multidegree ROADM (Figures 10.16 and 10.17). Considering the future photonic mesh networks and the requirements of optical amplifiers that can fulfill these architectures, we find three cases of use of photonic mesh technology with varying requirements on the optical amplification chain. The first case will be where the flexibility in the network is mainly used for service provisioning (time scale of hours/minutes), the second case will be re-provisioning after catastrophic failures (time scale of minutes/seconds), and the third case the full photonic mesh restoration (time scale of sub-second).
10.3.1 Provisioning through pre-cabling In an old-style network with manual add/drop, provisioning a wavelength over a large geographic area could take months, due to the many fiber jumpers that had to be
pre-cabling Turn up this path if needed
N-degree ROADM
N-degree ROADM
his
or t
N-degree ROADM
path if ne d
ede
N-degree ROADM
FIGURE 10.18 Pre-cabling at the end points for fast provisioning
N-degree ROADM
291
292
CHAPTER 10 Challenges for long-haul dynamic Networks
manually placed with a site visit. Modern photonic mesh networks reduce this provisioning time dramatically, since the ROADM sites can be provisioned remotely. Further improvement can be achieved, with additional up-front cost, by using tunable non-directional add/drops instead of fixed wavelengths. Carriers could also have precabled transponders to customers and regenerator-banks inside the network and turn up a wavelength without having to touch the network. This model is already well known from electrical cross-connects, where an extra bandwidth pool is kept reserved to turn up service quickly through point and click, without the need to first install any new hardware (Figure 10.18). The extra bandwidth could be, for example, an extra wavelength in every cross section to support new customer additions. Since the time scale of provisioning in this context is hours or minutes, amplifiers do not need to meet any more stringent requirements than today’s photonic mesh networks. The major improvements needed for this architecture are in higher port count wavelength selective switches, which are in development to improve the economics. We expect to see such architecture in a few years in all major carrier networks.
10.3.2 Simple photonic restoration techniques Another step toward greater flexibility and functionality in the photonic layer is the slow re-provisioning after catastrophic failures, like a fiber cut of amplifier failure. A large number of wavelengths will be affected that need to be re-routed through the network (Figure 10.19). The restoration speed in this case will be on the order of minutes or seconds, making sure that the existing service is not impacted by massive change in the loading of the amplifiers. To achieve this, the reconfigurations and channel turn-ups will be ramped sufficiently slowly that the amplifier control loops
Re-provisioning
N-degree ROADM
N-degree ROADM
N-degree ROADM Slow Re-provisioning
N-degree ROADM
FIGURE 10.19 Simple photonic restoration through re-provisioning
N-degree ROADM
10.3 The future: Requirements for a fully dynamic photonic mesh
will be able to follow. This form of restoration is not suited for sub-second guaranteed restoration, but more for a slower best-effort service (like, e.g., web traffic). Still, being able to restore best-effort traffic can reduce the capacity (and therefore cost) in, for example, the IP router layer, as any failure can be restored within minutes, which may be good enough for certain types of traffic. A tiered system could also make sure that more important real-time traffic gets restored first within seconds and the less important best-effort traffic later. For this case, amplifiers and link control loops need to be able to suppress large changes in channel loading. The re-provisioning algorithm can reduce some of these requirements for example by time staggering at the cost of restoration time.
10.3.3 The holy grail: photonic restoration The third case is full fast photonic mesh restoration with sub-second time scales depending on the application (Figure 10.20). For most modern services, around 100 to 200 millisecond switching time is tolerable. Translating that requirement into the photonic layer means that most elements in today’s transmission system need to be completely re-designed. For example, laser switching speeds of a few milliseconds are required, as well as wavelength selective switches that can re-configure all the channels within tens of milliseconds (today’s devices have a limited control that can re-configure one channel at a time within tens of milliseconds). Transponders would have to be synchronized within milliseconds. Similarly tunable chromatic dispersion elements like 40 G temperature tuned chromatic dispersion devices or coherent signal processing filter adjustments would need to be adjusted within milliseconds. Moreover, the amplifier chain would have to be extremely stable against significant variations in channels in a few milliseconds and still keep the existing channels error free.
Photonic Mesh Restoration
N-degree ROADM
N-degree ROADM
N-degree ROADM
FIGURE 10.20 Full photonic mesh restoration
N-degree ROADM
N-degree ROADM
293
294
CHAPTER 10 Challenges for long-haul dynamic Networks
The scenario outlined here is still years away, but there is a lot of research going on to ultimately make it happen.
10.4 SUMMARY Optical networks have evolved and continue to evolve in ways that are more and more challenging for the system designers. In large photonic mesh networks, the amplifiers must account for both static and dynamic behavior of the transmission line and accommodate changes without disturbing the traffic quality of service. As more traffic at higher bit rates is placed on these networks, survivability in the face of upstream failures becomes even more important, and the design and control can differentiate one solution from another. Carriers are looking for solutions that have been designed and tested against all possible failure scenarios and can handle network additions and deletions seamlessly. Finally, it is desirable to have a true photonic mesh-restorable network where wavelengths can be treated the same way as the electrical bandwidth and protected in a shared way against catastrophic failure.
ACRONYMS DCM DEMUX DWDM IP ITU MUX OSNR ROADM SONET SRS ULH WDM
Dispersion compensation modules Demultiplexer Dense wavelength division multiplexing Internet protocol International Telecommunication Union Multiplexer Optical signal-to-noise ratio Reconfigurable optical add/drop multiplexer Synchronous optical network Stimulated Raman scattering Ultra-long-haul Wavelength division multiplexing
References [1] J. Berthold, A.A.M. Saleh, L. Blair, J.M. Simmons, Optical Networking: Past, Present, and Future, Journal of Lightwave Technology 26 (9) (2008) 1104e1118. [2] A.A.M. Saleh, Overview of MONET project, Lasers and Electro-Optics. CLEO/ Pacific Rim ’97., Pacific Rim Conference, July 14e18, (1997) pp. 12e12
References
[3] K. Tse, AT&T’s Photonic Network. Optical Fiber Communication/National Fiber Optic Engineers Conference. February 24e28, (2004) pp. 1e6. [4] P.M. Krummrich, M. Birk, 2004. Compensation of Raman transients in optical networks. In: OFC 2004, paper MF 82. [5] P. Wysocki, V. Mazurczyk, Polarization dependent gain in erbium-doped fiber amplifiers: computer model and approximate formulas, Journal of Lightwave Technology 14 (4) (1996) 572e584. [6] P.M. Krummrich, M. Birk, Experimental investigation of compensation of Ramaninduced power transients from WDM channel interactions, IEEE Photonics Technology Letters 17 (5) (2005) 1094e1096. [7] A.K. Srivastava, Y. Sun, J.L. Zyskind, J.W. Sulhoff, EDFA transient response to channel loss in WDM transmission system, IEEE Photonics Technology Letters, 9 (3) (1997) 386e388. [8] Y. Sun, A.K. Srivastava, J.L. Zyskind, J.W. Sulhoff, C. Wolf, R.W. Tkach, Fast power transients in WDM optical networks with cascaded EDFAs, Electronics Letters 33 (4) (1997) 313e314. [9] G. Luo, J.L. Zyskind, J.A. Nagel, M.A. Ali, Experimental and theoretical analysis of relaxation-oscillations and spectral hole burning effects in all-optical gain-clamped EDFA’s for WDM networks, Journal of Lightwave Technology 16 (4) (1998) 527e533. [10] S.L. Woodward, M.D. Feuer, J. Calvitti, K. Falta, J.M. Verdiell, A High-Degree Photonic Cross-Connect for Transparent Networking, Flexible Provisioning and Capacity Growth. 32nd European Conference on Optical Communications. ECOC2006, paper Th1.2.2, (2006) Cannes, France.
295
This page intentionally left blank
CHAPTER
11
Transport Solutions for Optically Amplified Networks
Werner Weiershausen*, Malte Schneiders *
Luxdyne Ltd., Germany
CHAPTER OUTLINE HEAD 11.1. Introduction .............................................................................................. 11.2. Optical transport networks ......................................................................... 11.2.1. Transport and aggregation networks ......................................... 11.2.1.1. Access networks ........................................................... 11.2.1.2. Metro aggregation networks .......................................... 11.2.1.3. Backbone transport networks ........................................ 11.3. Signal degradation and temporal fluctuations.............................................. 11.3.1. Residual chromatic dispersion ................................................ 11.3.2. Polarization-mode dispersion .................................................. 11.3.2.1. Temporal properties of PMD ......................................... 11.4. Raman amplification in WDM networks ....................................................... 11.4.1. EDFA, Raman and hybrid amplification scenarios ..................... 11.4.2. Laser safety and network implementation issues ....................... 11.5. Summary................................................................................................... Abbreviations...................................................................................................... References .........................................................................................................
297 298 299 301 303 307 313 314 317 320 323 323 332 334 335 338
11.1 INTRODUCTION The chapter gives an overview of state-of-the-art optical transport networks (OTNs) that are introduced by the carriers to adequately address future broadband service aggregation and transport and the growing packet centric traffic. The dramatic traffic increases and special requirements to support new services cost efficiently are driving the evolution of new enabler technologies at all engaged network layers. This chapter mainly focuses on the optical layer encompassing modern OTN/optical transport hierarchy (OTH) architectures and its physical layer, i.e., wavelength division multiplex (WDM) system requirements and approaches from the network point of view. The following section provides an overview of some basic network architectures, which are being developed for national networks in Europe. They will represent Optically Amplified WDM Networks. DOI: 10.1016/B978-0-12-374965-9.10011-1 Copyright Ó 2011 Elsevier Inc. All rights reserved.
297
298
CHAPTER 11 Transport solutions for optically amplified networks
cost-efficient solutions for converged network platforms that carry both circuit and packet switched traffic. Special attention in this section will be on the optically transparent island solutions. The following part will outline some major physical layer problems that limit WDM transmission distances and the size of reconfigurable optically transparent islands, particularly for high-speed channels. In 40 Gbit/s and 100 Gbit/s systems, polarization-mode dispersion (PMD) represents a fundamental limitation for transmission. Furthermore, temporal fluctuations of chromatic dispersion (CD) and PMD require sophisticated and robust transceiver solutions. Some basic approaches, such as adaptive equalization and robust optical modulation formats, are covered in this part. The last part describes the application of different Raman amplification schemes for long-distance links including those for reconfigurable optical add-drop multiplexer (ROADM) based transparent network islands. Distributed and lumped Raman, and hybrid amplification schemes (combination of Raman and erbiumdoped fiber amplification) are introduced, and high power safety related aspects are discussed.
11.2 OPTICAL TRANSPORT NETWORKS In this section we focus on optical network architectures of the modern successors to the first-generation dense wavelength division multiplex (DWDM) networks that were based on simple point-to-point links with electrical termination at each node. Telecommunication services are undergoing fundamental changes due to the dominance of packet switched traffic over the circuit switched traffic. For residential customers, new broadband services such as new video applications are gaining momentum; for enterprises it’s virtual private networks. The enormous growth in bandwidth demand and the increasing need for service transport flexibility, while controlling the costs requires the whole telecommunications industry to optimize transport efficiency of the core and aggregation networks for functionality, flexibility, and scalability. These new requirements of telecommunication networks lead to an overall platform design with strongly interacting parameters from different network layers, local network domains, various service classes, dynamic traffic patterns, and temporal migration scenarios. This makes the optimization of network architectures including the optical layer much more complex when compared with the DWDM networks from the first generation. In legacy networks, the demarcation between their open systems interconnection (OSI) layers allows for a relatively simple individual layer based optimization of traffic engineering, network management and network costs. The optical layer of legacy networks mainly consists of transparent point-to-point links, which are terminated by the terminal multiplexers and transceivers at each switching node, excluding traffic branching within the optical multiplexers or line sections. All traffic switching and routing is done in the electrical domain, e.g., by synchronous optical network (SONET) / synchronous digital
11.2 Optical transport networks
hierarchy (SDH) based cross connects. Some optical add-drop multiplexers (OADM) have been deployed that allow simple all-optical bypassing of the traffic at nodes and permit only a certain percentage of traffic to be dropped. The new network architectures, related to enhanced flexibility and optimized costs, demand common optimization of different entangled protocol layers and a complex physical layer plane. To reduce cost, remotely configurable add-dropmultiplexers (ROADM) are being introduced into the networks so that the traffic switching is no longer restricted to the electrical domain. All-optical switching of WDM channels by ROADMs enables network operators to reduce their capital expenditure (CapEx) and power consumption by avoidance of optical-to-electricalto-optical (OEO) conversions. Unlike the older-generation optical add drop multiplexers (OADMs) with fixed filters, ROADMs can be switched remotely, thus leading to lower operational expenditure (OpEx). In modern transport networks, colorless and directionless ROADM nodes with degrees higher than two can be applied as photonic cross connects (PXC), which allow for low cost traffic routing of wavelength channels. This network approach can, however, benefit only from traffic granularities available in DWDM systems. Data rates at 10 Gbit/s, 40 Gbit/s, and 100 Gbit/s are seen as the wavelength channel bandwidths that will be efficiently transported and switched through the first generation of all-optical meshed networks. In second generation systems, flexible bandwidth support may be possible by techniques like bandwidth adaptable multilevel quadrature amplitude modulation (QAM) formats, sub carrier multiplexing (SCM), or optical orthogonal frequency-division multiplexing (OFDM) for an efficient exploitation of the bandwidth-distance resources near the Shannon limit. The extension of today’s networks from point to pointebased DWDM links toward more seamless meshed all-optical networks, the deployment of new modulation formats such as coherent digital signal processor (DSP)esupported polarization multiplexed quadrature phase shift keying (PM-QPSK), and more flexible bandwidth assignment schemes have consequences on the design of amplification maps and the use of certain types of lumped or distributed optical amplification.
11.2.1 Transport and aggregation networks Network operators have the common basic target to produce cost-efficient telecommunication services. When considering operators from different nations including carriers operating worldwide, a variety of network architecture designs need to be considered. The suitable network design depends on the individual national properties with respect to the telecommunication services to be provided, such as the local population density distributions, the characteristic local residential consumer behavior (e.g., demand for voice telephony, internet protocol [IP], or broadband TV) or the distribution and service level agreement (SLA) requirements of the business customers. The design of the networks is governed by the topology (e.g., ring, star,
299
300
CHAPTER 11 Transport solutions for optically amplified networks
mesh), by the purpose (access, aggregation, transport), by the mean and maximum link distances, and by the density and degree of switching or grooming nodes. All this has a direct impact on the choice of amplification in the optical multiplex section (OMS) of DWDM systems and on the local placement of optical amplifiers. The diameter of networks is one of the most obvious distinctions. Nationwide networks in the United States follow engineering rules different from those applicable to the national backbones in European countries (see Figure 11.1), especially when the design of amplifier maps and the positioning of photonic cross connect (PXC)/ROADM based nodes are considered. The largest diameters within alloptical transport is achieved in submarine cable networks that deploy lumped amplifier span designs with very short distances between adjacent EDFAs and eventually supported by additional distributed Raman amplification. Besides the distance, many other parameters influence decisions for special network layouts, e.g., the local distribution of population and industry to be connected, the traffic patterns and capacity evolution, the telecommunication service kinds and classes, and much more. Also, the deployment choice of lumped inline amplifiers, distributed Raman amplification or hybrid schemes, gain equalizing devices, electrical or optical inline regenerators, and electrical grooming nodes or optically amplified multidegree ROADM nodes is strongly dependent on these multiple factors. In the following part, some network options with consequences for optical amplifier applications will be described against the background of European national networks. Here a variety of requirements force operators to select many different network architectures for different local domains with suitable primary foci to meet optimum transport efficiency and operational performance. The present trend is to consolidate different network domains into a converged platform to simplify the overall network management process.
FIGURE 11.1 A comparison of characteristic link statistics of reference networks in North America and Europe. The distribution function with peak at 600 km belongs to pan-European backbone networks, the other one to backbone networks in North America (the small left satellite peak in the European distribution is not real; it is due to the applied fit algorithm only).
11.2 Optical transport networks
FIGURE 11.2 Schematic of German core network (public reference network [1]) with a table showing summary of the topological network characteristics [1]
European networks cover many scenarios of possible architectures, for ultralong-haul (ULH) pan-European backbone to national European backbone, metro, and access networks. The typical distance characteristics of link lengths between major backbone nodes for North America and pan-European networks are compared by Figure 11.1. In North America the geographical dimensions of backbone networks require ULH systems. The same is applicable in pan-European networks, but the distances are significantly shorter. The backbone links of national networks of the different European states like Germany are even shorter. Figure 11.2 gives characteristic values for a public German reference network. Here the mean fiber link distance between major cities and thus backbone nodes is about 400 km which could be still called “metro.” However, as for the next generation architecture it is intended to intensively apply optically transparent transit nodes (ROADM/PXC), future national networks will also demand systems with a longer reach. In the following sub-sections we will focus on typical modern intranational European network architectures. Future converged telecommunication platforms will comprise access, aggregation, and transport networks. Their design rules depend on their primary purpose: either traffic aggregation or distribution from and to customers, or the transport and routing of large amounts of combined capacity.
11.2.1.1 Access networks Today’s residential access networks deliver services for the mobile and fixed-line telecommunication with the latter dominating in overall traffic capacity. Different
301
302
CHAPTER 11 Transport solutions for optically amplified networks
digital subscriber line technologies (DSL) such as asymetric DSL (ADSL) and very high bit rate DSL (VDSL), demand up to 50 Mbit/s peak traffic per subscriber. DSL backhaul is already realized by fiber-to-the-cabinet (FTTC) solutions, where cabinets are placed in the streets of urban and sub-urban regions. Besides this, fiber-tothe-building (FTTB) and fiber-to-the-home (FTTH) deployments have started in Europe and will secure further bandwidth growth driven by future consumer broadband services. While the carriers migrate from DSL to FTTB/H, they will try to reduce the operational expenditure, which is especially important for the incumbent European carriers. As a consequence, a lot of central offices will eventually be closed. Typical scenarios show a trend toward closure of 90% of the sites, meaning only 10% of the active sites will remain. The extended distance that has to be bridged between the FTTx access nodes and the remaining 10% of active central office sites for backhauling leads to severe problems for network operators when using passive optical network (PON) solutions, which employ gigabit PON (GPON) and Ethernet PON (EPON) standards with tree-type topology. The problem is a combination of the extended distances, greater than 40 km, and a high PON splitting ratio to provide coverage to a large area with many distributed homes and buildings around the little central offices. In addition, when combined with high bandwidth demand, the long distances and high splitting ratios cannot be handled with the classic GPON/EPON technologies. Amplifier extenders can extend the power budget of PON networks with high splitter and fiber transmission losses. These extenders are positioned in semi-active sites using simple outdoor cabinets with a small footprint, without air conditioning or personnel. This way the desired operational cost reduction can be achieved. However, even single-wavelengthebased PON tree networks will not be able to shoulder the whole problem space. An alternative will be colored systems like WDM-PON, or hybrid systems like colored time division multiple access (TDMA)e PON channels on a WDM system. These WDM-PON systems have to be amplified by optical amplifiers (EDFA). Generally, the EDFA is followed by a WDM splitter, e.g., a demultiplexer (DEMUX). It is important that both DEMUX and EDFA can be put in outdoor cabinets without air conditioning. They have to be very tolerant of temperature variations; consequently, athermal arrayed waveguide grating (AWG) filters are used for DEMUX filters. Besides the star and tree topology for WDM-PON, ring topology is also a possibility. WDM rings with fixed optical add-drop multiplexers (OADM, fixed OADM [FOADM], simple filters) can be used to collect the traffic from different access sites and to drop it at a common hub node (hub&spoke to the metro network). The advantage is efficient exploitation of the existing fiber plant, and these rings allow for simple 1þ1 or 1:1 protection by using both ring directions. As the cost of transport of transit traffic through the OADM nodes are relatively small, this is an interesting alternative to stars and trees. To support the ring attenuation budget, for this architecture optical amplifiers (EDFA) have to be used. They can be placed at the multiplexer/demultiplexer/OADM sites.
11.2 Optical transport networks
Low-cost EDFAs are preferred in relatively small access rings with comparably small optical signal-to-noise ratio (OSNR) degradation. Application of the very lowcost passive coarse wavelength division multiplex (CWDM) technology is not suitable because the distances are too long for purely passive rings. This argument, however, applies for the before-mentioned scenario with the strongly reduced number of central offices.
11.2.1.2 Metro aggregation networks The next important network domain of a hierarchical platform is the metro aggregation network. This network connects the regional active central offices to the core sites of a national backbone. The aggregation network has several tasks. First, it is used to aggregate all incoming traffic from the residential customers in the previously described access networks. The aggregation is carried out such that the bandwidth granularity with common destinations is large enough to efficiently fill wavelength channels of DWDM systems. The minimum reasonable channel bandwidth today is 10 Gbit/s (10.7 Gbit/s according to optical channel transport unit (OTU) 2 within the ITU-T G.709 standard for the optical transport hierarchy OTH). Second, the aggregation network can be used to connect network termination (NT) points from business customers to the backbone or other NT points within the same region. Business customers need not be served by the same network like the residential customers, so two parallel network platforms can be dedicated to both applications. While the traffic pattern of business and residential connections are generally very different, requiring a separate network platform definition for them, a trend toward common transport platforms is more suitable due to the uncertainty regarding the future traffic patterns and capacity evolution. Often the future bandwidth evolution, especially of business traffic, is not very distinct, so it is an interesting option to share major traffic platform between the residential and business customers. For the aggregation network, the following architectural alternatives can be seen as major solutions for European national networks.
11.2.1.2.1 Star and tree topologies If a metro network is exclusively dedicated to backhauling of traffic, such as DSL, FTTx or mobile backhaul, a star or tree topology can be applied. The problem is that WDM applications for fiber sharing are not very effective since the traffic is collected by the star from different locations; i.e., the amount of commonly transported traffic is low. Furthermore, a lot of fiber infrastructure is needed, especially if a double star or tree is needed for protection purposes. For intra-regional business traffic with an any-to-any traffic matrix, such networks are not suitable.
11.2.1.2.2 Ring topologies WDM rings with active FOADM or ROADM nodes can solve the previously described problems of star or tree topologies. They are suitable for both service
303
304
CHAPTER 11 Transport solutions for optically amplified networks
classes, residential and business, and the fiber infrastructure can be efficiently exploited by active DWDM (with amplification, e.g., by EDFA). Ring type networks are very suitable for residential traffic aggregation. The traffic pattern within the topological ring can still follow a logical star; i.e., DSL/ FTTx or mobile traffic is backhauled and condensed in a hub node before being transferred into the longer-distance transport backbone. A hub&spoke DWDM ring constellation with one hub OADM and a number of satellite OADM nodes provides good architecture for handling startype traffic pattern including a cost-efficient and easy to manage 1þ1 or 1:1 protection scheme. Protection can be carried out in different layers, for instance in the WDM layer (L1), which is directly concerned about the fast ring direction switching in