UNIVERSITY OF CALIFORNIA - UCSB · 2005. 11. 10. · Communication Conference (OFC´05), paper...
Transcript of UNIVERSITY OF CALIFORNIA - UCSB · 2005. 11. 10. · Communication Conference (OFC´05), paper...
UNIVERSITY OF CALIFORNIA
Santa Barbara
Optical Signal Processing Using Traveling-Wave
Electroabsorption Modulators
A Dissertation submitted in partial satisfaction of the
requirements for the degree Doctor of Philosophy
in Electrical and Computer Engineering
by
Hsu-Feng Chou
Committee in charge:
Professor John E. Bowers, Chair
Professor Kevin C. Almeroth
Professor Daniel J. Blumenthal
Professor Nadir Dagli
December 2005
The dissertation of Hsu-Feng Chou is approved.
Kevin C. Almeroth
Daniel J. Blumenthal
Nadir Dagli
John E. Bowers, Committee Chair
October 2005
iii
Optical Signal Processing Using Traveling-Wave Electroabsorption Modulators
Copyright © 2005
by
Hsu-Feng Chou
iv
This dissertation is dedicated to my family
for their love and support
v
Acknowledgements
Staying five years in Santa Barbara in the name of completing a Ph.D.
degree is a wonderful deal. It is even more enjoyable under the advisory of Prof.
John Bowers, who allowed me plenty of freedom to pursuit new frontiers in
fiber-optic communication while providing guidance and support. In addition,
his enthusiasm, energy, and entrepreneurship are inspirational to me in many
dimensions beyond academic research. The collaborations with Prof. Daniel
Blumenthal and his research group were instrumental to this dissertation. I really
appreciate the opportunity to carry out my dissertation work under the expertise
of both professors and within the research environment they have established,
which covers a full spectrum of fiber-optic communication. I would also like to
thank Prof. Kevin Almeroth and Prof. Nadir Dagli for serving in my committee
and providing valuable comments.
It is not possible to implement any experiment in this dissertation
without the wonderful devices fabricated by Yi-Jen Chiu, Matthew Sysak, and
Sheng Zhang. Their generous supply of device gave me the luxury to squeeze
out all possibilities of TW-EAM without worrying too much about potential
risks. This is what made all the innovations in this dissertation possible.
On the system side, I would like to thanks Zhaoyang Hu for close
collaborations over the years, especially on clock recovery related researches.
vi
The collaborations with Kohsuke Nishimura, Ryo Inohara, and Masashi Usami
in KDDI R&D Laboratories Inc. on the 160-Gb/s clock recovery experiment are
also highly appreciated.
There are many other researchers in UCSB to whom I would like to
express my sincere gratitude. Volkan Kaman and Lavanya Rau taught me how
to do system experiments during my early days in UCSB. The first 160-Gb/s
experiment cannot be completed without the help of Wei Wang. The
collaboration with Gustaf Smedman is a memorable experience, not to mention
his local guide in Stockholm. I would also like to thank Henrik Poulsen, Suresh
Rangarajan, and Roopesh Doshi for their help in the lab.
The support and friendship from the current and previous members of
the Ultrafast Optoelectronic Research Group are greatly appreciated. In
particular, I would like to thank Kian-Giap Gan, Gehong Zeng, Jonathan Geske,
and Satoshi Kodama for inspirational discussions and countless helps. The
friendship of Yu-Chia Chang, Dallas Hinds, Chiou-Fu Wang, Heng-Kuang Lin,
and Katharina Rauscher among many others made my days in UCSB more
colorful. Finally, I would like to thank my wife, Shu-Chuan, for her love and
companion over these years. It is UCSB that brought us together.
vii
Curriculum Vitæ Hsu-Feng Chou
Personal Born July 29, 1974 Taipei, Taiwan
Education 2000-2005 Ph.D. in Electrical and Computer Engineering University of California, Santa Barbara, U.S.A. 1996-1998 Master of Science in Electro-Optical Engineering National Taiwan University, Taipei, Taiwan 1992-1996 Bachelor of Science in Physics National Taiwan University, Taipei, Taiwan
Award
Bor-Uei Chen Memorial Scholarship, Year 2005 Awarded by The Photonics Society of Chinese-Americans
Publications Journal Papers
1. Zhaoyang Hu, Roopesh Doshi, Hsu-Feng Chou, Henrik N. Poulsen, David
Wolfson, John E. bowers, and Daniel J. Blumenthal, “Optical Label Swapping Using Payload Envelope Detection Circuits”, IEEE Photonics Technology Letters, vol. 17, no. 7, pp. 1537-1539, 2005
2. Zhaoyang Hu, Hsu-Feng Chou, Kohsuke Nishimura, Masashi Usami, John E. Bowers, and Daniel J. Blumenthal, “Optical Clock Recovery Circuits Using Traveling-Wave Electroabsorption Modulator-Based Ring Oscillators for 3R Regeneration”, IEEE Journal of Selected Topics in Quantum Electronics, vol. 11, no. 2, pp. 329-337, 2005
viii
3. Hsu-Feng Chou, and John E. Bowers, “Simplified optoelectronic 3R regenerator using nonlinear electro-optical transformation in an electroabsorption modulator”, Optics Express, vol. 13, no. 7, pp. 2742-2746, April 2005
4. Wei Wang, Henrik N. Poulsen, Lavanya Rau, Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal, “Raman-Enhanced Regenerative Ultrafast All-Optical Fiber XPM Wavelength Converter”, Journal of Lightwave Technology, vol. 23, no. 3, pp. 1105-1115, 2005
5. Hsu-Feng Chou, Zhaoyang Hu, John E. Bowers, and Daniel J. Blumenthal, “Compact Optical 3R Regeneration Using a Traveling-Wave Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 17, no. 2, pp. 486-488, 2005
6. Toshio Kimura, Staffan Bjorlin, Hsu-Feng Chou, Qi Chen, Shaomin Wu, and John E. Bowers, “Optically Preamplified Receiver at 10, 20, and 40 Gb/s Using a 1550-nm Vertical-Cavity SOA”, IEEE Photonics Technology Letters, vol. 17, no. 2, pp. 456-458, 2005
7. Zhaoyang Hu, Kohsuke Nishimura, Hsu-Feng Chou, Lavanya Rau, Masashi Usami, John E. Bowers, and Daniel J. Blumenthal, “40-Gb/s Optical Packet Clock Recovery Using a Traveling-Wave Electroabsorption Modulator-Based Ring Oscillator”, IEEE Photonics Technology Letters, vol. 16, no. 12, pp. 2640-2642, 2004
8. Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal, “Compact 160-Gb/s Add-Drop Multiplexer With a 40-Gb/s Base Rate Using Electroabsorption Modulators”, IEEE Photonics Technology Letters, vol. 16, no. 6, pp. 1564-1566, 2004
9. Zhaoyang Hu, Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal, “40-Gb/s Optical Clock Recovery Using a Traveling-Wave Electroabsorption Modulator-Based Ring Oscillator”, IEEE Photonics Technology Letters, vol. 16, no. 5, pp. 1376-1378, 2004
10. Bin Liu, Jongin Shin, Yi-Jen Chiu, Hsu-Feng Chou, Joachim Piprek, and John E. Bowers, “Slope Efficiency and Dynamic Range of Traveling-Wave Multiple-Quantum-Well Electroabsorption Modulators”, IEEE Photonics Technology Letters, vol. 16, no. 2, pp. 590-592, 2004
ix
11. Hsu-Feng Chou, Zhaoyang Hu, John E. Bowers, Daniel J. Blumenthal, Kohsuke Nishimura, Ryo Inohara, amd Masashi Usami, “Simultaneous 160-Gb/s Demultiplexing and Clock Recovery by Utilizing Microwave Harmonic Frequencies in a Traveling-Wave Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 16, no. 2, pp. 608-610, 2004
12. Hsu-Feng Chou, Yi-Jen Chiu, Adrian Keating, John E. Bowers, and Daniel J. Blumenthal, “Photocurrent-Assisted Wavelength (PAW) Conversion With Electrical Monitoring Capability Using a Traveling-Wave Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 16, no. 2, pp. 530-532, 2004
13. Lavanya Rau, Suresh Rangarajan, Wei Wang, Hsu-Feng Chou, Henrik N. Poulsen, John Bowers, and Daniel J. Blumenthal, “High-speed optical time-division-multiplexed/WDM networks and their network elements based on regenerative all-optical ultrafast wavelength converters”, OSA Journal of Optical Networking, vol. 3, no. 2, pp. 100-118, 2004
14. Zhaoyang Hu, Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal, “40-GHz Optical Pulse Generation Using Strong External Light Injection of a Gain-Switched High-Speed DBR Laser Diode”, IEEE Photonics Technology Letters, vol. 15, no. 12, pp. 1767-1769, 2003
15. Hsu-Feng Chou, Yi-Jen Chiu, Wei Wang, John E. Bowers, and Daniel J. Blumenthal, “Compact 160-Gb/s Demultiplexer Using a Single-Stage Electrically-Gated Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 15, no. 10, pp. 1458-1460, 2003
16. Wei Wang, Henrik N. Poulsen, Lavanya Rau, Hsu-Feng Chou, John E. Bowers, Daniel J. Blumenthal, and Lars Gruner-Nielsen, “Regenerative 80-Gb/s Fiber XPM Wavelength Converter Using a Hybrid Raman/EDFA Gain Enhanced Configuration”, IEEE Photonics Technology Letters, vol. 15, no. 10, pp. 1416-1418, 2003
17. Hsu-Feng Chou, Yi-Jen Chiu, and John E. Bowers, “Standing-Wave Enhanced Electroabsorption Modulator for 40-GHz Optical Pulse Generation”, IEEE Photonics Technology Letters, vol. 15, no. 2, pp. 215-217, 2003
x
18. Hsu-Feng Chou, Yi-Jen Chiu, Lavanya Rau, Wei Wang, Suresh Rangarajan, John E. Bowers, and Daniel J. Blumenthal, “Low Power Penalty 80 to 10 Gbit/s OTDM Demultiplexer Using Standing-Wave Enhanced Electroabsorption Modulator With Reduced Driving Voltage”, Electronics Letters, vol. 39, no. 1, pp. 94-95, 2003
19. Daniel J. Blumenthal, John E. Bowers, Lavanya Rau, Hsu-Feng Chou, Suresh Rangarajan, Wei Wang, and Henrik N. Poulsen, “Optical Signal Processing for Optical Packet Switching Networks”, IEEE Opitcal Communications, vol. 1, no. 1, pp. S23-S29, 2003
20. Lavanya Rau, Wei Wang, Bengt-Erik Olsson, Yi-Jen Chiu, Hsu-Feng Chou, Daniel J. Blumenthal, and John E. Bowers, “Simultaneous All-Optical Demultiplexing of a 40-Gb/s Signal to 4 x 10 Gb/s WDM Channels Using an Ultrafast Fiber Wavelength Converter”, IEEE Photonics Technology Letters, vol. 14, no.12, pp. 1725-1727, 2002
21. Yi-Jen Chiu, Hsu-Feng Chou, Volkan Kaman, Patrick Abraham, John E. Bowers, “High Extinction Ratio and Saturation Power Traveling-Wave Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 14, no. 6, pp. 792-794, 2002
22. Hsu-Feng Chou, Yi-Jen Chiu, and John E. Bowers, “40GHz Optical Pulse Generation Using Sinusoidally-Driven Traveling-Wave Electroabsorption Modulator”, Electronics Letters, vol. 38, no. 8, pp. 379-380, 2002
23. Shing Mou, Ching-Fuh Lin, and Hsu-Feng Chou, “Quasi-3-D Beam-Propagation Method for Modeling Nonlinear Wavelength Conversion”, Journal of Lightwave Technology, vol. 19, no. 5, pp. 772-779, 2001
24. Hsu-Feng Chou,, Ching-Fuh Lin, and Shing Mou, “Comparisons of Finite Difference Beam Propagation Methods for Modeling Second-Order Nonlinear Effects”, Journal of Lightwave Technology, vol. 17, no. 8, pp. 1481-1486, 1999
25. Hsu-Feng Chou, Ching-Fuh Lin, and Gin-Chung Wang, “An Iterative Finite Difference Bean Propagation Method for Modeling Second-Order Nonlinear Effects in Optical Wavelguides”, Journal of Lightwave Technology, vol. 16, no. 9, pp. 1686-1693, 1998
xi
Conference Presentations 1. Zhaoyang Hu, Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal,
“40-Gb/s Optical 3R Regeneration Using a Traveling-wave Electroabsorption Modulator-Based Optical Clock Recovery”, Optical Fiber Communication Conference (OFC´05), paper OTuO5, Mar. 2005, Anaheim, CA
2. John E. Bowers, and Hsu-Feng Chou, “Applications of Traveling-Wave
Electroabsorption Modualtors in 160-Gbit/s Systems”, European Conference on Optical Communication (ECOC´04), Symposium Tu4.1.1, Sept. 2004, Stockholm, Sweden (invited talk)
3. Toshio Kimura, Staffan Bjorlin, Garrett Cole, Hsu-Feng Chou, and John E.
Bowers, “1550-nm Vertical Cavity SOAs for Optically Preamplified High Bit Rate Receivers”, European Conference on Optical Communication (ECOC´04), paper We4.P.070, Sept. 2004, Stockholm, Sweden
4. Zhaoyang Hu, Kosuke Nishimura, Hsu-Feng Chou, Lavanya Rau, Masashi
Usami, John E. Bowers, and Daniel J. Blumenthal, “40-Gb/s Optical Packet Clock Recovery Using a Traveling-wave Electroabsorption Modulator-Based Ring Oscillator”, European Conference on Optical Communication (ECOC´04), paper We3.5.4, Sept. 2004, Stockholm, Sweden
5. Y. L. Okuno, K.-G. Gan, H.-F. Chou, Y.-J. Chiu, C. Wang, S. Wu, J. Geske,
E. S. Bjorlin, J. E. Bowers, “Stable polarization operation of 1.3-µm wavelength vertical cavity surface emitting laser (VCSEL) fabricated by orientation-mismatched wafer bonding”, IEEE 19th International Semiconductor Laser Conference, Conference Digest., pp. 117-118, Sept. 2004
6. Zhaoyang Hu, Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal,
“A Compact All-Optical 40-Gb/s Clock Recovery Using a Traveling-wave Electroabsorption Modulator-Based Ring Oscillator With a Chip Coplanar Q-Filter”, Conference on Laser and Electro-Optics (CLEO´04), paper CTuW5, May 2004, San Francisco, CA
7. Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal ,"Compact 160-
Gb/s Add-Drop Multiplexing With a 40-Gb/s Base Rate", Optical Fiber Communication Conference (OFC´04), postdeadline paper PDP28, Feb. 2004, Los Angeles, CA
xii
8. Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal ,"Novel Photocurrent-assisted wavelength (PAW) converter using a traveling-wave electroabsorption modulator with signal monitoring and regeneration capabilities", Optical Fiber Communication Conference (OFC´04), paper FD4, Feb. 2004, Los Angeles, CA
9. Hsu-Feng Chou, Zhaoyang Hu, John E. Bowers, Daniel J.Blumenthal,
Kohsuke Nishimura, Ryo Inohara, and Masashi Usami, “Simultaneous Demultiplexing, Electrical Clock Recovery, and Optical Clock Generation Using Photocurrent and Subharmonic Frequency in a Traveling-Wave Electroabsorption Modulator”, European Conference on Optical Communication (ECOC´03), Sept. 2003, Rimini, Italy
10. Bin Liu, Jongin Shin, Yi-Jen Chiu, Hsu-Feng Chou, Kian-Giap Gan,
Joachim Piprek, and John E. Bowers, “Dynamic Range of Traveling-Wave Multiple-Quantum-Well Electroabsorption Modulators”, Conference on Lasers and Electro-Optics (CLEO ’03), paper CTuJ3, June 2003
11. Hsu-Feng Chou, Yi-Jen Chiu, John E. Bowers, Wei Wang, and Daniel J.
Blumenthal ,"160Gb/s to 10Gb/s OTDM Demultiplexing Using a Traveling-Wave Electroabsorption Modulator", Optical Fiber Communication Conference (OFC´03), vol. 2, paper ThX2, pp. 583-584, Mar. 2003, Atlanta, GA
12. Wei Wang, Henrik N. Poulsen, Lavanya Rau, Daniel J. Blumenthal, Hsu-
Feng Chou, John E. Bowers, L Grüner-Nielsen, “80-Gb/s Regenerative Wavelength Conversion Using a Hybrid Raman/EDFA Gain-Enhanced XPM Converter with Highly-Nonlinear-Fiber”, Optical Fiber Communication Conference (OFC´03), vol. 1, paper TuH2, pp. 193-194, Mar. 2003, Atlanta, GA
13. Hsu-Feng Chou, Yi-Jen Chiu, John E. Bowers, Lavanya Rau, Suresh
Rangarajan, and Daniel J. Blumenthal, “Standing-Wave Enhanced Electroabsorption Modulator for 80 to 10 Gb/s OTDM Demultiplexing”, European Conference on Optical Communication (ECOC´02), Sept. 2002, Copenhagen, Denmark
14. Daniel J. Blumenthal, John E. Bowers, Yi-Jen Chiu, Hsu-Feng Chou, Bengt-
Erick Olsson, Suresh Rangarajan, Lavanya Rau, Wei Wang, “Fast Optical Signal Processing in Optical Transmission”, IEEE/LEOS Summer Topical Meeting, Jul. 2002, Mont-Tremblant, Canada (invited)
xiii
15. Daniel J. Blumenthal, John E. Bowers, Yi.-Jen Chiu, Hsu-Feng Chou, Bengt-Erik Olsson, Suresh Rangarajan, Lavanya Rau, and Wei Wang, “Optical Packet Switching and Associated Optical Signal Processing”, IEEE/LEOS Summer Topical Meeting, Jul. 2002, Mont-Tremblant, Canada (invited talk)
16. Hsu-Feng Chou, Yi-Jen Chiu John E. Bowers, “Using Standing-wave
Electroabsorption Modulators to Generate 40GHz Optical Pulses”, Conference on Laser and Electro-Optics (CLEO´02), paper CMI1, pp. 41-42, Jun. 2002, Long Beach, CA
17. Hsu.-Feng Chou, Yi-Jen Chiu, and John E. Bowers, “40GHz Optical Pulse
Generation Using Traveling-Wave Electroabsorption Modulator”, Optical Fiber Communication Conference (OFC´02), paper WV2, pp. 338-339, Mar. 2002, Anaheim, CA
18. Lavanya. Rau, Suresh Rangarajan, Daniel J. Blumenthal, Hsu-Feng Chou,
Yi-Jen Chiu, John E. Bowers, “Two-Hop All-Optical Label Swapping with Variable Length 80Gb/s Packets and 10Gb/s Label using Nonlinear Fiber Wavelength Converters, Unicast/Multicast Output and a single EAM for 80- to 10Gb/s Packet Demultiplexing”, Optical Fiber Communication Conference (OFC´02), postdeadline paper, FD2, Mar. 2002, Anaheim, CA
19. Ching-Fuh Lin, Shin Mou, Hsu-Feng Chou, “Quasi-3D beam-propagation
method for simulating quasi-phase-matched second-order nonlinear interaction”, OSA Topical Meeting on Nonlinear Optics: Materials, Fundamentals, and Applications, paper TuB25, 2000, Kaua'i-Lihue, HI
20. Hsu-Feng Chou, Gin-Chung Wang, and Chin-Fuh Lin, “A Study of Quasi-
Phase-Matched Second-Order Nonlinear Effects in Optical Waveguides Using an Iterative Finite-Difference Beam Propagation Method”, European Conference on Laser and Electro-Optics (CLEO/Europe´98), paper CThH39, 1998, Glasgow, UK
21. Hsu-Feng Chou, Ye-Wen Hong, Ching-Fuh Lin, “The Influence of Non-
ideal Imple-mentation and Noise on Quasi-Phase-Matched Parametric Interactions in Optical Waveguides”, 1998 International Photonics Conference (IPC’98), December 1998, Taipei, Taiwan
xiv
22. Hsu-Feng Chou, Shing Mou, Ching-Fuh Lin, “Comparisons of Various Finite Difference Beam Propagation Methods for the Modeling of Second-Order Nonlinearities”, 1998 International Photonics Conference (IPC’98), December 1998, Taipei, Taiwan
23. Hsu-Feng Chou, Ching-Fuh Lin, “An Iterative Finite Difference Beam
Propagation Method to Model Quasi-Phase-Matched Second-Order Nonlinear Effects in Optical Waveguides”, Optics and Photonics / Taiwan’97, December 1997, Hsinchu, Taiwan
24. Ching-Fuh Lin, Hsu-Feng Chou, Fuh-Hsiang Yang, “A Study of Bending
Waveguide for Semiconductor Photonics”, Progress in Electromagnetics Research Symposium (PIERS 1997), January 1997, Hong Kong
Book Chapter 1. John E. Bowers, and Hsu-Feng Chou, “Wavelength Division Multiplexing”,
Encyclopedia of Modern Optics, Elsevier, December 2004
xv
Abstract
Optical Signal Processing Using Traveling-Wave Electroabsorption Modulators
by
Hsu-Feng Chou
Internet traffic is expected to increase by a thousand times in the next ten
years, which poses a severe challenge to the present optical transport network.
To scale for future traffic, the optical network needs to handle higher data-rates
while offer more flexibility. The most promising solution is a network based on
the mixing of optical time-division multiplexing (OTDM), wavelength-division
multiplexing (WDM), and photonic cross-connect (PXC) technologies that
involve optical signal processing in the time, wavelength and space domains. At
the same time, these technologies are required to achieve higher performance
with lower cost and smaller size.
The goal of this dissertation is to meet the demands of future optical
networks by developing compact and high-performance optoelectronic sub-
systems using traveling-wave electroabsorption modulators (TW-EAMs). The
optical signal processing capabilities studied in this dissertation include
demultiplexing, add-drop multiplexing, and clock recovery for OTDM and
wavelength conversion and 3-R regeneration for WDM. To ensure that the
proposed approaches meet the demands in speed and functionality with the least
xvi
amount of complication, the unique properties of the TW-EAM are explored and
utilized in full to provide novel solutions. These properties include (1) the
distributed effect due to the traveling-wave electrodes; (2) the photocurrent
signal generated inside the device; (3) the nonlinear electro-optical transfer
function of the TW-EAM; and (4) the integrability with other optoelectronic
devices. Many resulting sub-systems have reached record operation speed and
compactness. For example, the first 160-Gb/s OTDM add-drop multiplexing
with a 40-Gb/s base-rate was demonstrated with TW-EAMs operated in the
standing-wave enhanced mode, which is one of the new inventions proposed in
this dissertation. Less than 1-dB of power penalty was achieved for all
operations. On the other hand, a very compact optical 3-R regenerator was
demonstrated by incorporating three essential functions (clock recovery, pulse
generation, and nonlinear gating) into a single TW-EAM through the utilization
of unique properties of the device. In addition, extensive numerical simulations
are presented throughout the dissertation to model and study the proposed
concepts.
xvii
Contents 1 Introduction
1.1 Evolution of fiber-optic communications……………...................... 1.2 Traveling-wave electrabsorption modulator……………...……...…
1.2.1 Electroabsorption modulator………………..……...………. 1.2.2 Traveling-wave electrode…………………………………... 1.2.3 Characteristics of the implemented TW-EAM……………..
1.3 This Dissertation…………………………………………………… 1.3.1 Approach and philosophy………………………………….. 1.3.2 Overview of chapters……………………………………….
References……………………………………………………………….
1 8 9 10 12 15 15 21 25
2 Optical Pulse Generation
2.1 Introduction………………………………………………………... 29 2.2 Traveling-wave phenomena in optical pulse generation…………... 36
2.2.1 Experimental observations…………………………………. 36 2.2.2 Modeling of traveling-wave phenomena………………….. 40
2.3 Standing-wave enhanced mode of TW-EAM……………………… 44 2.3.1 Concept…………………………………………………….. 45 2.3.2 Experimental and theoretical verifications………………… 2.3.3 Generating shorter pulses…………………………………...
47 52
2.4 Summary…………………………………………………………… References……………………………………………………………….
55 56
3 OTDM Gating Operations 3.1 Introduction………………………………………………………… 3.2 Optical demultiplexing……………………………………………...
3.2.1 Demultiplexing using the standing-wave enhanced mode….3.2.2 Demultiplexing using a shaped electrical driving signal…...
59 69 70 74
xviii
3.3 Time-domain add-drop multiplexing………………………………. 3.4 Summary…………………………………………………………… References……………………………………………………………….
82 89 90
4 Clock Recovery Techniques 4.1 Introduction………………………………………………………… 4.2 Scaled OTDM clock recovery with simultaneous demultiplexing…
4.2.1 Concept…………………………………………………….. 4.2.2 40-Gb/s operation………………………………….………..4.2.3 160-Gb/s operation………………………………………….
4.3 Modeling of scaled clock recovery………………………………… 4.3.1 Model………………………………………………………. 4.3.2 Lock-in range………………………………………………. 4.3.3 Shift of free-running frequency……………………………. 4.3.4 Phase locking transient…………………………………...... 4.3.5 Small-signal analysis……………………………………….
4.4 Clock recovery by injection locking a ring oscillator……………… 4.4.1 Concept…………………………………………………….. 4.4.2 Characteristics measured at 10-Gb/s………………………..
4.5 Summary………………………………………………………….... References………………………………………………………………..
93 101102103108112112117120122129132133136146147
5 Wavelength Conversion 5.1 Introduction………………………………………………………… 5.2 Photocurrent-Assisted Wavelength Conversion
(PAW-Conversion)………………………………………………… 5.2.1 Concept…………………………………………………….. 5.2.2 Proof of concept (2.5 Gb/s NRZ operation)………………...5.2.3 Speed limitations…………………………………………… 5.2.4 10-Gb/s NRZ operation using shallow QWs……………….
5.3 Modeling of PAW-Conversion…………………………………….. 5.3.1 Model………………………………………………………. 5.3.2 Simulation of cascaded conversions………………………..
5.4 RF-Driven PAW-Conversion (PAW-Regeneration)………………. 5.4.1 Concept…………………………………………………….. 5.4.2 10-Gb/s RZ operation……………………………………… 5.4.3 Regenerative properties…………………………………….
5.5 Summary…………………………………………………………… References………………………………………………………………..
151 162162167174176190191196201202204208210212
xix
6 3-R Regeneration 6.1 Introduction………………………………………………………… 6.2 Compact 3-R PAW-Regeneration………………………………….
6.2.1 Concept…………………………………………………….. 6.2.2 10-Gb/s RZ signal regeneration……………………………. 6.2.3 Performance limitations…………………………………….
6.3 Simplified optoelectronic 3-R regeneration………….…………….. 6.3.1 Concept………………………………………..…………… 6.3.2 10-Gb/s RZ signal regeneration……………………………. 6.3.3 Modeling…………………………………………………… 6.3.4 Simulation of cascaded performance………………………. 6.3.5 Simulation of regeneration capabilities…………………….
6.4 Integrated optoelectronic 3-R regeneration………………………... 6.4.1 Experimental setup………………………………………….6.4.2 10-Gb/s RZ signal regeneration…………………………….
6.5 Summary…………………………………………………………… References………………………………………………………………..
215226227231239257259262273277281288289292300302
7 Summary and Future Work 7.1 Summary…………………………………………………………… 7.2 Suggestions for future research……………………………………..
7.2.1 Techniques for higher OTDM line-rate speeds……….…… 7.2.2 Improved QW design for better performances…………….. 7.2.3 Monolithic and hybrid integrations for higher speed……….
References………………………………………………………………..
305312312314314315
1
Chapter 1 Introduction 1.1 Evolution of fiber-optic communications
Over the past few decades, fiber-optic communications have made
significant progress in providing higher transmission capacity with lower cost.
The success of the Internet, which already changed human society globally,
benefited a lot from the developments in fiber-optics technology. Given the
115% per year growth observed in the past few years [1], the Internet traffic is
expected to increase by a thousand times in the next ten years, which poses a
severe challenge to the present optical network. To scale for future traffic, the
optical network needs to handle higher data-rates while offer more flexibility.
Up to date, the most promising solution is an optical network that utilizes optical
signal processing in the time, wavelength and space domains.
CHAPTER 1: Introduction
2
There are three general ways to increase the transmission capacity in
fiber-optic communications. The first is space division multiplexing (SDM),
which simply means increasing the number of fibers in the deployed cable or
increasing the number of cables. In a typical installation, there may be between
8 and 1000 fibers in parallel. This approach, while straight forward, may not be
very cost-effective when the demand of bandwidth grows rapidly. This is
particularly true once the cable has been laid. On the other hand, the other two
approaches, wavelength-division multiplexing (WDM) and time-division
multiplexing (TDM), work on increasing the capacity of each fiber.
Figure 1.1 Concept of wavelength-division multiplexing (WDM)
In WDM transmission, as shown in Figure 1.1, signals are sent with
different wavelengths and the aggregate capacity is increased accordingly. The
low loss transmission band of fiber extends from 1200 nm to 1600 nm, which
corresponds to over 40 THz of available bandwidth. Multiplexing (MUX) and
demultiplexing (DMUX) components based on technologies such as arrayed-
CHAPTER 1: Introduction
3
waveguide grating (AWG) are essential to enable WDM systems. The erbium-
doped fiber amplifier (EDFA) can amplify a large number of channels
simultaneously, which is a critical technology to reduce the cost of regeneration
for long-haul transmission and was the key driver for cost effective WDM
transmission.
WDM has been quite successful in the past decade and is widely
deployed in optical networks worldwide. With the advancement in related
technologies, particularly new optical amplifiers to cover wider bandwidths, a
record 10.92 Tb/s (273×40Gb/s) transmission through 117 km of fiber was
demonstrated in 2001 [2].
The second approach, which can be used in conjunction with WDM, is
time division multiplexing (TDM), consisting of interleaving multiple data
streams in time into one higher speed data stream. For example, 672 voice
channels (DS0) at 64 Kb/s are multiplexed into one T1 data stream at 1.544
Mb/s. 84 of these T1 channels are then multiplexed into a SONET (Synchronous
Optical NETwork) OC-3 signal at 155 Mb/s. With the advances in electronics,
the data-rate of electrical TDM (ETDM) has been increased up to 40 Gb/s and
commercial deployment is technologically feasible. Many WDM systems have
adopted such a high ETDM base-rate in order to reduce the number of optical
channels to be managed.
CHAPTER 1: Introduction
4
Figure 1.2 Concept of optical time-division multiplexing (OTDM)
To increase the single wavelength data-rate beyond the speed of
electronics, optical time-division multiplexing (OTDM) can be utilized, as
shown in Figure 1.2. The prerequisite is that return-to-zero (RZ) data format
must be adopted and the pulsewidth of each base-rate channel must be short
enough so that they can be interleaved in time and combined together without
interference by an OTDM multiplexer. On the receiver side, a demultiplexer
composed of synchronized optical gates is needed to bring the high-speed line-
rate signal down to base-rate tributaries. The critical components for OTDM are
the pulse source at the transmitter and the optical gate at the receiver. They
determine how many channels can be multiplexed in time. Most demonstrations
in recent years are based on 10-Gb/s or 40-Gb/s ETDM technologies and OTDM
is employed to increase the single wavelength data-rate. As in the case of WDM,
adopting a higher ETDM rate can reduce the number of optical channels, which
means a reduced complexity. Compared to WDM, OTDM is more demanding
on dispersion management since short optical pulses are used, which have a
CHAPTER 1: Introduction
5
wide spectrum. Nevertheless, with the same aggregate bit-rate, OTDM may
have a better spectral efficiency since wavelength channels are separated by
empty guard bands in order to reduce crosstalk in WDM. The highest OTDM
line-rate demonstrated so far is 1.28 Tb/s with the aid of polarization
multiplexing [3].
Figure 1.3 Optical add-drop multiplexing (OADM) in (a) WDM network; (b) OTDM network
Point-to-point optical transmission is widely deployed in opaque
networks with electronic switching and regeneration at each node. Enabled by
advances in fiber-optic technology, more and more functionalities can be
implemented in the optical domain without converting the signal back and forth
between the optical and the electrical domain at each node. This promises a
more transparent optical network in terms of bit-rate, modulation format, and
protocol. A great reduction in cost has also occurred and additional cost
CHAPTER 1: Introduction
6
reduction is expected. Therefore, the next stage in optical networking is the
deployment of transparent ring and mesh networks.
Figure 1.3(a) shows a simple ring network where most of the optical
signals pass through the nodes transparently. The optical add-drop multiplexing
(OADM) elements drop some wavelengths at each node and add new channels.
When more and more wavelength channels are deployed, it may become
necessary to dynamically reallocate wavelengths. Therefore, reconfigurable
optical add-drop multiplexers (ROADMs) are needed to enable a more dynamic
network, where the configuration of the wavelengths to be processed can be
changed by the service provider. Tunable lasers and transmitters are key
components to enable this technology. A similar OADM concept can be
implemented in the OTDM network, where the channels are manipulated in the
time domain, as indicated in Figure 1.3(b).
OTDM and WDM can be employed at the same time to make full usage
of the available fiber bandwidth. As the data traffic continues to increase, a
tremendous number of channels will have to be handled at each network node
due to the use of all three multiplexing schemes (OTDM, WDM and SDM). The
total capacity at each node can be as high as hundreds of Tb/s and is expected to
reach Pb/s or even Eb/s (1 Ebits =1 million Tbits) [1]. To deal with such huge
amount of traffic, a network node based on a wavelength-selective large-scale
photonic cross-connects (WSPXC) can be used, as shown in Figure 1.4. The
CHAPTER 1: Introduction
7
PXC is realized with the three-dimensional microelectromechanical system (3-D
MEMS) and is capable of non-blocking switching with low-loss [4]. In this
optically transparent network node, all the OTDM and WDM channels can be
switched in space, time, and wavelength domains, providing full management of
the data traffic without converting each channel into the electrical domain.
Additional optical signal processing capabilities such as regeneration and
wavelength conversion can also be implemented at either the OTDM line-rate or
the base-rate, depending on the requirement of the network. These resources can
be shared by all the wavelength channels through the dynamic switching of
WSPXC.
Figure 1.4 An optical cross-connect node in transparent OTDM-WDM network
CHAPTER 1: Introduction
8
The realization and deployment of such high-speed and high-capacity
transparent networks still require many research efforts to make the associated
technologies more mature and cost-effective. For OTDM, even though many
high-speed systems have been demonstrated, the associated complexity and
high-cost have made practical deployment difficult. Therefore, one mission of
this dissertation is to come up with compact and efficient sub-systems to enable
high-speed OTDM with reduced complexity and cost. The covered sub-systems
include optical pulse generator, time-domain add-drop multiplexer, optical
demultiplexer and scaled clock recovery for OTDM systems up to 160 Gb/s. On
the other hand, low-cost and high-performance wavelength converters and 3-R
regenerators at the ETDM rate will also be proposed, demonstrated and studied
for WDM systems. All of these compact sub-systems are based on the traveling-
wave electroabsorption modulators designed and fabricated at the University of
California, Santa Barbara.
1.2 Traveling-wave electroabsorption modulator
The development of the central element of this dissertation, the
traveling-wave electro-absorption modulator (TW-EAM), is briefly reviewed in
this section. The general characteristics of the implemented device utilized in
this dissertation are also presented.
CHAPTER 1: Introduction
9
1.2.1 Electroabsorption modulator
The electroabsorption modulator (EAM) is a compact, efficient, and
integrable component for fiber-optic communications. It utilizes either the
Franz-Keldysh effect in bulk material or the quantum confined Stark effect in a
quantum-well (QW) structure to change the absorption of light according to the
applied electric field [5]. QW structures are widely used for their higher
modulation efficiency but the wavelength dependence is generally larger than
the bulk material. Just like other external optical intensity modulators, the chirp
induced by EAM is much lower than that of directly modulated lasers. The
advantages of EAMs over other types of external modulators include: (1) a
smaller device dimension, usually on the order of a few hundred micrometers;
(2) a lower driving voltage because of its highly efficient and nonlinear electro-
optical (E-O) transfer function; and (3) the possibility of monolithic integration
with other semiconductor components. EAMs with lumped electrodes have been
demonstrated to exhibit a small-signal 3-dB bandwidth over 40 GHz [6].
A critical design issue of the EAM is the trade-off between the
modulation bandwidth and the extinction ratio. The bandwidth of the
conventional lumped-electrode EAM is determined by the RC-time constant. To
enable high-speed broadband modulation, the active length of the EAM must be
short in order to keep the capacitance low. For example, in Ref. [6], a 40-Gb/s
device is only 63-µm long and extra passive optical waveguides are necessary to
CHAPTER 1: Introduction
10
extend the device size for easier handling. The main concern of a short active
length is the relatively low modulation efficiency (expressed in dB/V), which
can result in a higher driving voltage. Also, the total extinction ratio is also
limited by a short device length. For data modulation, around 10 dB of dynamic
extinction ratio may be adequate, but for applications like optical gating, higher
extinction ratios are generally required. Another issue for a short EAM is the
reduced optical power handling capability.
1.2.2 Traveling-wave electrode
To overcome the trade-off between bandwidth and device length, EAMs
with traveling-wave (TW) electrodes were developed [7]-[13], where the
bandwidth is not RC limited and the device length can be longer without
sacrificing the bandwidth. In the TW design, the microwave driving signal
propagates through the active waveguide, which is part of the transmission line.
The capacitance of the EAM is no longer a lumped load but becomes
distributed. With a longer device, the modulation efficiency can be higher and
the driving voltage is reduced. Ideally, the bandwidth and the device length are
only limited by the microwave loss and the velocity mismatch between the
lightwave and the microwave. The later determines how long the lightwave can
be effectively modulated. Another limiting factor is the increased optical loss
CHAPTER 1: Introduction
11
with a longer device, which is related to the optical signal-to-noise ratio (OSNR)
of the modulated signal.
A design challenge for the TW-EAM is a low microwave impedance of
25 Ohm or below in the active waveguide, which causes reflections when driven
by a 50 Ohm source and limits the modulation bandwidth. The low impedance
of the active waveguide is caused by the need to maintain adequate optical
confinement, which is critical to the modulation efficiency. Therefore, low
impedance terminations in the range of 12 to 35 Ohm are required to obtain the
maximum bandwidth. For example, a bandwidth of 43 GHz was measured for a
450-µm device with a 13 Ohm termination [12]. One way to overcome the
impedance mismatch is to use a segmented transmission line design [14], where
alternating high impedance passive segments and low impedance active
segments are fabricated to obtain a composite impedance of 50 Ohm. Flat
response up to 50 GHz was measured and extrapolated to 90 GHz by a
theoretical model. In practice, a low impedance driver can be used to drive the
TW-EAM.
Even though the traveling-wave design has been quite successful in
extending the modulation bandwidth, its benefits in applications other than
broadband modulation are rarely explored, which is one of the central missions
of this dissertation.
CHAPTER 1: Introduction
12
1.2.3 Characteristics of the implemented TW-EAM
The general characteristics of the TW-EAM utilized in this dissertation
are summarized in this section. More details can be found in Ref. [5] and [11].
The epi-layer used for the TWEAM is an InGaAsP-based material grown by
MOCVD on a semi-insulated InP substrate. To achieve high modulation
efficiency and saturation power, wider wells and lower barriers were designed
and grown to improve the quantum-confined Stark effect and to reduce the
bandgap offset. The active region consists of 10 tensile-strain wells and 11
compressive-strain barriers, where the thickness of each period of well and
barrier is 19 nm. Cladding layers of p-InP (top) and n-InP (bottom) are grown
and sandwich the active region. As shown in Figure 1.5(a), a 300-µm long and
2-µm wide optical ridge-waveguide is fabricated, where the gases of CH4/H2/Ar
was used for reactive-ion-etching. The deposited metals on the N and the P
contacts form a coplanar waveguide transmission line (CPW line) for guiding
the microwave signal. Therefore, the lightwave and the microwave can interact
with each other along the waveguide through the electroabsorption effect of the
active material. As a result, this waveguide is called an active waveguide. PMGI
is utilized for passivating the etched surface and for bridging the ground metal
on both sides of the active waveguide. Two sections of passive CPW lines are
used to connect both ends of the active waveguide to outside circuits.
CHAPTER 1: Introduction
13
Figure 1.5 (a) SEM picture of the fabricated TW-EAM; (b) schematic drawing of the TW-EAM
that is used for illustration throughout the dissertation. G:ground, S:signal.
Figure 1.6 Typical characteristics of the TW-EAM (a) Optical transmission at 1555 nm; (b) Microwave S-parameters; (c) Measured and modeled small-signal E-O response with a 35-Ohm
termination; (d) measured small-signal O-E response with a 50-Ohm termination
CHAPTER 1: Introduction
14
Typical static optical transmission curves for the TE and the TM
polarizations are shown in Figure 1.6(a) (TW-EAMs with different compositions
of QWs may be used in the dissertation. In that case, the transmission curves
will be presented). The peak modulation efficiency can be as high as 30 dB/V
for the TM polarization. The polarization dependence can be greatly reduced by
properly compensating the strains in the QWs [5]. It was demonstrated that the
implemented TW-EAM is capable of 10 Gb/s operation with only 1 Vpp of
driving voltage, and over 14 dBm of optical saturation power [11].
Figure 1.6(b) shows the measured microwave S-parameters of the TW-
EAM up to 40 GHz. Note that the impedance of the active waveguide is 25 Ohm
or less, while that of the passive CPW lines is about 50 Ohm. Reflections occur
at the joints and become a limiting factor at elevated frequencies. The increase
of S11 is mainly caused by these reflections. On the other hand, the decrease of
S21 with frequency is mainly caused by microwave loss and in a smaller part by
the reflection.
The small-signal electrical to optical (E-O) response is plotted in Figure
1.6(c) with a 35-Ohm termination, which gives the highest bandwidth over 20
GHz [11]. Depending on the termination of the TW-EAM, the 3-dB bandwidth
of the E-O response can vary dramatically.
The TW-EAM can also be used as a photodetector to generate
photocurrent. The detected photocurrent signal in the TW-EAM is intensively
CHAPTER 1: Introduction
15
utilized in this dissertation for many novel applications. Figure 1.6(d) shows that
with a 50-Ohm termination, the 3-dB small signal optical to electrical (O-E)
response is about 10 GHz. This is smaller than the E-O response with the same
termination, which should be caused by the relatively long carrier sweep-out
time of the active material. This issue will be studied in more detail in Chapter
5.
1.3 This Dissertation
This dissertation is dedicated to meet the system-level demands of next
generation optical networks by exploring the unique device-level properties of
the TW-EAM. Through the study and utilization of these unique properties, the
proposed sub-systems based on TW-EAMs can meet system requirements on
speed and functionality in more compact and efficient configurations. In Section
1.3.1, the unique properties of the TW-EAM utilized in this dissertation are
briefly reviewed, together with the philosophies that guide the direction of
research. The organization of the chapters is given in Section 1.3.2.
1.3.1 Approach and philosophy
Several unique properties of the TW-EAM are studied and utilized in
this dissertation in order to build novel sub-systems for OTDM and WDM
CHAPTER 1: Introduction
16
systems. These properties include (1) the distributed effect in the traveling-wave
device; (2) the photocurrent signal generated by the TW-EAM; (3) the nonlinear
E-O transformation of the TW-EAM; (4) the integrability of TW-EAM with
other optoelectronic devices.
Distributed effect — The traveling-wave design overcomes the RC-
limited bandwidth of a lumped device by making the TW-EAM as part of the
microwave transmission line. In the TW-EAM, the microwave “propagates”
through the device instead of being terminated. The immediate benefit is that a
longer device is possible for high-speed operation with lower driving voltage
and higher extinction ratio. Another less explored consequence of a longer
device is that the device can no longer be modeled as a lumped element with no
physical length.
Figure 1.7 Distributed effect of the TW-EAM. The two configurations may have different results in high-frequency operation
The length of the TW-EAM makes particular importance when the
microwave wavelength is comparable to the device dimension. In this case, the
CHAPTER 1: Introduction
17
microwave distribution inside the device is not even and distributed effects may
occur. Figure 1.7 shows an example that is studied in Chapter 2, where the
results of the two configurations may be different, depending on if the optical
wave propagates in the same or in the opposite direction with respect to the
electrical signal. This is not expected from a lumped device. In addition, even if
the microwave wavelength is much longer than the device dimension, a long
traveling-wave device still makes a difference when combined with the next
property.
Photocurrent signal — A reverse-biased EAM is in fact a
photodetector. The only difference is that the absorption of EAM is usually
actively modulated and more attention is paid on the transmitted optical signal
instead of the detected electrical signal. In this dissertation, several novel
simultaneous operations enabled by the photocurrent signal are proposed and
demonstrated in order to incorporate more functions into a single device.
Figure 1.8 Generation of photocurrent signal in the TW-EAM
CHAPTER 1: Introduction
18
In the TW-EAM, the photocurrent signal can lead to interesting effects.
As shown in Figure 1.8, the generated photocurrent signal propagates in both
directions along the traveling-wave electrodes and eventually couples to outside
electronics. However, because of the long active device length made possible by
the traveling-wave design, the photocurrent signal may go through a non-
negligible portion of active waveguide before coupling into passive CPW lines.
During the propagation through the active waveguide, the photocurrent signal
can modulate the local voltage of the active waveguide. As a result, the
absorption of the TW-EAM can be changed if the photocurrent signal is strong
enough. This mechanism is not expected in a lumped EAM with no traveling-
wave electrodes.
Nonlinear E-O transformation — Generally speaking, the electrical to
optical (E-O) transformation of the EAM is nonlinear. A typical curve in linear
scale is shown in Figure 1.9. It can be pretty linear in some voltage ranges but
overall it is nonlinear. By utilizing only the linear region, fairly linear analog RF
transmission has been achieved [15]. However, for digital transmission, the
nonlinear property is particularly useful.
The high efficiency region (high dB/V) is useful for generating short
optical pulses and gating windows. This is due to the fact that an EAM with a
higher dB/V value takes less voltage to achieve a given modulation depth. For
example, if the driving signal is a sinusoidal wave, a more efficient EAM
CHAPTER 1: Introduction
19
requires less duration in time to change the output power by 50% and generates
shorter pulses.
Figure 1.9 Nonlinear E-O transfer function of the EAM
The step-like transfer function is extremely useful for improving digital
signals because it can be utilized to remove the amplitude fluctuations on the
“1” and the “0” levels of the electrical signal and generate an optical signal with
better quality. Nevertheless, to obtain this benefit it is required that the
amplitude of the electrical driving signal be large enough to sweep both of the
flat regions. It goes without saying that the EAM must have a step-like E-O
transfer function in the first place.
Integrability with other optoelectronic devices — One of the most
attractive features of EAM is the potential to be integrated with other
optoelectronic devices in order to shrink a complicated sub-system on a single
chip. Reduction in size and cost with integration technology is highly desired in
CHAPTER 1: Introduction
20
future fiber-optic communications. For example, Figure 1.10 shows a possible
configuration for an integrated tunable transmitter for return-to-zero data format
[16]. Great savings in coupling loss, power consumption, and size can be
achieved. Large-scale optoelectronic integration may dramatically improve the
economics of many conventional O-E-O-style sub-systems and make them a
competitive solution. For example, the regeneration of signals in a WDM system
with high port counts [17] may be economically viable.
Figure 1.10 Integration of TW-EAM with other optoelectronic devices
There are two general philosophies that guide the selection of technology
and the direction of research throughout this dissertation.
(1) Higher performance and more function with less complexity — The
speed and functionality of the proposed approaches should be improved with
minimal complication. Performing simultaneous operations in one device is a
desired feature if the associated degradation in performance is not significant.
For best efficiency, the functionality of each individual component should be
maximized before implementing large-scale integration.
(2) Utilize the most of what electronics provides (without violating
philosophy 1) — This is based on the belief that electronics can be competitive
CHAPTER 1: Introduction
21
in cost, reliability and flexibility when the technology is matured. The right
blend of electronics and optics should provide the best solution. By this
argument, optoelectronic approaches should be pushed to their limits before
resorting to all-optical approaches. In the scope of this dissertation, this
philosophy has several interpretations. For OTDM systems, it would be
advantageous to adopt the highest available ETDM base-rate, which is presently
40-Gb/s. In addition, optical gates based on electrically-driven EAMs should be
improved to their limits so that all-optical approaches can aim for speeds beyond
these limits. For clock recovery and 3-R regeneration, optoelectronic approaches
also have their advantages, as will be discussed in detail in the following
chapters.
1.3.2 Overview of chapters
This dissertation can be divided into three parts: optical signal processing
for OTDM (Chapter 1 and 2), optical signal processing for WDM (Chapter 5
and 6), and clock recovery techniques (Chapter 4).
In Chapter 1, 40-GHz optical pulse generation using the TW-EAM is
studied, which is critical for 160-Gb/s OTDM systems with a 40-Gb/s ETDM
base-rate. The traveling-wave phenomena due to the distributed effect in TW-
EAM is observed experimentally and supported by numerical simulations based
on a transmission line model [18]. In order to obtain shorter pulses, a standing-
CHAPTER 1: Introduction
22
wave enhanced mode of the TW-EAM is proposed to make the best use of the
distributed effect [19]. 40-GHz optical pulses as short as 2.4 ps are generated
with a single TW-EAM without post-compression. These studies also make the
TW-EAM a high-speed optical gate for OTDM signal processing.
In Chapter 2, the two most essential optical signal processing for OTDM,
demultiplexing and add-drop multiplexing, are studied. With the standing-wave
enhanced mode, a low-power-penalty 80- to 10-Gb/s demultiplexer is
demonstrated with reduced driving power [20]. The first 160- to 10-Gb/s
demultiplexer using a single EAM is achieved by adding microwave harmonic
frequencies as the driving signal [21]. Experimental comparison with the
previously reported two-stage design shows that better performance is obtained
with fewer components. On the other hand, the first 160-Gb/s OTDM add-drop
multiplexer with a 40-Gb/s base-rate is demonstrated with two standing-wave
enhanced TW-EAMs [22]. Up to date, this remains the only semiconductor-
based time-domain add-drop multiplexer that can operate with a 40-Gb/s base-
rate.
Clock recovery techniques are studied in Chapter 4. Simultaneous
operation of clock recovery and demultiplexing is first demonstrated at 40 Gb/s
by incorporating three simultaneous functions into one TW-EAM (detection,
demultiplexing, and pulse generation). An optoelectronic phase-locked loop is
employed for recovering stable and low-jitter clocks. To increase the line-rate to
CHAPTER 1: Introduction
23
160-Gb/s, a scaled OTDM clock recovery technique is applied and successful
operation is demonstrated [23]. Detailed theoretical modeling of this technique
is presented, with emphasis on how the locking dynamics scales with the line-
rate speed. Another optoelectronic clock recovery technique based on injection-
locking is also studied and characterized experimentally [24]-[25]. The
configuration of this approach is much simpler than the approach based on
phase-locked loop and is utilized in Chapter 6 to realize a very compact 3-R
regenerator.
Chapter 5 studies the use of TW-EAM for wavelength conversion in
WDM systems. A novel photocurrent-assisted mechanism is proposed and
verified for cross-absorption modulation in a TW-EAM and the result is a
photocurrent-assisted wavelength converter (PAW-Converter) [26]. By
optimizing the material and microwave properties of the TW-EAM in great
detail, the operation speed of PAW-Conversion for NRZ data is increased by 4
times from 2.5 Gb/s to 10 Gb/s. Numerical simulations are carried out to study
the cascaded performance of the wavelength converter. An RF-driven approach
is introduced to improve the performance for converting RZ data. It is
demonstrated that the RF-driven PAW-Conversion not only significantly
increases the operating speed but also provides additional re-shaping and re-
timing capabilities [27].
CHAPTER 1: Introduction
24
Several unique 3-R Regeneration approaches based on TW-EAMs are
proposed and demonstrated in Chapter 6. Based on the results of Chapter 5, the
RF-driven PAW-Conversion is merged with the injection locking clock recovery
studied in Chapter 4 to realize a compact 3-R PAW-Regenerator, where thee
essential functionalities required by the 3-R regeneration of RZ data (clock
recovery, pulse generation, and nonlinear gating) are implemented with a single
TW-EAM [28]. Successful regeneration at 10 Gb/s is demonstrated and the
performance limitations are studied in detail with numerical simulations. For
even stronger regeneration capabilities, a simplified optoelectronic 3-R
regenerator is proposed by utilizing the nonlinear E-O transfer function of the
TW-EAM. Very successful 3-R regeneration of ASE noise degraded RZ signal
is obtained [29]. This optical regenerator can operate with the same input and
output wavelength or as a regenerative wavelength converter. Numerical
simulations show excellent cascadibility and very strong regenerative
capabilities on amplitude noise and timing jitter degraded signals. To further
reduce the size of the simplified optoelectronic 3-R regenerator, a monolithically
integrated tunable transmitter, which contains a tunable laser, a semiconductor
optical amplifier, and two TW-EAMs on a single InP chip, is used to replace
discrete devices. The injection locking clock recovery technique is also
incorporated so that only electrical amplifiers and integrable optoelectronic
CHAPTER 1: Introduction
25
devices are needed for all the required 3-R functionalities. Regeneration of ASE
degraded signals is demonstrated.
This dissertation is summarized in Chapter 7 together with suggestions
for future research.
References
[1] E. Deseveire, “Optical Communications in 2025”, plenary talk, Mo2.1.3, European Conference on Optical Communication 2005, Glasgow, Scotland
[2] K. Fukuchi, T. Kasamatsu, M. Morie, R. Ohhira, T. Ito, K. Sekiya, D. Ogasahara, and T. Ono, “10.92 Tb/s (273x40Gb/s) triple-band/ultra-dense WDM optical-repeatered transmission experiment”, Optical Fiber Communication Conference (OFC), paper PD24, 2001
[3] M. Nakazawa, T. Yamamoto, and K. R. Tamura, “1.28Tbit/s-70km OTDM transmission using third- and fourth-order simultaneous dispersion compensation with a phase modulator”, Electron. Lett., vol. 36, no. 24, pp. 2027-2028, 2000
[4] X. Xheng, V. Kaman, S. Yuan, Y. Xu, O. Jerphagnon, A. Keating, R. C. Anderson, H. N. Poulsen, B. Liu, J. R. Sechrist, C. Pusarla, R. Helkey, D. J. Blumenthal, and J. E. Bowers, “Three-Dimensional MEMS Photonic Cross-Connect Switch Design and Performance”, IEEE Journal of Selected Topics in Quantum Electronics, vol. 9, no. 2, pp. 571-578, 2003
[5] S. Zhang, “Traveling-wave Electroabsorption Modulators”, Ph. D. dissertation, University of California, Santa Barbara, July 1999
[6] T. Ido, S. Tanaka, M. Suzuki, Mari Koizumi, H. Sano, and H. Inoue, “Ultra-High-Speed Multiple-Quantum-Well Electro-Absorption Optical Modulators with Integrated Waveguides”, J. Lightwave Technol., vol. 14, no. 9, 2026-2034, 1996
[7] K. Kawano, M. Kohtoku, M. Ueki, T. Ito, S. Kondoh, Y. Noguchi, and Y. Hasumi, “Polarisation-insensitive traveling-wave electrode electroabsorption (TW-EA) modulator with bandwidth over 50GHz and driving voltage less than 2V”, Electron. Lett., vol. 37, no. 5, pp. 299-300,
[8] S. Z. Zhang, Y.-J. Chiu, P. Abraham, and J. E. Bowers, “25-GHz Polarization-Insensitive Electroabsorption Modulators with Traveling-Wave Electrodes”, IEEE Photon. Technol. Lett., vol. 11, no. 2, pp. 191-193, 1999
[9] Y. Akage, K. Kawano, S. Oku, R. Iga, H. Okamoto, Y. Miyamoto, and H. Takeuchi, “Wide bandwidth of over 50GHz traveling-wave electrode electroabsorption modulator integrated DFB lasers”, Electron. Lett., vol. 33, no. 18, pp. 1580-1581, 1997
CHAPTER 1: Introduction
26
[10] G. L. Li, S. A. Pappert, P. Mages, C. K. Sun, W. S. C. Chang, and P. K. L. Yu, “High-Saturation High-Speed Traveling-Wave InGaAsP-InP Electroabsorption Modulator”, IEEE Photon. Technol. Lett., vol. 13, no. 10, pp. 1076-1078, 2001
[11] Y.-J. Chiu, H.-F. Chou, V. Kaman, P. Abraham, and J. E. Bowers, “High Extinction Ratio And Saturation Power Traveling-Wave Electroabsorption Modulator”, IEEE Photon. Technol. Lett., vol. 14, no. 6, pp. 792-794, 2002
[12] S. Irmscher, R. Lewen, and U. Eriksson, “InP-InGaAsP High-Speed Traveling-Wave Electroabsorption Modulators With Integrated Termination Resisrors”, IEEE Photon. Technol. Lett., vol. 14, no. 7, pp. 923-925, 2002
[13] J. Lim, Y.-S. Kang, K.-S. Choi, J.-H. Lee, S.-B. Kim, and J. Kim, “Analysis and Characterization of Traveling-Wave Electrode in Electroabsorption Modulator for Radio-on-Fiber Application”, J. Lightwave Technol., vol. 21, no. 12, pp.3004-3010, 2003
[14] R. Lewen, S. Irmscher, U. Westergren, L. Thylen, and U. Eriksson, “Segmented Transmission-Line Electroabsorption Modulators”, J. Lightwave Technol., vol. 22, no. 1, pp. 172-179, 2004
[15] B. Liu, J. Shim, Y.-J. Chiu, A. Keating, J. Piprek, J. E. Bowers, “Analog Characterization of Low-Voltage MQW Traveling-Wave Electroabsorption Modulators”, J. Lightwaave Technol., vol. 21, no. 12, pp. 3011-3019, 2003
[16] M. N. Sysak, J. S. Barton, L. A. Johansson, J. W. Raring, E. J. Skogen, M. L. Masanovic, D. J. Blumenthal, and L. A. Coldren, “Single-Chip Wavelength Conversion Using a Photocurrent-Driven EAM Integrated With a Widely Tunable Sampled-Grating DBR Laser”, IEEE Photon. Technol. Lett., vol. 16, no. 9, pp. 2093-2095, 2004
[17] J. Livas, “Optical Transmission Evolution: From Digital to Analog to ? Network Tradeoffs Between Optical Transparency and Reduced Regeneration Cost”, J. Lightwaave Technol., vol. 23, no. 1, pp. 219-224, 2005
[18] Hsu-Feng Chou, Yi-Jen Chiu, and John E. Bowers, “40GHz Optical Pulse Generation Using Sinusoidally-Driven Traveling-Wave Electroabsorption Modulator”, Electronics Letters, vol. 38, no. 8, pp. 379-380, 2002
[19] Hsu-Feng Chou, Yi-Jen Chiu, and John E. Bowers, “Standing-Wave Enhanced Electroabsorption Modulator for 40-GHz Optical Pulse Generation”, IEEE Photonics Technology Letters, vol. 15, no. 2, pp. 215-217, 2003
[20] Hsu-Feng Chou, Yi-Jen Chiu, Lavanya Rau, Wei Wang, Suresh Rangarajan, John E. Bowers, and Daniel J. Blumenthal, “Low Power Penalty 80 to 10 Gbit/s OTDM Demultiplexer Using Standing-Wave Enhanced Electroabsorption Modulator With Reduced Driving Voltage”, Electronics Letters, vol. 39, no. 1, pp. 94-95, 2003
[21] Hsu-Feng Chou, Yi-Jen Chiu, Wei Wang, John E. Bowers, and Daniel J. Blumenthal, “Compact 160-Gb/s Demultiplexer Using a Single-Stage Electrically-Gated Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 15, no. 10, pp. 1458-1460, 2003
[22] Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal, “Compact 160-Gb/s Add-Drop Multiplexer With a 40-Gb/s Base Rate Using Electroabsorption Modulators”, IEEE Photonics Technology Letters, vol. 16, no. 6, pp. 1564-1566, 2004
CHAPTER 1: Introduction
27
[23] Hsu-Feng Chou, Zhaoyang Hu, John E. Bowers, Daniel J. Blumenthal, Kohsuke Nishimura, Ryo Inohara, amd Masashi Usami, “Simultaneous 160-Gb/s Demultiplexing and Clock Recovery by Utilizing Microwave Harmonic Frequencies in a Traveling-Wave Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 16, no. 2, pp. 608-610, 2004
[24] Zhaoyang Hu, Hsu-Feng Chou, Kohsuke Nishimura, Masashi Usami, John E. Bowers, and Daniel J. Blumenthal, “Optical Clock Recovery Circuits Using Traveling-Wave Electroabsorption Modulator-Based Ring Oscillators for 3R Regeneration”, IEEE Journal of Selected Topics in Quantum Electronics, vol. 11, no. 2, pp. 329-337, 2005
[25] Zhaoyang Hu, Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal, “40-Gb/s Optical Clock Recovery Using a Traveling-Wave Electroabsorption Modulator-Based Ring Oscillator”, IEEE Photonics Technology Letters, vol. 16, no. 5, pp. 1376-1378, 2004
[26] Hsu-Feng Chou, Yi-Jen Chiu, Adrian Keating, John E. Bowers, and Daniel J. Blumenthal, “Photocurrent-Assisted Wavelength (PAW) Conversion With Electrical Monitoring Capability Using a Traveling-Wave Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 16, no. 2, pp. 530-532, 2004
[27] Hsu-Feng Chou, John E. Bowers, and Daniel J. Blumenthal ,"Novel Photocurrent-assisted wavelength (PAW) converter using a traveling-wave electroabsorption modulator with signal monitoring and regeneration capabilities", Optical Fiber Communication Conference (OFC´04), paper FD4, Feb. 2004, Los Angeles, CA
[28] Hsu-Feng Chou, Zhaoyang Hu, John E. Bowers, and Daniel J. Blumenthal, “Compact Optical 3R Regeneration Using a Traveling-Wave Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 17, no. 2, pp. 486-488, 2005
[29] Hsu-Feng Chou, and John E. Bowers, “Simplified optoelectronic 3R regenerator using nonlinear electro-optical transformation in an electroabsorption modulator”, Optics Express, vol. 13, no. 7, pp. 2742-2746, 2005
CHAPTER 1: Introduction
28
This page intentionally left blank
29
Chapter 2 Optical Pulse Generation 2.1 Introduction
In optical fiber communication, optical pulse generators are needed for
all kinds of return-to-zero (RZ) data format. They also serve as an indispensable
building block for all-optical signal processing in advanced optical
communication systems. At lower transmission speeds such as 10 Gb/s, there is
not much technical difficulty in generating pulses for RZ signal since the
modulator used for modulating the data most likely can generate less than 50%
duty-cycle pulses, especially when aided by the nonlinear electro-optical (E-O)
transformation. For example, by using a LiNbO3 Mach-Zehnder modulator with
a half-rate sinusoidal driving signal, pulses as short as 33% duty-cycle can be
obtained at the data-rate. However, in the case of optical time-division
CHAPTER 2: Optical Pulse Generation
30
multiplexing (OTDM) system, the requirements on pulse generators are much
more stringent and high-performance optical pulse generators are needed.
For OTDM systems, the optical pulses required must have low jitter, a
repetition at the base-rate, and a pulsewidth short enough so that they can be
multiplexed to the line-rate without serious interference between channels. This
means that the duty cycle or the full-width at half-maxima (FWHM) of the
pulses must be scaled down with the number of channels multiplexed. In
addition, not only is the width of the pulses important, but the extinction ratio is
also critical for OTDM applications. A limited extinction ratio will result in
interference between channels. If coherence is not maintained, which is very
likely to happen in a non-integrated multiplexer or in an add-drop node, the
amplitude fluctuation due to interference becomes very fast and behaves like
noise or incoherent crosstalk. A theoretical study showed that for extremely high
extinction ratio pulses, the power penalty due to multiplexing is negligible when
the FWHM of the pulses are less than 40% of the timeslot at line-rate [1]. On the
other hand, for finite extinction ratios, a 4×40 Gb/s OTDM system would
require over 27 dB of extinction in order to keep a power penalty less than 1 dB.
If the FWHM is raised to 50%, the requirement goes up to 29 dB. As the
multiplexing number increases, so does the required extinction ratio [1].
The most widely used optical pulse source is the mode-locked laser. It
can be passively [2] or actively [3] mode-locked. The gain material can be
CHAPTER 2: Optical Pulse Generation
31
Erbium-doped fiber [3], semiconductor optical amplifier (SOA) [4], or solid-
state crystal [2]. The repetition rate can be 10 GHz or 40 GHz and several
products are commercially available nowadays. The pulse qualities of mode-
locked lasers in terms of pulsewidth, extinction ratio, and output power are
generally excellent and they have been used in many high-speed OTDM system
demonstrations. The main concerns of mode-locked lasers are the size of the
unit, the stability and difficulty of locking, and the tuning range in output
wavelength and repetition frequency. To cope with the stability requirement of
communication systems, phase-locked loops and temperature controls are
incorporated in commercial products for long-term stability. The size of mode-
locked lasers can be drastically reduced by monolithic integration. For example,
a semiconductor laser can be integrated with an electroabsorption modulator and
a chirped grating to generate 40 GHz pulses with a 2-ps pulsewidth [5]. An
external cavity version of the mode-locked laser diode can offer more tunability
in output wavelength [6].
To obtain a pulse source that is easier to operate and with more
flexibility in tuning, several direct-modulation pulse sources have also been
studied as alternatives. Those based on electroabsorption modulators (EAM) are
among the most successful ones for OTDM systems. Figure 2.1 shows that with
a simple electrical driving signal at the desired repetition rate, optical pulses can
be obtained by operating the EAM as an optical gate to the continuous wave
CHAPTER 2: Optical Pulse Generation
32
(CW) input [7]. The pulsewidth can be much shorter than that of the electrical
driving signal because of the highly nonlinear E-O transfer function.
Figure 2.1 Generating optical pulses using an electroabsorption modulator
The advantages of using the EAM as an optical pulse generator are the
simplicity and stability of operation, high tunability in wavelength and repetition
rate, and the potential of integration with other optoelectronic components.
Unlike gain-switched lasers that increase timing jitter by itself, the jitter of
EAM-generated pulses is mainly limited by that of the electrical driving signal.
However, the most severe challenge for applications in OTDM systems with
line-rates of 160 Gb/s and beyond is the requirement on pulsewidth.
The duty-cycle of the optical pulse generated by an EAM is normally
10% to 20% with a sinusoidal driving signal, depending on the bandwidth and
CHAPTER 2: Optical Pulse Generation
33
efficiency of the device. Therefore, for 160-Gb/s OTDM with a 10-Gb/s base-
rate (16 times of multiplexing), the pulsewidth is far from adequate. The
required pulsewidth would be around 3 ps FWHM (50% timeslot at 160-Gb/s).
Increasing the repetition rate (base-rate) of the driving signal to 40-GHz (40-
Gb/s) can reduce the multiplexing factor for the same line-rate and relax the
duty-cycle requirement. However, the bandwidth of the EAM may become a
limiting factor at the elevated driving frequency.
Several techniques have been proposed to reduce the pulsewidth of EAM
generated pulses. Depending on the design of the material, if the chirp of the
generated pulses is high, linear compression using dispersion compensating
fibers can be applied and 7-ps pulses were reported at 10 GHz [8]. If nonlinear
compression techniques are utilized so that soliton-like pulses are generated, the
pulsewidth can be further reduced to below 1 ps [9]-[10]. OTDM systems up to
320 Gb/s has been demonstrated with optical pulses compressed by nonlinear
self-phase modulation [11]. However, the size, power consumption, and
complexity of the EAM-based pulse generators are increased significantly.
Another approach is to use tandem EAMs to shorten the pulsewidth,
where the effective gating window can be manipulated by the phase difference
between the two EAMs [12]. An integrated version of this is also reported where
5-ps pulses can be generated at 40 GHz [13]. A 160-Gb/s OTDM system is
demonstrated using two discrete EAMs as the pulse generator, which generates
CHAPTER 2: Optical Pulse Generation
34
3-ps pulses with a 40-Gb/s base-rate [14]. However, the pulses must be
multiplexed with orthogonal polarizations in order to reduce the crosstalk
between channels. In practice, 2-ps pulses would be required for single
polarization multiplexing at 160 Gb/s.
Electrical signal processing techniques can also be utilized to shorten the
pulses generated by the EAM. By adding higher-order harmonics, the original
sinusoidal driving signal becomes pulse-like (if the phases among the harmonics
are properly adjusted) and the carved optical pulses are shortened compared to
the original case. For example, in Ref. [15] the third harmonic frequency is
added and 6% duty-cycle pulses are generated at a 5 GHz repetition rate. The
advantage of this approach is that only one EAM is required. However, the
application is limited by the harmonic frequencies manageable by the EAM and
the driving electronics. For example, at a 40-GHz repetition rate, it is difficult to
apply higher harmonics in a cost-effective way. Therefore, no OTDM systems
were demonstrated with such kind of pulse generator. Nevertheless, in Chapter
3, this technique will be applied at a 10 GHz base-rate to shorten the gating
window width of an optical demultiplexer for 160- to 10-Gb/s demultiplexing
using only one EAM.
This chapter will focus on direct short pulse generation using a single
traveling-wave electroabsorption modulator (TW-EAM). No post-compression
or tandem configuration is considered in order to obtain the simplest
CHAPTER 2: Optical Pulse Generation
35
configuration, which is the most important feature of EAM-based pulse
generator. It is envisioned that 40-Gb/s base-rate is a more economical solution
for practical OTDM systems at 160 Gb/s and beyond. Therefore, the repetition
frequency of particular interest is 40 GHz in this chapter.
There are four factors that determine the pulse shape generated by an
EAM: the modulation efficiency and the total extinction ratio of the EAM, and
the shape and the amplitude of the driving signal. TW-EAMs can be longer than
a lumped EAM and have advantages in modulation efficiency and extinction
ratio, as discussed in Chapter 1. As a result, the first two factors are secured. At
high-frequencies such as 40 GHz, the microwave driving signal is usually a
simple sinusoidal wave so that the shape of the driving signal is fixed. Adding
higher order harmonics is not practical at 40 GHz, as just mentioned above.
Therefore, the remaining factor to be worked on is the amplitude of the driving
signal.
To obtain the best possible pulse generation performance using the TW-
EAM, section 2.2 starts from operating the device in the traveling-wave mode
and studying the phenomena that is associated with the traveling-wave design. A
theoretical mode is also established to explain the observed phenomena and
confirms the traveling-wave operation of the device. In section 2.3, a novel
standing-wave enhanced mode of TW-EAM is proposed and verified both
experimentally and numerically in order to increase the amplitude of the
CHAPTER 2: Optical Pulse Generation
36
electrical driving signal inside the TW-EAM for generating shorter pulses using
only a single device. The improvement in performance provided by the
standing-wave enhanced mode not only benefits short pulse generation but can
also be applied to a broad range of time-domain gating operations in OTDM
systems, as will be explored further in Chapter 3. This is not obtainable with
post-compression techniques. Much of the research presented in this chapter was
published in [16]-[20].
2.2 Traveling-wave phenomena in optical pulse generation
In this section, 40-GHz optical pulse generation using a TW-EAM is
demonstrated and studied. The traveling-wave phenomena observed at high-
frequency is supported by a theoretical model based on the transmission line
theory. The match between theory and experiment not only verifies the
traveling-wave operation of the device but also indicates that the distributed
effect is an important factor in the design of TW-EAM for high-speed operation.
2.2.1 Experimental observations
For traveling-wave optical pulse generation, the TW-EAM is terminated
with a 50-Ohm load and driven by a 5.6Vpp, 40-GHz sinusoidal microwave
signal. A 2-dBm CW at 1555 nm is coupled into the optical input port. As
CHAPTER 2: Optical Pulse Generation
37
shown in Figure 2.2, there are two possible configurations: (1) the optical wave
propagates in the same direction as the microwave along the active waveguide
of TW-EAM, which is termed as co-propagating configuration; (2) the optical
wave propagates in the direction opposite to the microwave, which is termed as
counter-propagating configuration.
Figure 2.2 Configurations of traveling-wave optical pulse generation at 40 GHz. Solid black (gray) arrows: direction of optical transmission in co-propagating (counter-propagating)
configuration; dashed black arrows: direction of microwave propagation
The E-O transfer functions of the TE and the TM polarizations measured
at 2 dBm of input power are shown in Figure 2.3. The TM polarzation is more
efficient and can generate shorter pulses compared to the TE polarization.
However, the loss is higher on the TM polarization, which has a fiber-to-fiber
loss of 13.5 dB. This comes from several origins including the coupling loss, the
scattering loss, and the residual absorption loss of the optical waveguide.
CHAPTER 2: Optical Pulse Generation
38
Figure 2.3 Normalized optical transmission of the TW-EAM with 2 dBm of input power
at 1555 nm for the TE and the TM polarizations
Figure 2.4 Results of 40-GHz optical pulse generation for the two propagation configurations
(a) pulsewidth; (b) output power.
The pulse generation results using the TM polarization mode are shown
in Figure 2.4. The output power of the pulse is measured directly while two
stages of EDFA pre-amplification are used to measure the pulsewidth with a
second-harmonic generation (SHG)-based autocorrelator. Gaussian pulse shapes
are assumed in converting the pulsewidths. It is observed that the results for
these two configurations are different, which is not expected for lumped
CHAPTER 2: Optical Pulse Generation
39
modulators. This is a very unique property of the TW-EAM. In the next section,
a theoretical model will be presented to explain the difference between the two
configurations.
Figure 2.5 (a) optical spectrum with 0.1-nm resolution; (b) autocorrelation trace of the optical
pulse generated at 1 V of reverse bias in the co-propagating configuration
The spectrum and the autocorrelation trace of the pulses generated with
the co-propagating configuration at 1 V of reverse bias are shown in Figure 2.5.
The 40 GHz tones spaced by 0.32 nm are clearly seen in the spectrum. The
autocorrelation trace suggests a high extinction ratio of these pulses. From the
simulation results that will be presented in the next section, the extinction ratios
of the pulses are greater than 30 dB for the co-propagating configuration and 3
to 4 dB smaller for the counter-propagating configuration.
The chirp property can be estimated by measuring the time-bandwidth
product. The value varies from 0.49 to 0.44 depending on the bias voltage,
which indicates the low-chirp feature of these pulses (A transform-limited
Gaussian pulse has a time-bandwidth product of 0.44).
CHAPTER 2: Optical Pulse Generation
40
2.2.2 Modeling of traveling-wave phenomena
To further investigate the traveling-wave phenomena, a theoretical
model based on the microwave transmission line theory is used, which takes into
account the temporal and spatial interactions between the microwave and the
optical wave through the E-O transfer function shown in Figure 2.3.
Figure 2.6 Transmission line model of the TW-EAM operating in the traveling-wave mode
Figure 2.6 shows the transmission line model, where the traveling-wave
electrodes of the TW-EAM are modeled by three sections of transmission line,
each with different microwave properties. The microwave parameters are
obtained from S-parameter measurements and theoretical calculations [21]. Both
the forward-propagating and the backward-propagating sinusoidal waves are
included in the model.
CHAPTER 2: Optical Pulse Generation
41
The microwave voltage distribution along the active waveguide can thus
be expressed as:
( ) )]ee(eRe[Vt,zV )L2z(ztj a⋅−⋅γ⋅γ−ω+ ⋅Γ+⋅⋅=
(2.1)
where V+ is the amplitude of the forward-propagating microwave; Γ is the
reflection coefficient at the end of the active waveguide; γ and La are the
complex propagation constant and the length of the active waveguide; z denotes
the position. The instantaneous optical output power can be derived by summing
over all the losses the optical wave encountered during the propagation through
the active waveguide:
( ) ( )
⋅′α∆−⋅=
aL
0in dzt,zexpPtP
(2. 2)
where Pin is the input CW optical power; t´ is the time at which the optical wave
propagates to position z; ∆α is the incremental loss at position z, which can be
obtained from Figure 2.3 as a function of local voltage. It is assumed that the
transfer function measured at DC is applicable at high frequency. The local
voltage is the sum of the bias voltage and the microwave voltage. Only V+ is
adjusted to match the experimental results, which accounts for the microwave
coupling loss. All other parameters are pre-determined from measurement or fair
CHAPTER 2: Optical Pulse Generation
42
estimation. Note that the velocity mismatch between the microwave and the
optical wave is included in the model through the respective refractive indexes.
The measured results in Figure 2.4 are simulated and the best fitted V+ is
1.8 Vpp for both configurations, showing excellent consistency of the model. The
simulated and the measured results are plotted in Figure 2.7. The close
agreement between the theoretical and experimental results supports the
traveling-wave operation of the device.
(a)
(b)
Figure 2.7 Simulated and measured results of 40-GHz optical pulse generation using a TW-
EAM (a) co-propagating configuration; (b) counter-propagating configuration
CHAPTER 2: Optical Pulse Generation
43
In the proposed transmission line model, the main difference from a
lumped model is the distributed interaction between the microwave and the
optical wave along the active waveguide. By simply considering this effect, the
difference between the two propagation configurations can be explained,
meaning that the distributed effect is a critical factor in describing the operation
of the traveling-wave device at high frequencies.
For the current device, the microwave wavelength in the active
waveguide at 40 GHz is 1500 µm, which means that the 300-µm active
waveguide is about one quarter-wavelength long. This is what makes the
distributed effect noticeable: During the transit through the active waveguide, if
the optical wave propagates in the opposite direction to the microwave, it will
experience significant difference in microwave amplitude. On the contrary, if
they propagate in the same direction (and the velocity mismatch can be
neglected), the optical wave sees the same microwave amplitude all the way
throughout the active waveguide. The difference between the two cases is
further enlarged by the nonlinear E-O transformation. As a result, the output
optical pulses are not the same. The distributed effect takes place when the
microwave wavelength is close to the device dimension. If the device is short or
the microwave wavelength is long (such as in low-speed operation), this
traveling-wave phenomena would not be observed.
CHAPTER 2: Optical Pulse Generation
44
The co-propagating configuration gives about 4 dB higher output power
than the counter-propagating configuration at the same bias. It also generates the
shortest pulse, which is about 4 ps for the current setup. The difference in output
power can be explained by the fact that in the co-propagating configuration the
peak of the pulse propagates with the peak of the microwave and experiences
less absorption. Therefore, it follows naturally that the velocity mismatch
between the optical wave and the microwave ultimately limits the useful device
length.
The fitted V+ value of 1.8 Vpp is smaller than the measured 5.6 Vpp at the
microwave amplifier output (measured with a 50-Ohm load). The difference in
amplitude should mainly come from the microwave coupling loss. Note that the
impedance of the active waveguide is only 25 Ohm, which causes impedance
mismatch. The simulation also suggests that if the microwave amplitude can be
increased inside the TW-EAM, shorter pulses with higher output power can be
obtained.
2.3 Standing-wave enhanced mode of TW-EAM
In order to increase the microwave amplitude inside the device for
generating shorter pulses, a novel standing-wave enhanced mode of the TW-
EAM is proposed. The successful operation of this approach is demonstrated
CHAPTER 2: Optical Pulse Generation
45
experimentally and also verified with numerical simulations. The spirit of this
new operating mode is to make the best use of the distributed effect for single
frequency operation.
2.3.1 Concept
The distributed effect verified in the previous section proves that the
traveling-wave electrode in the TW-EAM behaves like a transmission line.
Then, it follows naturally that one way to increase the amplitude (with the same
input power) is to generate a standing-wave along the transmission line. Ideally,
the maximal voltage swing can be doubled. To increase the microwave
amplitude in the TW-EAM, a dual-drive scheme was proposed where two
synchronized microwave amplifiers are used to drive the TW-EAM from both
ends of the traveling-wave electrodes and 3.6 ps pulses at 40 GHz were
generated [22]. Although not explicitly pointed out in the paper, the dual-drive
scheme equivalently forms a standing-wave pattern along the traveling-wave
electrodes and increases the microwave swing in the active waveguide.
Since the standing-wave concept is supported by the success of the dual-
drive scheme, a standing-wave enhanced mode of the TW-EAM for single
frequency operation is proposed, which reduces the number of microwave
amplifiers and connections by half while generating shorter optical pulse.
CHAPTER 2: Optical Pulse Generation
46
Figure 2.8 shows the microwave envelopes in the cases where the active
waveguide is about one quarter (microwave) wavelength long. Without loss of
generality, the effects of microwave loss and impedance mismatch are neglected
in these plots. Figure 2.8(a) represents the traveling-wave mode of operation.
The microwave envelope, which is defined as the trace of the maximal
microwave swing in space, is flat in this case since the sinusoidal wave travels
trough the device.
Figure 2.8 Microwave envelope in various configurations (a) traveling-wave mode; (b) and (c) standing-wave enhanced mode with different termination line length
However, if the other end of the traveling-wave electrode is left open, as
shown in Figure 2.8(b), it will reflect the microwave back to the input and forms
CHAPTER 2: Optical Pulse Generation
47
a standing-wave pattern along the transmission line. However, in this particular
case, a node happens to fall in the center of the active waveguide and the
effective microwave swing is low therein. It is then necessary to move the
position of the open so that the maximal microwave swing lies within the active
waveguide. Practically, it can be done by ribbon bonding an extra coplanar
waveguide transmission line (CPW line) to the TW-EAM.
2.3.2 Experimental and theoretical verifications
Four devices are prepared with different extension lengths, ranging from
0 to 750 µm. A 2 dBm CW optical input at 1555 nm and a 5.6 Vpp sinusoidal
microwave at 40 GHz are coupled into the TW-EAM. The transmission line
model presented in section 2.2.2 is modified to simulate the standing-wave
enhanced mode as shown in Figure 2.9. The open is assumed to be ideal. The
main difference in actual calculation from the traveling-wave case is that the
value of Γ is changed according to the position of the open. Otherwise, the
programming is basically the same. Co-propagating configuration is assumed,
even though the propagation direction is less sensitive in the standing-wave
enhanced mode. The E-O transfer function that determines the relationship
between ∆α and the local voltage is measured for each TW-EAM individually,
even though there are only minor differences in the high extinction (high bias)
CHAPTER 2: Optical Pulse Generation
48
region. In addition to V+, the termination line length, Lt, is also used as a fitting
parameter due to possible imperfections of the open and the ribbon bonding.
Figure 2.9 Transmission line model of the TW-EAM operating in the standing-wave enhanced mode
The experimental and the simulated results for the four devices are
shown in Figure 2.10(a)-(d), in the order of increasing extension length. Very
close agreement between the simulation and the experiment is obtained. The
termination line length, Lt, obtained by fitting the simulation with the
experiment is 510, 790, 980 and 1050 µm for the four devices, respectively. The
peak-to-peak amplitude of the forward-propagating microwave amplitude, V+, is
also fitted to be 2.8, 2.4, 1.5 and 1.0 Vpp, respectively for the four devices. The
shortest pulse is 3.4 ps, generated by the Lt = 980 µm TW-EAM.
CHAPTER 2: Optical Pulse Generation
49
Figure 2.10 The pulse width and the average power obtained from the experiment (circle) and
the simulation (triangle) for the four devices: (a) Lt = 510 µm, Vpp = 2.8V; (b) Lt = 790 µm, Vpp = 2.4 V; (c) Lt = 980 µm, Vpp = 1.5 V; (d) Lt = 1050 µm, Vpp = 1.0 V.
Solid lines: pulsewidth. Dotted lines: average power
An interesting point of these results is that the TW-EAM with 1.5 Vpp
microwave amplitude performs better than the TW-EAMs with higher
microwave amplitudes, in terms of the shortest possible pulsewidth and the
output power at the same pulsewidth. This can only be explained by the
distributed effect as shown in Figure 2.11. In this figure, the simulated
microwave envelope along the active waveguide is plotted for the four TW-
EAMs. Even though the EAMs with 510 µm and 790 µm termination line
lengths have higher microwave amplitudes, their microwave distributions are
very uneven, which makes it difficult to find one bias voltage to optimize
CHAPTER 2: Optical Pulse Generation
50
different portions of the active waveguide. Those portions with very small
microwave amplitude only give loss to the optical wave and contribute very
little in shaping the output pulse. On the other hand, the device with 1050 µm
termination line length has a very even microwave distribution but its
microwave amplitude is the lowest due to the microwave coupling loss.
Therefore, the best performance is achieve by the device with the Lt = 980 µm
TW-EAM. From Figure 2.11(c), the maximal peak-to-peak microwave swing in
the active waveguide is 2.5 V, which is higher than 1.8 V obtained in the
traveling-wave case (with the same driving amplifier).
Figure 2.11 The simulated microwave envelopes in the active waveguide with parameters
obtained by matching the experimental results with the theoretical model for the four devices. (a) Lt = 510 µm, Vpp = 2.8 V; (b) Lt = 790 µm, Vpp = 2.4 V; (c) Lt = 980 µm, Vpp = 1.5 V;
(d) Lt = 1050 µm, Vpp = 1.0 V.
CHAPTER 2: Optical Pulse Generation
51
It is not difficult to infer from Figure 2.11 that for the standing-wave
enhanced mode, the length of the active waveguide cannot be too long.
Otherwise, the microwave envelope unavoidably varies along the active
waveguide with ineffective low-amplitude portions. The current case of a one
quarter-wavelength long device seems to be a reasonable limit. For device
lengths longer than that, the traveling-wave mode would be more advantageous
but it is under the constraints of optical and microwave propagation losses as
well as velocity mismatch.
Figure 2.12 Spectrum and autocorrelation trace of the pulses generated by the TW-EAM
with Lt = 790 µm at 1.8 V reverse bias
Figure 2.12 shows a typical optical spectrum and the autocorrelation
trace for the 40-GHz pulses generated using the standing-wave enhanced mode.
The measured time-bandwidth product is around 0.43, which is very close to a
transform-limited Gaussian pulse. The low-chirp property of the generated
CHAPTER 2: Optical Pulse Generation
52
pulses comes from the fact that the TW-EAM used in the experiment only
requires a low driving voltage (inside the device) to operate. The microwave
sweeps only around the sharp transition edge of the E-O transfer function, where
the associated index change is the smallest [21].
2.3.3 Generating shorter pulses
Since the theoretical model matches the experiment well, it can be
applied to predict the performance of 40-GHz optical pulse generation using the
standing-wave enhanced mode.
Figure 2.13 Theoretical predictions for the TW-EAM with Lt = 1050 µm at different microwave amplitudes (2V+). Solid lines: pulse width. Dotted lines: average power
Figure 2.13 shows the simulated pulsewidth and the average output
power for a 1050-µm termination line length at several microwave amplitudes.
CHAPTER 2: Optical Pulse Generation
53
To generate less than 3 ps pulses, the peak-to-peak microwave amplitude has to
be over 2 V. Also, to obtain higher output power with the same pulsewidth, the
driving amplitude should be increased.
The current TW-EAM can reach a performance close to the 3 Vpp case in
Figure 2.13 and less than 3 ps pulses can be generated in the wavelength range
of 1545 nm to 1565 nm, as shown in Figure 2.14. In this particular
measurement, a 40-GHz narrowband electrical isolator is inserted between the
TW-EAM and the microwave amplifier for better coupling. The 40-GHz
narrowband amplifier is made by Narda Microwave (Model: DBM-3740N330),
which has an output power over 30 dBm with a 50-Ohm load. The shortest pulse
generated is 2.4 ps with a 40 GHz repetition rate. This is the shortest pulsewidth
reported using a single EAM without any post-compression.
Figure 2.14 Wavelength dependence of 40-GHz optical pulse generation using a standing-wave enhanced TW-EAM regarding (a) pulsewidth; (b) output power. The CW input power is 2 dBm.
The arrows in the right figure indicate the points where the pulsewidth is 3 ps.
CHAPTER 2: Optical Pulse Generation
54
The output power of the current device has wavelength dependence. The
arrows in Figure 2.14(b) indicate the points where less than 3-ps pulses can be
generated with the highest possible output power. As the wavelength decreases,
so does the output power. The current TW-EAM seems to work better in the
higher C-band (or possibly in L-band, which is not tested because of the limited
bandwidth of the measurement system). Band-gap engineering is required to
shift the optimal operating wavelength range.
The maximal output power of around – 20 dBm at 1565 nm is barely
enough for practical applications, not to mention the cases at shorter
wavelengths. The main issue of a low output power is that even with optical
amplification, the OSNR can be degraded and the extinction ratio of the pulse
can be significantly reduced. This could be a problem in OTDM, where about 27
dB of extinction ratio is required for 4×40 Gb/s multiplexing, as discussed in
section 2.1. Increasing the input power can increase the output level, but it is
limited by the power handling capability of the TW-EAM (saturation and
heating effects will eventually occur). The reductions in coupling loss, scattering
loss, and residue absorption loss are required to increase the output power.
A parallel way to improve the output power is to increase the microwave
amplitude inside the TW-EAM. With a properly tuned termination line length,
the spatial phase of the standing-wave pattern can be optimized for the active
waveguide. However, at the optimal termination line length, the microwave
CHAPTER 2: Optical Pulse Generation
55
coupling from the amplifier to the TW-EAM may not also be optimized. That is
the reason why only the 3 Vpp case in Figure 2.13 is experimentally obtained
even with a high-power amplifier. Narrowband impedance matching network is
thus required at the input port of the TW-EAM to improve the coupling
efficiency. The problem in microwave coupling would be greatly reduced if the
characteristic impedance of the active waveguide is 50 Ohm, which still remains
a challenge in the design of TW-EAM [21].
2.4 Summary
Optical pulse generation using the TW-EAM is studied in this chapter
with particular focus at 40 GHz. The traveling-wave phenomena caused by the
distributed effect is demonstrated experimentally and successfully explained
with a transmission line model. To increase the effective microwave amplitude
inside the TW-EAM for generating shorter pulses, a standing-wave enhanced
mode is proposed and verified experimentally and numerically. With a properly
tuned termination line length, pulses as short as 2.4 ps with a 40 GHz repetition
rate were obtained, which is believed to be the shortest optical pulse ever
generated so far by an EAM without post-compression. The proposed standing-
wave enhanced mode is not only useful for short pulse generation but will also
be applied in Chapter 3 for several kinds of optical signal processing in OTDM
systems.
CHAPTER 2: Optical Pulse Generation
56
References
[1] A. T. Clausen, H. N. Poulsen, L. K. Oxenløwe, A. I. Siahlo, J. Seoane, and P. Jeppensen, “Pulse source requirements for OTDM systems”, IEEE LEOS annual meeting, paper TuY2, pp. 382-383, 2003
[2] Ch. Erny, G. J. Spuhler, L. Krainer, R. Paschotta, K. J. Weingarten, and U. Keller, “Simple repetition rate tunable picosecond pulse-generating 10 GHz laser”, Electron. Lett, vol. 40, no. 14, July 2004
[3] A. D. Ellis, R. J. Manning, I. D. Phillips, and D. Nesset, “1.6ps pulse generation at 40GHz in phaselocked ring laser incorporating hightly nonlinear fibre for application to 160Gbit/s OTDM networks”, Electron. Lett, vol. 35, no. 8, pp. 645-646, 1999
[4] J. M. Roth, T. G. Ulmer, N. W. Spellmeyer, S. Constantine, and M. E. Grein, “Wavelegnth-Tunable 40-GHz Picosecond Harmonically Mode-Locked Fiber Laser Source”, IEEE Photon. Technol. Lett., vol. 16, no. 9, pp. 2009-2011, 2004
[5] K. Sato, A. Hirano, M. Asobe, and H. Ishii, “Chirp-compensated 40 GHz semiconductor modelocked lasers integrated with chirped gratings”, Electron. Lett, vol. 34, no. 20, pp. 1944-1946, 1998
[6] Y. Hashimoto, H. Yamada, R. Kuribayashi, and H. Yokoyama, “40-GHz Tunable Optical Pulse Generation from a Highly-Stable External-Cavity Mode-Locked Semiconductor Laser Module”, Optical Fiber Communication Conference (OFC), paper WV5, 2002
[7] S. Oshiba, K. Nakamura, and H. Horikawa, “Low-Drive-Voltage MQW Electroabsorption Modulator for Optical Short-Pulse Generation” J. Quantum Electron., vol. 34, no. 2, pp. 277-281, 1998
[8] D. G. Moodie, A. D. Ellis, and C. W.Ford, “Generation of 6.3ps optical pulses at a 10GHz repetition rate using a packaged electroabsorption modulator and dispersion compensating fibre”, Electron. Lett, vol. 30, no. 20, pp. 1700-1702, 1994
[9] M. J. Guy, S. V. Chernikov, and J. R. Taylor, “A Duration-Tunable, Multiwavelength Pulse Source for OTDM And WDM Communications Systems”, IEEE Photon. Technol. Lett., vol. 9, no. 7, pp. 1017-1019, 1997
[10] P. C. Reeves-Hall and J. R. Taylor, “Wavelength and duration tunable sub-picosecond source using adiabatic Raman compression”, Electron. Lett, vol. 37, no. 7, pp. 417-418, 2001
[11] B. Mikkelsen, G. Raybon, R.-J. Essiambre, A. J. Stentz, T. N. Nielsen, D. W. Peckham, L. Hsu, L. Gruner-Nielsen, K. Dreyer, and J. E. Johnson, “320-Gb/s Single-Channel Pseudolinear Transmission over 200 km of nonzero-Dispersion Fiber”, IEEE Photon. Technol. Lett., vol. 12, no. 10, pp. 1400-1402, 2000
[12] H. Tanaka, S. Takagi, M. Suzuki, and Y. Matsushima, “Optical short pulse generation by double date operation of tandem connected electroabsorption modulators driven by sinusoidal voltages”, Electron. Lett, vol. 29, no. 16, pp. 1449-1451, 1993
[13] V. Kaman, Y.-J. Chiu, T. Liljeberg, S. Z. Zhang, and J. E. Bowers, “Integrated Tandem Traveling-Wave Electroabsorption Modulators for > 100 Gbit/s OTDM applications”, IEEE Photon. Technol. Lett., vol. 12, no. 11, pp. 1471-1473, 2000
CHAPTER 2: Optical Pulse Generation
57
[14] E. Lach, M. Schmidt, K. Schuh, B. Junginger, G. Veith, and P. Nouchi, “Advanced 160 Gbit/s OTDM system based on wavelength transparent 4×40 Gbit/s ETDM transmitters and receivers”, Optical Fiber Communication Conference (OFC), paper TuA2, 2002
[15] H. Tanaka and Y. Matsushima, “Novel Method for Optical Short Pulse Generation Using an EA Modulator Incorporating Microwave Resonators with 3rd Harmonic Sinusoidal Voltage”, Eur. Conf. Optical Communication (ECOC), 1996, Paper ThC.1.4
[16] H.-F. Chou, Y.-J. Chiu, and J. E. Bowers, “Standing-Wave Enhanced Electroabsorption Modulator for 40-GHz Optical Pulse Generation”, IEEE Photonics Technology Letters, vol. 15, no. 2, pp. 215-217, 2003
[17] H.-F. Chou, Y.-J. Chiu, and J. E. Bowers, “40GHz Optical Pulse Generation Using Sinusoidally-Driven Traveling-Wave Electroabsorption Modulator”, Electronics Letters, vol. 38, no. 8, pp. 379-380, 2002
[18] J. E. Bowers, and H.-F. Chou, “Applications of Traveling-Wave Electroabsorption Modualtors in 160-Gbit/s Systems”, European Conference on Optical Communication (ECOC´04), Symposium Tu4.1.1, Sept. 2004, Stockholm, Sweden (invited)
[19] H.-F. Chou, Y.-J. Chiu J. E. Bowers, “Using Standing-wave Electroabsorption Modulators to Generate 40GHz Optical Pulses”, Conference on Laser and Electro-Optics (CLEO´02), paper CMI1, pp. 41-42, Jun. 2002, Long Beach, CA
[20] H.-F. Chou, Y.-J. Chiu, and J. E. Bowers, “40GHz Optical Pulse Generation Using Traveling-Wave Electroabsorption Modulator”, Optical Fiber Communication Conference (OFC´02), paper WV2, pp. 338-339, Mar. 2002, Anaheim, CA
[21] S. Zhang, “Traveling-wave Electroabsorption Modulators”, Ph.D. dissertation, University of California, Santa Barbara, July 1999
[22] V. Kaman, Y.-J. Chiu, S. Z. Zhang, and J. E. Bowers, “3.7 ps pulse generation at
30GHz by dual-drive electroabsoprtion modulator”, Electron. Lett., vol. 36, pp. 1130-1132, 2000
CHAPTER 2: Optical Pulse Generation
58
This page intentionally left blank
59
Chapter 3 OTDM Gating Operations 3.1 Introduction
Time-domain optical signal processing is the core of an optical time-
division multiplexing (OTDM) system, which enables single-wavelength
transmission beyond the speed of available electronics. The two most important
signal processing for OTDM are illustrated in Figure 3.1: demultiplexing and
add-drop multiplexing.
Figure 3.1 Illustration of the two fundamental optical gating operations in an OTDM system: demultiplexing and add-drop multiplexing
CHAPTER 3: OTDM Gating Operations
60
While OTDM multiplexing can be achieved simply by using passive
components such as delay-lines and combiners, demultiplexing of the high-
speed line-rate signal back into base-rate tributaries is quite involved. The
demultiplexing capability directly limits the achievable line-rate speed, without
which the transmitted OTDM signal cannot be converted back into lower-speed
tributaries that can be handled by electronics.
Figure 3.2 Implementation of OTDM signal processing with optical gates (a) demultiplxing and the drop function in ADM; (b) the add function in ADM
With a set of matched multiplexer and demultiplexer, point-to-point
OTDM transmission can be realized. However, to enable networking based on
OTDM, more signal processing capabilities are required. Time-domain add-drop
multiplexing (ADM) is the most important OTDM networking function. There
are two operations performed in an ADM: (1) a base-rate channel is extracted
from the high-speed line-rate signal (the drop function); (2) the timeslot of the
CHAPTER 3: OTDM Gating Operations
61
dropped channel is cleared and a new channel is inserted while the through
channels remain undisturbed (the add function). The drop function is essentially
the same as demultiplexing. These two operations are illustrated in Figure 3.2,
which be implemented with two complementary optical gates. The timing of the
gates must be properly aligned with the channel of interest. Clock recovery is
necessary to acquire synchronization with the input signal, which is the subject
of Chapter 4. The general requirements of these gates are described in Figure
3.3.
Figure 3.3 Requirements of the gating window shape for (a) demultiplexing and the drop function in ADM; (b) the add function in ADM
For demultiplexing and the drop function in ADM (Figure 3.3(a)), the
transmission window should have a width around the timeslot at the line-rate but
with a repetition at the base-rate. This means that when the multiplexing factor
(N) increases, the duty cycle of the optical gate should decrease accordingly.
The extinction ratio of the optical gate must be high enough so that the
neighboring channels are suppressed sufficiently. The exact value of the
required extinction ratio for a given power penalty is related to the quality of the
CHAPTER 3: OTDM Gating Operations
62
input signal (in terms of pulsewidth and extinction ratio). An estimation
provided in Figure 3.37 of Ref. [1] indicates that for N = 4, over 17.5 dB of
extinction ratio is required to achieve less than 1 dB of power penalty. The
requirement in terms of gating window width is more difficult to specify since
there is no general relation between the width and the extinction ratio for all
optical gates. In practice, a rule of thumb for a Gaussian-like gating window is a
full-width at half-maxima (FWHM) at least less than the timeslot at the line-rate
Nevertheless, the extinction ratio must be appropriate.
For the add function in ADM (Figure 3.3(b)), the required gating
window shape is basically complementary to that of the drop function: instead of
one channel extracted, one channel is dropped. Since a new channel will be
inserted into the timeslot of the dropped channel, the extinction ratio for the add
function should be high so that interference between the dropped and the added
channels is minimized. In addition, the transmission for the through channels
must be flat so that variations in power among the through (non-dropped)
channels do not occur, which is important for cascaded operations. Generally
speaking, the gating window of the add function is more demanding than that of
the drop function. In a full add-drop node, both functions must be implemented.
The technologies that have been reported for demultiplexing and ADM are
reviewed below.
CHAPTER 3: OTDM Gating Operations
63
Demultiplexing technologies — OTDM demultiplexers have been
proposed and demonstrated using optical fibers, semiconductor optical
amplifiers (SOAs), and electroabsorption modulators (EAMs). Most of the
demonstrated demultiplexers have a 10 Gb/s base-rate. However, since the 40
Gb/s electrical time-division multiplexing (ETDM) technology is becoming
mature and even commercially available, it is of high interest in recent years to
operate the optical demultiplexer with a 40 Gb/s base-rate.
Fiber-based demultiplexers generally utilize the third-order nonlinearities
in fiber and have excellent potential for ultra-high speed operation because the
response time of fiber nonlinearities is on the order of femto seconds. The gating
operation of the demultiplexer is controlled by optical short pulses so that the
line-rate speed is not limited by electronics. The ultimate limitation on speed is
the walk-off between the OTDM signal and the control pulses. This means that
dispersion must be properly managed for high-speed operations. Highly
nonlinear fibers are generally used so that the fiber length can be kept short
(below 1 km) while keeping a moderate pumping power and well-managed
dispersion.
By using 9 sections of 50 m long dispersion flattened fibers in a
nonlinear optical loop mirror (NOLM), 640- to 10 Gb/s demultiplexing can be
achieved [2]. The total walk-off of the NOLM was designed to be less than 100
fs in order to operate at such high speed.
CHAPTER 3: OTDM Gating Operations
64
Other types of fiber-based optical demultiplexer generally involve
wavelength conversion of the demultiplexed signal. The nonlinearities utilized
include four-wave mixing (FWM) [3] and cross-phase modulation (XPM) [4].
The former was demonstrated up to a 500 Gb/s line-rate while the later was 160
Gb/s. The base-rates were all 10 Gb/s.
FWM in SOAs can also be utilized and both 160- to 10-Gb/s and 160- to
40-Gb/s demultiplexers were demonstrated [5]. However, FWM in SOA is more
complicated than FWM in fibers because of the carrier dynamics that can cause
pattern dependence and optical signal to noise ratio (OSNR) degradation. The
wavelengths of the interacting waves must be properly chosen to optimize the
performance [5]. By using the planar lightwave circuit (PLC) technology, eight
SOAs can be integrated on a single chip with couplers and delay lines to
demultiplex a 160 Gb/s signal into eight 20 Gb/s signals simultaneously [6]. The
possibility of high-level integration is the most attractive advantage of
semiconductor-based technologies.
SOAs can also be configured as an optical switch without wavelength
converting the demultiplexed signals. One advantage is that the demultiplexer
performance has little dependence on the input power of the data since the
gating window is determined by the control pulses. The most successful ones
include SOA-Mach-Zehnder interferometer (SOA-MZI) [7]-[9] and gain-
transparent ultrafast nonlinear interferometer (GT-UNI) [10]. Both of them
CHAPTER 3: OTDM Gating Operations
65
utilize the cross phase modulation (XPM) in SOAs with differential
configurations and have been demonstrated up to 160 Gb/s. The most severe
challenge for SOA-based optical switch is a 100-ps scale carrier lifetime. The
use of the differential scheme is critical to mitigate this problem and obtain high
line-rate operations. In addition, the gain-transparent technique, where the gain
peak of the SOA is shifted from the data wavelength, also helps to reduce the
pattern dependence [9]-[10]. Nevertheless, operating with a 40-Gb/s base-rate
still remains a challenge because of the reduced nonlinear phase shift caused by
the carrier lifetime of SOA even with a differential configuration [8].
The electroabsorption modulator (EAM) can be either an all-optical or
an electrically-driven demultiplexer. In the all-optical case, optical short pulses
are used as control signals. Both FWM and cross-absorption modulation (XAM)
in EAM have been utilized. By using FWM, 80- to 10-Gb/s demultiplexing was
demonstrated with simultaneous timing extraction [11]. On the other hand,
XAM induced by strong control pulses can also make the EAM an all-optical
gate. However, the width of the gating window is limited by the absorption
recovery time. For 160- to 10-Gb/s demultiplexing, XAM-based EAM
demultiplexer must be followed by a fiber-based SPM discriminator in order to
reduce the crosstalk from adjacent channels [12].
An interesting development of EAM-based optical gate is the monolithic
integration of a uni-traveling carrier photodiode (UTC-PD) and a traveling-wave
CHAPTER 3: OTDM Gating Operations
66
EAM (TW-EAM). 320- to 10-Gb/s demultiplexing has been demonstrated [13].
The device as a whole looks like all-optical since it is controlled by optical short
pulses. However, inside the device the TW-EAM is electrically-driven by the
photocurrent generated from the UTC-PD. The monolithic integration makes it
possible to handle the microwave coupling issues between the two elements for
high-speed operation.
OTDM demultiplexers based on electrically-driven EAMs provide the
simplest configuration, where the EAM is driven by a simple electrical signal at
the base-rate. In fact, it works exactly the same way as an optical pulse generator
in Chapter 2 except that the CW input is changed to an OTDM signal. In an
OTDM receiver, clock recovery is required for synchronization. As will be
discussed in detail in Chapter 4, optoelectronic approaches are generally needed
to recover a base-rate clock and the recovered clock is electrical. For all-optical
demultiplexers, a high-quality optical pulse source at the base-rate is needed in
order to generate a synchronized control signal using the recovered electrical
clock. This would greatly increase the complexity and cost of the system. As a
result, electrically-driven EAMs are ideal as compact and low cost
demultiplexers.
The main issue of electrically-driven EAM demultiplexer is that the
gating window width may not be short enough for high-speed operation. With a
10-Gb/s base-rate, the highest line-rate demonstrated was 80 Gb/s, where
CHAPTER 3: OTDM Gating Operations
67
tandem EAMs [14] or a single EAM with a high bias (10 V) and a high driving
voltage (11 Vpp) [15] must be used. To operate at 160 Gb/s, EAM
demultiplexers must operate with a 40 Gb/s base-rate in order to obtain shorter
gating windows. This would not be a problem if the OTDM system is based
entirely on 40-Gb/s ETDM technologies [16]. Otherwise, an additional optical
demultiplexer would be required to bring the signal down to 10 Gb/s [17].
Add-drop multiplexing technologies — An ADM requires both gating
windows shown in Figure 3.3, which means that it is more demanding than a
demultiplexer. ADMs can also be implemented with optical fibers, SOAs, and
EAMs. Using an XPM-based wavelength converter, the drop and the add
functions can be realized by converting the OTDM signal into different
wavelengths with channel configurations specified by the patterns of the control
pulses [18]. Operation at a 40-Gb/s line-rate was demonstrated and this approach
in principle can scale to higher speeds. However, in practice, the flexibility of
this approach is constrained by the need to arrange the control pulses for the two
gating functions individually. A modified approach using band-pass filters to
separate the wavelength shifted and un-shifted signal was proposed and
demonstrated at 80 Gb/s [19]. Only one set of control pulses at the base-rate was
required, reducing the complexity of the previous approach.
Optical switches based on polarization rotation are particularly suitable
for ADM. Once the polarization of the channel intended to be dropped is rotated
CHAPTER 3: OTDM Gating Operations
68
by 90 degrees, a polarization beam splitter can be used to obtain the dropped and
the through channels simultaneously. However, in practice, the polarization may
not be perfectly rotated and two polarizers are needed to optimize the two
functions individually. The main drawback is that the polarization of the OTDM
input signal must be uniform and well controlled. Using a Kerr-gate, 320 Gb/s
ADM with a 40 Gb/s base-rate was demonstrated [20]. On the other hand, 160
Gb/s ADM with a 40 Gb/s base-rate was realized in a NOLM configuration [21].
Semiconductor-based technologies can also be applied for ADM. The
SOA-based GT-UNI introduced above for all-optical demultiplexing can be
modified as an ADM since it also involves polarization rotation [22]-[23].
Successful field trial of this kind of ADM was reported [23]. However, it is very
challenging for these SOA-based ADM to operate with a 40 Gb/s base-rate
because of the limitations imposed by SOA carrier dynamics.
EAM is another choice for semiconductor-based optical switching, and a
40-Gb/s ADM was demonstrated with a 10-Gb/s base-rate [24]. The advantages
of using EAMs include: (1) the switching window is generated without an
interferometer; (2) only an electrical control signal is required. Consequently,
EAMs are promising as a very compact ADM. Nevertheless, the width of the
switching window needed to be shortened in order to scale the operation to
higher line-rates such as 160 Gb/s.
CHAPTER 3: OTDM Gating Operations
69
In this chapter, based on the studies of Chapter 2, the TW-EAMs are
utilized to advance the state of art in OTDM gating operations. In Section 3.2.1
the driving voltage required for EAM-based 80- to 10-Gb/s demultiplexing is
significantly reduced by using a standing-wave enhanced EAM (SW-EAM). In
Section 3.2.2 the first 160- to 10-Gb/s demultiplexer using electrically-driven
EAM is presented by using shaped electrical driving signal. On the other hand,
the first 160-Gb/s ADM with a 40-Gb/s base-rate is demonstrated in Section 3.3
using the standing-wave enhanced TW-EAM developed and optimized in
Chapter 2. Even though several fiber-based approaches with a 40-Gb/s base-rate
were reported after this work, it remains the only semiconductor-based ADM
with a 40-Gb/s base-rate. Much of the research presented in this chapter was
published in [25]-[27].
3.2 Optical demultiplexing
This section is focused on increasing the line-rate speed of OTDM
demultiplexers based on a single electrically-driven EAM. The motivation is the
same as for optical pulse generation in Chapter 2: to achieve faster operation
with a more compact configuration. First, a standing-wave enhanced TW-EAM
is optimized for optical demultiplexing with a 10-Gb/s base-rate. Low-power
penalty 80-Gb/s operation is demonstrated with a reduced driving voltage. To
further increase the line-rate speed to 160 Gb/s, the electrical driving signal is
CHAPTER 3: OTDM Gating Operations
70
shaped by adding harmonic frequencies. Compared to the previously reported
dual-stage approach, a better bit-error-rate (BER) performance is obtained
because of a simpler configuration.
3.2.1 Demultiplexing using the standing-wave enhanced mode
For demultiplexing with a 10-Gb/s base-rate, the EAM must generate a
gating window shorter than the timeslot at the line-rate with a 10-GHz repetition
rate. A simple 10-GHz sinusoidal electrical wave is used as the driving signal.
There are two operation modes available for the TW-EAM: the traveling-wave
mode and the standing-wave enhanced mode, as shown schematically in Figure
3.4.
Figure 3.4 Operation modes for demultplexing with a 10-Gb/s base-rate (a) the traveling-wave
mode; (b) the standing-wave enhanced mode
CHAPTER 3: OTDM Gating Operations
71
In the traveling-wave mode, the EAM is terminated with a 50-Ohm load,
whereas in the standing-wave enhanced mode, an open termination is used
instead to reflect the microwave and forms a standing-wave along the traveling-
wave electrodes. At 10 GHz, the active waveguide (300 µm long) is only 1/20 of
the microwave wavelength. To optimize the microwave distribution in the active
waveguide, the termination CPW line is cleaved shorter so that the open is
closer to the active waveguide (about 150 µm apart). In this way, the
deliberately formed standing-wave pattern can increase the microwave swing
inside the device even at the same driving amplitude.
Figure 3.5 Shortest pulse generated as a function of microwave driving voltage
The advantage of the standing-wave enhanced mode is clearly
demonstrated with a 10 GHz optical pulse generation experiment, which is a
direct measurement of the gating window shape. Figure 3.5 shows the shortest
CHAPTER 3: OTDM Gating Operations
72
10-GHz pulses generated by these two operation modes. The E-O transfer
functions of the TW-EAM at 1555 nm are plotted in Figure 2.3. The modulation
efficiency of the TM polarization is more efficient and generates the shortest
pulses. It is clear from Figure 3.5 that the standing-wave enhanced mode
requires a lower driving voltage for the same pulsewidth. This shows the
advantage of the standing-wave enhanced mode.
Even though the standing-wave enhanced mode can generate pulses
close to 6 ps, which is about the timeslot at 160 Gb/s, the total loss is too high
for good OSNR. Therefore, the demultiplexer is tested at 80 Gb/s, where the
timeslot is 12.5 ps. Figure 3.6 shows the experimental setup for 80- to 10-Gb/s
demultiplexing. A 10-GHz, 5-ps pulse train at 1555 nm is generated from a
mode-locked fiber ring laser and compressed nonlinearly to 2 ps with 5 km of
dispersion shifted fiber (DSF). It is modulated by a LiNbO3 modulator with 231-
1 pseudo-random binary sequence (PRBS) and then optically multiplexed to 80
Gb/s. The standing-wave enhanced TW-EAM is biased at – 4 V and driven by a
6.4Vpp, 10-GHz microwave. The gating window is 7.3 ps for the TM
polarization and 10.6 ps for the TE polarization. Although the gating window of
the TM polarization is shorter, the output power is 10 dB lower than that of the
TE polarization. Therefore, TE polarization is chosen in the experiment to get a
better OSNR. A phase shifter is adjusted manually to synchronize the gating
window with the channel to be demultiplexed.
CHAPTER 3: OTDM Gating Operations
73
Figure 3.6 Experimental setup of 80- to 10-Gb/s demultiplexing using a standing-wave enhanced TW-EAM
Figure 3.7 shows both the input 80-Gb/s and the demultiplexed 10-Gb/s
eyes, detected with a 50-GHz oscilloscope and a 40-GHz photodetector. The
demultiplexed 10Gb/s eye is clean and open. The ripples are due to the response
of the photodetector. Bit error rate (BER) curves and receiver sensitivities at 10-9
BER are shown in Figure 3.8. The power penalty is low and varies slightly from
0.4 to 0.7 dB. The averaged value is 0.55 dB. No error-floor is observed and the
slope of the demultiplexed curve is the same as the back-to-back curve,
indicating excellent SNR. The previously reported 80- to 10-Gb/s EAM-based
demultiplexer requires a high reverse bias of – 10 V and a high driving voltage
of 11 Vpp [15]. These results well demonstrate the advantage of the standing-
wave enhanced design.
CHAPTER 3: OTDM Gating Operations
74
Figure 3.7 Eye diagrams of the 80-Gb/s line-rate signal and the demultiplexed 10-Gb/s signal
Figure 3.8 BER curves of the back-to-back 10-Gb/s signal and the 10-Gb/s signal demultiplexed from the 80-Gb/s OTDM signal. Insert: individual receiver sensitivity of the eight channels.
3.2.2 Demultiplexing using a shaped electrical driving signal
To extend the line-rate of the electrically-driven EAM demultiplexer to
160 Gb/s, the gating widow width needs to be shortened to less than 6.25 ps. In
[16] and [17], the base-rate is increased to 40-Gb/s to this purpose because the
gating window width decreases as the driving frequency increases. However,
another stage of either electrical [16] or optical [17] demultiplexer is required to
bring the signal down to 10 Gb/s. For an OTDM system with a 10-Gb/s base-
CHAPTER 3: OTDM Gating Operations
75
rate, this may not be an efficient solution. Therefore, in this section, microwave
harmonic frequencies are mixed to generate a shortened gating window [28] and
a single-stage 160- to 10-Gb/s demultiplexer is obtained.
As shown in Figure 3.9(a), when a 10 GHz and a 20 GHz microwave are
added in phase, the result is a pulse-like electrical waveform with a repetition
rate of 10 GHz but a width close to that of the 20 GHz wave (depending on the
relative strengths of the two waves). Adding more or higher order harmonics can
in principle shorten the pulsewidth further but at the expense of a more
complicated setup.
Figure 3.9 (a) Signal shaping by adding harmonic frequencies; (b) driving configuration of the TW-EAM
The two microwave frequencies are coupled into the TW-EAM from two
opposite electrical ports as shown in Figure 3.9(b). This is a unique
configuration for the TW-EAM where the microwave loss that may occur when
CHAPTER 3: OTDM Gating Operations
76
the two waves are combined with a coupler can be eliminated. This ensures the
shortest possible pulsewidth with the same driving amplifiers.
The gating window of this harmonically driven TW-EAM can be
accessed through the pulse generation experiment, where the TW-EAM is
driven by a 24.6 dBm, 10 GHz microwave from one port and a 23.4 dBm, 20
GHz microwave from the other. The optical wave is arranged to co-propagate
with the 20 GHz microwave. Figure 3.10(a) shows both the pulsewidth and the
output power of the generated 10-GHz pulses. The input optical power is 2 dBm
at 1555 nm. At 6.5 V of reverse bias, the pulsewith of the TE polarization is 5.7
ps and 4.0 ps for the TM polarization, both are shorter than the timeslot at 160
Gb/s. Although the TM polarization has shorter pulsewidth, its output power is
more than 10 dB lower then the TE polarization. Therefore, TE polarization is
adopted to ensure adequate OSNR.
Figure 3.10 Pulsewidth and output power in the TE and the TM polarizations at 1555 nm using (a) a TW-EAM driving by both 10 GHz and 20 GHz microwaves in the traveling-wave mode;
(b) a standing-wave enhanced TW-EAM driven at 40 GHz
CHAPTER 3: OTDM Gating Operations
77
The setup of the 160- to 10-Gb/s single-stage demultiplexer is shown in
Figure 3.11(a). Two electrical isolators are used as terminations for the two
counter-propagating microwave harmonics.
Figure 3.11 Setup of (a) the single-stage demultiplexer; (b) the dual-stage demultiplexer
for 160- to 10 Gb/s demultiplexing
For comparison, a dual-stage 160- to 10-Gb/s demultiplexer similar to
the one in [17] is also constructed, as shown in Figure 3.11(b), using two
standing-wave enhanced TW-EAMs (SW-EAMs). SW-EAM1 and SW-EAM2
are optimized for 40 GHz and 10 GHz gating operations with termination line
CHAPTER 3: OTDM Gating Operations
78
lengths of 840 µm and 150 µm, respectively. The microwave driving powers are
19 dBm at 40 GHz and 20 dBm at 10 GHz. Figure 3.10(b) shows the 40-GHz
pulse generation results of SW-EAM1. The gating window is 4.6 ps (TE
polarization) at 1.6 V of reverse bias. For SW-EAM2, the reverse bias is set at
4.1 V and the gating window is 11 ps (TE polarization), suitable for 40- to 10
Gb/s demultiplexing. In the experiment, the output power of SW-EAM1 is no
more than – 20 dBm and an EDFA is inserted between the two SW-EAMs to
compensate for the loss.
Figure 3.12 Experimental setup for 160- to 10 Gb/s demultiplexing
The system configuration of the experiment is shown in Figure 3.12. 10-
GHz, 5-ps optical pulses are generated at 1555 nm using a mode-locked fiber
ring laser (MLFRL). These pulses are nonlinearly compressed using 5 km of
dispersion-shifted fiber (DSF). The pulse width after compression is around 2
ps. The 10GHz pulse train is encoded with 231-1 PRBS by a LiNbO3 modulator
(LN-MOD) and then optically multiplexed to 160 Gb/s with passive delay lines
CHAPTER 3: OTDM Gating Operations
79
and couplers. In the experiment, the 160 Gb/s signal is multiplexed with
alternating polarization because the compressed pulse has relatively long tails.
Single polarization multiplexing in this case would lead to severe intersymbol
interference (ISI) and error-free operation cannot be obtained due to the
limitation of the pulse source.
The eye diagrams at different bit-rates are shown in Figure 3.13. They
are taken with a 50 GHz digital sampling scope and a 40 GHz photodetector.
The 160 Gb/s eye can hardly be seen without using histogram. The directly
demultiplexed 10-Gb/s eye (10:160 Gb/s eye) using the single-stage
demultiplexer is relatively clear and open. The bumps on the ground level are
caused by the electrical reflections in the photodetector which is evident from its
impulse response.
Figure 3.13 Eye histograms of the input 160 Gb/s OTDM signal and the demultiplexed signals at 40 Gb/s and 10 Gb/s
CHAPTER 3: OTDM Gating Operations
80
In the dual-stage demultiplexer, the 160 Gb/s signal is first demultplexed
to 40 Gb/s by SW-EAM1 and then to 10 Gb/s by SW-EAM2. The response of
the photodetector leads to a partial closure of the 40 Gb/s eye (40:160 Gb/s eye)
but when the signal is further demultiplexed by SW-EAM2, the 10 Gb/s eye
(10:40:160 Gb/s eye) opens up. However, the 10:40:160 Gb/s eye is noisier than
the 10:160 Gb/s eye.
Figure 3.14 shows the BER curves of the two demultiplexers. Error-free
operations are achieved and no error-floor is observed. The receiver sensitivities
of all 16 channels were measured individually and the variation among them is
caused by the imperfection of the multiplexer. The averaged power penalty for
the single-stage demultiplexer is 1 dB and 2.8 dB for the dual-stage
demultiplexer. It is evident from Figure 3.14 that the slope of the BER curve of
the single-stage demultiplexer is almost the same as that of the 10 Gb/s back-to-
back signal (without multiplexing, demultiplexing, and two EDFAs). On the
other hand, the slope of the dual-stage demultiplexer decreases, indicating the
increase of noise in the demultiplexed signal.
The 1-dB power penalty of the single-stage demultiplexer mainly comes
from the ISI caused by the long tails of the compressed pulse and a relatively
wide gating window width (5.7 ps). The polarization dependence of the TW-
EAM actually helped to minimize the effect of ISI because the adjacent channels
are TM polarized, which experiences higher loss. However, in the experiment,
CHAPTER 3: OTDM Gating Operations
81
the polarizations in the multiplexer may not be perfectly aligned, so some
residual ISI may still occur.
Figure 3.14 BER curves of the single-stage and the dual-stage 160- to 10 Gb/s demultiplexers.
The insert shows the receiver sensitivity of all 16 channels for both demultiplexers.
For the dual-stage demultiplexer, the EDFA for loss compensation can
degrade the OSNR because the output power from SW-EAM1 is no more than –
20 dBm. Therefore, the power penalty is higher and the demultiplexed eye
shows more beating noise. Increasing the optical input power beyond 5 dBm to
SW-EAM1 might improve the OSNR but the TW-EAM will suffer from pattern
dependence caused by the stronger photocurrent bouncing back and fourth along
the traveling-wave electrodes (note: this photocurrent induced effect will be
utilized in later chapters). This clearly shows the advantage of the single stage
demultiplexer since the extra EDFA and optical filter are not needed, which
yields better performance with a simpler configuration.
CHAPTER 3: OTDM Gating Operations
82
The operating wavelength range of the single-stage demultiplexer can be
measured by systematic pulse generation experiments at different wavelengths.
By setting the gating window width to be less than 5.7 ps and the output power
higher than the receiver sensitivity, the single-stage demultiplexer is able to
operate from 1545 nm to 1565 nm. The current polarization dependence of the
EAM can be reduced by properly compensating the strain in the quantum wells
[29]. To further shorten the gating window of the single-stage demulitplexer, the
amplitude of the 20 GHz driving signal inside the TW-EAM should be
increased. This can be done with a higher power amplifier but a more efficient
way is to improve the microwave coupling efficiency at 20 GHz. Another
approach is to add higher harmonics at 30 GHz or 40 GHz, but the coupling
problem would be more severe at elevated frequencies.
The results presented in this section demonstrate that electrically-driven
EAMs are capable of demultiplexing 160 Gb/s signals down to either 40 Gb/s or
10 Gb/s efficiently using a single EAM.
3.3 Time-domain add-drop multiplexing
In this section, a full 160-Gb/s time-domain add-drop multiplexer
(ADM) is demonstrated with a 40-Gb/s base-rate using standing-wave enhanced
TW-EAMs introduced and studied in Chapter 2. The standing-wave enhanced
CHAPTER 3: OTDM Gating Operations
83
design is essential to push the operating speed of the EAM to such high-bit rates.
The previously demonstrated EAM-based ADM only operated at a 40 Gb/s line-
rate with a 10 Gb/s base-rate [24]. Even though SOA-based technology can
operate at a 160-Gb/s line-rate, the base-rate is limited to 10-Gb/s due to the
carrier lifetime in the SOA. Compared to fiber-based ADMs, the electrically-
driven, EAM-based ADM have advantages of compactness and efficiency.
Figure 3.15 Experimental setup of the 160-Gb/s ADM with a 40-Gb/s base-rate using standing-wave enhanced TW-EAMs. Solid line: optical link. Dotted line: electrical link
The configuration of the 40Gb/s-based ADM is shown schematically in
Figure 3.15. Two standing-wave enhanced TW-EAMs are used to implement the
drop function (EAM1) and the add function (EAM2) individually. The spatial
phase of both EAMs is adjusted with a 500-µm extension CPW line so that the
microwave amplitude and distribution can be optimized in the active waveguide.
A new channel can be added with the through channels passively with a coupler
and a delay line. The only differences between the two EAMs are the bias
CHAPTER 3: OTDM Gating Operations
84
voltage and the phase of the 40-GHz driving signal, which implies that single-
chip integration is possible.
The static fiber to fiber transmissions for the TE and TM polarizations at
1557.5 nm are shown in Figure 3.16(a). The TE polarization is chosen for the
add function and TM for the drop function, which optimizes the gating windows
for the respective operations. The bias voltage is 0 V for the add function and –
1 V for the drop function. The EAMs are driven by a 6 Vpp, 40-GHz sinusoidal
microwave.
Figure 3.16 (a) Static transmissions of the TE and the TM polarizations. The bias points for the add and the drop functions are also indicated; (b) Dynamic transmission at 40 GHz measured by
scanning a 1.5 ps pulse train. The vertical lines are spaced by 6.25 ps (timeslot at 160 Gb/s)
By scanning a 1.5-ps, 10-GHz optical pulse train through the EAMs, the
gating windows can be probed as shown in Figure 3.16(b), which represents the
closest estimation of the actual gating performance since the pulse source is the
same as that used in the transmitter. The traces are shifted by 180 degrees in
CHAPTER 3: OTDM Gating Operations
85
phase (12.5 ps) relative to each other for clear illustration. Due to the high
modulation efficiency of the TW-EAM, the gating window shape can be
switched between the two operating modes (add and drop) by changing only 1 V
in bias voltage.
The 3-dB width of the gating window is 11.7 ps for the add function and
5.2 ps for the drop function. If the polarizations are switched, the gating
windows will be broadened to 13.3 ps and 6.8 ps, respectively. The suppression
of adjacent channels is over 16 dB in the drop function and the clearing of the
targeted timeslot is over 21 dB in the add function. However, there can be a
variation of 1.9 to 2.6 dB in power among the through channels in the add
function. This should not be a fundamental limitation of EAM-based ADM and
can be reduced by moving the bias point upwards along the transmission curve
in Figure 3.16(a). However, to maintain the same extinction ratio, the effective
driving voltage must be increased by either a higher driving power or an
impedance matching network, as discussed in Section 2.3.3.
Referring to Figure 3.15, a 40-Gb/s NRZ electrical signal is generated
from four 10-Gb/s, 27-1 pseudo-random-binary-sequence (PRBS) tributaries
using an electrical multiplexer (SHF 4005A). Operation with a longer pattern
length imposes a power penalty because of the pattern dependency of the
electrical multiplexer. The 40-Gb/s electrical NRZ signal drives a LiNbO3
modulator to encode a 40-GHz optical pulse train, which is passively
CHAPTER 3: OTDM Gating Operations
86
multiplexed from a 10-GHz, 1.5-ps pulse train generated by a mode-locked fiber
laser centered at 1557.5 nm. The pulse tail extinction ratio is better than that
used in the 160-Gb/s demultiplexing experiment in the previous section. The
resulting 40-Gb/s RZ optical signal is split into two parts. One is further
multiplexed to 160-Gb/s with single polarization and the other is used as the add
channel. The PRBS character is not truly preserved at 160-Gb/s but several
meters of delay are used to de-correlate the tributaries.
The phase delay, φ, of the 40-GHz driving signal of EAM1 is 180
degrees shifted relative to that of EAM2 in order to implement ADM on the
same channel. An advantage of the presented approach is the flexibility of
simultaneous add and drop on different channels by changing the value of φ.
On the receiver side, EAM3 is used to demultiplex 160-Gb/s signals
back to 40-Gb/s. Note that EAM3 in the receiver (for demultiplexing) and
EAM1 in the ADM (for the drop function) have the same gating window shape.
The 40-Gb/s receiver is composed of a 40- to 10-Gb/s optical demultiplexer
(EAM4) and a 10-Gb/s electrical receiver. BER is measured at 10 Gb/s for all 16
channels. The transmitter and the receiver systems can be fully 40-Gb/s if a 40-
Gb/s pattern generator and a 40-Gb/s BER tester are available. Nevertheless, the
ADM itself is completely working with a 40-Gb/s base-rate.
CHAPTER 3: OTDM Gating Operations
87
The input average power levels to EAM1, EAM2, and EAM3 in the
experiment are approximately – 1 dBm per 40-Gb/s channel, corresponding to a
peak power of 14.2 dBm.
Figure 3.17 40-Gb/s eye diagrams of the four channels in the 160-Gb/s signal (a) back to back;
(b)after ch.2 is dropped; (c) after a new 40-Gb/s channel is added.
The eye diagrams of the 160-Gb/s signals cannot be fully resolved by
using a 50-GHz electrical sampling scope with a 40-GHz photodetector, as is
evident in Figure 3.13. To mitigate this problem, the 160-Gb/s signals are
sampled after the 160- to 40-Gb/s demultiplexer (EAM3), which extracts the
four 40-Gb/s channels individually. Figure 3.17(a) shows the eye diagrams of
the four channels in the back-to-back 160-Gb/s signal. The bumps on the bottom
of the eyes are caused by the response of the photodetector. After EAM2, a 40-
Gb/s channel (ch.2) is cleared, as shown in Figure 3.17(b). The eye amplitude of
CHAPTER 3: OTDM Gating Operations
88
ch.4 is about 2 dB higher than the other two through channels, in agreement with
the pulse scanning results in Figure 3.16(b). A new 40-Gb/s channel is then
added to the cleared timeslot in the same polarization with the through channels.
Figure 3.17(c) shows no observable sign of interference.
Figure 3.18 BER curves measured at 10 Gb/s. The insert shows the receiver sensitivities of the
four 10-Gb/s tributaries in each 40-Gb/s channel
Figure 3.18 shows the results of BER measurements. The received power
is measured at the input of the 10-Gb/s receiver. Power penalty and receiver
sensitivity are measured at a BER of 10-9. The 40-Gb/s line in the figure is
obtained by sending a 40-Gb/s signal to the 160-Gb/s receiver (EAM3 plus the
40-Gb/s receiver). There is 1-dB power penalty when the input is changed to the
160-Gb/s back-to-back (BtB) signal. This penalty mainly comes from the finite
suppression ratio of the 160- to 40-Gb/s demultiplexer (Figure 3.16(b)). Note
CHAPTER 3: OTDM Gating Operations
89
that the 160-Gb/s BtB line also represents the BER result for the dropped
channel since the operation of EAM1 (drop) and EAM3 (demultiplexing) are
identical as mentioned above.
After ADM, the averaged power penalty for the four 40-Gb/s channels
(ch.1 to 4) are 1.3, 0.8, 1.4 and 0.7 dB, respectively, which results in an overall
power penalty of 1 dB. A low power penalty of 0.8 dB for the added channel
(ch.2) indicates that the clearing of the timeslot in the add function is complete
and the interference with the residue of the dropped channel is negligible. The 2-
dB variation in power among the through channels (ch.1, ch.3 and ch.4) only
results in a 0.6 to 0.7 dB difference in power penalty, which is believed to be the
consequence of the varied signal-to-noise ratio due to the power variation.
3.4 Summary
TW-EAMs are applied in this chapter to advance the state-of-the-art in
OTDM gating operations. The electrically-driven demultiplexers and add-drop
multiplexer presented in this chapter have great advantages in simplicity, size,
performance, and efficiency when compared to all-optical approaches.
The standing-wave enhanced mode proposed and studied in Chapter 2
helps to reduce the driving voltage required for 80- to 10-Gb/s demultiplexing.
For single-stage 160- to 10-Gb/s demultiplexing, microwave harmonics are
added in the traveling-wave mode to shape the driving signal for shorter gating
CHAPTER 3: OTDM Gating Operations
90
windows. On the other hand, the first 160-Gb/s ADM with a 40-Gb/s base-rate
is demonstrated with a pair of well-optimized standing-wave enhanced TW-
EAMs. All demonstrations presented in this chapter have an averaged power
penalty at or less than 1 dB for 80-Gb/s or 160-Gb/s line-rate operations,
showing excellent performance of these proposed approaches.
References
[1] V. Kaman, “High-Speed Optical Systems Based on Electroabsorption Modulators”, Ph.D. Dissertation, University of California, Santa Barbara, December 2000
[2] T. Yamamoto, E. Yoshida, and M. Nakazawa, “Ultrafast nonlinear optical loop mirror for demultiplexing 640 Gbit/s TDM signals”, Electron. Lett., vol. 34, no. 10, pp. 1013-1014, 1998
[3] T. Morioka, H. Takara, S. Kawanishi, T. Kitoh, and M. Saruwatari, “Error-free 500Gbit/s all-optical demultiplexing using low-noise, low-jitter supercontinuum short pulses”, Electron. Lett., vol. 32, no. 9, pp. 833-834, 1996
[4] J. Li, B.-E. Olsson, M. Karlsson, and P. Andrekson, “OTDM Demultiplexer Based on XPM-Induced Wavelegnth Shifting in Highly Nonlinear Fiber”, IEEE Photon. Technol. Lett., vol. 15, no. 12, pp. 1770-1772, 2003
[5] S. L. Jansen, G. Khoe, H. de Waardt, M. Heid, S. Spalter, E. Meissner, C. Weiske, and A. Schopflin, “Optimizing the Wavelength Configuration for FWM-Based Demultiplexing in a SOA”, Optical Fiber Communication Conference (OFC), paper ThO5, 2003
[6] I. Shake, H. Takara, K. Uchiyama, I. Ogawa, T. Kitoh, T. Kitagawa, M. Okamoto, K. Magari, Y. Suzuki, and T. Morioka, “160 Gbit/s fill optical time-division demultiplexing using FWM of SOA-array integrated on PLC”, Electron. Lett., vol. 38, no. 1, pp. 37-38, 2002
[7] S. Nakamura, Y. Ueno, K. Tajima, J. Sasaki, T. Sugimoto, T. Kato, T. Shimoda, M. Itoh, H. Hatakeyama, T. Tamanuki, and T. Sasaki, “Demultiplexing of 168-Gb/s Data Pulses with a Hybrid-Integrated Symmetric Mach-Zehnder All-Optical Switch”, IEEE Photon. Technol. Lett., vol. 12, no. 4, pp. 425-427, 2000
[8] M. Heid, S. L. Jansen, S. Spalter, E. Meissner, W. Vogt, H. Melchior, “160-Gbit/s demultiplexing to base rate of 10 and 40 Gbit/s with a monolithically integrated SOA-Mach-Zehnder Interferometer”, Proc. ECOC 2002, vol. 3, paper 8.4.3, 2002
[9] S. Diez, C. Schubert, R. Ludwig, H.-J. Ehrke, U. Feiste, C. Schmidt, and H. G. Weber, “160 Gbit/s all-optical demultiplexer using hybrid gain-transparent SOA Mach-Zehnder interferometer”, Electron. Lett., vol. 36, no. 17, pp. 1484-1486, 2000
CHAPTER 3: OTDM Gating Operations
91
[10] C. Schubert, S. Diez, J. Berger, R. Ludwig, U. Feiste, H. G. Weber, G. Toptchiyski, K. Petermann, and V. Krajinovic, “160-Gb/s All-Optical Demultiplexing Using a Gain-Transparent Ultrafast-Nonlinear Interferometer (GT-UNI)”, IEEE Photon. Technol. Lett., vol. 13, no. 5, pp. 475-477, 2001
[11] E. S. Awad, P. S. Cho, and J. Goldhar, “Simultaneous Four-Wave Mixing and Cross-Absorption Modulation Inside a Single EAM for High-Speed Optical Demultiplexing and Clcok Recovery”, IEEE Photon. Technol. Lett., vol. 17, no. 7, pp. 1534-1536, 2005
[12] T. Miyazaki and F. Kubota, “All-Optical Reshaping in a 160-Gb/s OTDM Receiver Using an XAM Gate Followed by an SPM Discriminator”, IEEE Photon. Technol. Lett., vol. 16, no. 8, pp. 1909-1911, 2004
[13] S. Kodama, T. Yoshimatsu, T. Ito, and H. Ito, “320-Gbit/s Error-Free Demultiplexing Operation of a monolithic PD-EAM Optical Gate”, Optical Fiber Communication Conference (OFC), paper ThX5, 2003
[14] J. Yu, K. Kojima, N. Chand, A. Ougazzaden, C. W. Lentz, J. M. Geary, J. M. Freund, and B. Mason, “Simultaneous Demultiplexing and Clock Recovery of 80 Gb/s OTDM Signals Using a Tandem Electro-absorption Modulator”, Proc. IEEE/LEOS Annual Meeting 2001, 1, paper TuBB4, pp. 358-359
[15] D. D. Marcenac, A. D. Ellis, and D. G. Moodie, “80Gbit/s OTDM using electroabsorption modulators”, Electron. Lett., vol. 34, pp. 101-103, 1998
[16] E. Lach, M. Schmidt, K. Schuh, B. Junginger, G. Veith, P. Nouchi, “Advanced 160 Gbit/s OTDM system based on wavelength transparent 4 × 40 Gbit/s ETDM transmitters and receivers”, OFC 2002, paper TuA2, 2002
[17] B. Mikkelsen, G. Raybon, and R.-J. Essiambre , “160 Gb/s TDM Transmission Systems”, Proc. ECOC 2000, vol. 2, pp. 125-128, 2000
[18] L. Rau, S. Rangarajan, W. Wang, and D. J. Blumenthal, “All-optical add-drop of an OTDM channel using an ultra-fast fiber based wavelength converter”, OFC 2002, paper WM1, 2002
[19] J. Li, B.-E. Olsson, M. Karlsson, and P. A. Andrekson, “OTDM Add-Drop Multiplexer Based on XPM-Induced Wavelegnth Shifting in Highly Nonlinear Fiber”, J. Lightwave. Technol., vol. 23, no. 9, pp. 2654-2661, 2005
[20] C. Schubert, C. Schmidt-Langhorst, K. Schulze, V. Marembert, and H. G. Weber, “Time Division Add-Drop Multiplexing up to 320 Gbit/s”, OFC 2005, apepr OThN2
[21] E. J. M. Verdurmen, Y. Zhao, E. Tangdiongga, J. P. Turkiewicz, G. D. Khoe, H. de Waardt, “Error-free all-optical add-drop multiplexing using HNLF in a NOLM at 160 Gbit/s”, Electron. Lett., vol. 41, no. 6, pp. 349-350, 2005
[22] C. Schubert, C. Schmidt, S. Ferber, R. Ludwig, and H. G. Weber, “Error-free all optical add-drop multiplexing at 160 Gbit/s”, Electron. Lett., vol. 39, no. 14, pp. 1074-1076, Jul. 2003
[23] J. P. Turkiewicz, E. Tangdiongga, G. Lehmann, H. Rohde, W. Schairer, Y. R. Zhou, E. S. R. Sirkra, A. Lord, D. B. Payne, G.-D. Khoe, and H. de Waardt, “160 Gb/s OTDM Networking Using Deployed Fiber”, J. Lightwave. Technol., vol. 23, no. 1, pp. 225-235, 2005
CHAPTER 3: OTDM Gating Operations
92
[24] I. D. Phillips, A.Gloag, D. G. Moodie, N. J. Doran, I. Bennion, and A. D. Ellis, “Drop and Insert Multiplexing with Simultaneous Clock Recovery Using an Electroabsorption Modulator”, IEEE Photon. Technol. Lett., vol. 10, no. 2, pp. 291-293, Feb. 1998
[25] H.-F. Chou, Y.-J. Chiu, L. Rau, W. Wang, S. Rangarajan, J. E. Bowers, and D. J. Blumenthal, “Low Power Penalty 80 to 10 Gbit/s OTDM Demultiplexer Using Standing-Wave Enhanced Electroabsorption Modulator With Reduced Driving Voltage”, Electronics Letters, vol. 39, no. 1, pp. 94-95, 2003
[26] H.-F. Chou, Y.-J. Chiu, W. Wang, J. E. Bowers, and D. J. Blumenthal, “Compact 160-Gb/s Demultiplexer Using a Single-Stage Electrically-Gated Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 15, no. 10, pp. 1458-1460, 2003
[27] H.-F. Chou, J. E. Bowers, and D. J. Blumenthal, “Compact 160-Gb/s Add-Drop Multiplexer With a 40-Gb/s Base Rate Using Electroabsorption Modulators”, IEEE Photonics Technology Letters, vol. 16, no. 6, pp. 1564-1566, 2004
[28] H. Tanaka and Y. Matsushima, “Novel Method for Optical Short Pulse Generation Using an EA Modulator Incorporating Microwave Resonators with 3rd Harmonic Sinusoidal Voltage”, Eur. Conf. Optical Communication (ECOC), 1996, Paper ThC.1.4
[29] S. Zhang, “Traveling-wave Electroabsorption Modulators”, Ph.D. dissertation, University of California, Santa Barbara, July 1999
93
Chapter 4 Clock Recovery Techniques 4.1 Introduction
In digital communication, clock recovery is an essential functionality to
acquire synchronization with the input signal at the receiver or any networking
node so that digital decision or time domain signal processing can be
implemented. The most fundamental specification for clock recovery is the jitter
transfer bandwidth, which determines the timing jitter of the recovered clock
with a given input signal quality. In general, a practical clock recovery
subsystem should allow low frequency jitter to be transferred to the recovered
clock so as to track with the slow timing drift of the input signal. On the other
hand, high frequency jitter should be suppressed to ensure the spectral purity of
the recovered clock. This requirement is particularly important for re-timing of
CHAPTER 4: Clock Recovery Techniques
94
signal in a regenerator, where undesired jitter caused by the dispersion, non-
linearity, and noise of the transmission link can be rectified.
The clock tone at the data rate in the incoming signal contains the
desired timing information. Depending on the format of the signal, the strength
of the clock tone can vary dramatically. For return-to-zero (RZ) data, strong
clock tones exist at the data rate and its harmonic frequencies. However, in a
perfect non-return-to-zero (NRZ) data, there is no clock tone at the data rate.
Consequently, nonlinear circuits are required to enhance or retrieve the clock
tone in the signal [1] and the square law is a desirable non-linearity for timing
extraction. There are also several optical methods to enhance the clock tone in
an optical NRZ signal, which include electrooptic multiplication [2] and
differentiating the leading edge by saturating a semiconductor optical amplifier
(SOA) [3].
The simplest way to extract the timing information from the incoming
signal is to use an electrical band-pass filter centered at the clock frequency.
Despite its simplicity there are two major drawbacks. First, the pass-band
bandwidth is finite which means that significant jitter transfer may occur. This is
particularly true at higher bit rates. For example, a high quality band-pass filter
can have a Q-value of 1000, and at 10 GHz it corresponds to a pass-band
bandwidth of 10 MHz, which is too high for low jitter clock recovery. As a
result, extremely high Q filters are necessary. Secondly, limiting amplifiers may
CHAPTER 4: Clock Recovery Techniques
95
be necessary to ensure a stable output level, which would also increase the
timing jitter. In addition, the center frequency of the electrical filter in most
cases cannot be easily tuned to track the change or drift in the clock frequency,
reducing the flexibility of this approach.
The most widely adopted electronic clock recovery technique is the
phase-locked loop (PLL) [4]. A typical PLL is composed of a phase comparator,
a loop filter and a voltage controlled oscillator (VCO). The output of the VCO is
compared with the clock tone in the input signal by the phase comparator. The
resulting error signal is filtered by a loop filter and used to control the output
frequency of the VCO in order to complete the loop. The jitter transfer
bandwidth is mainly determined by the loop filter, which also sets the maximum
detuning between the free-running VCO and the clock tone of the input signal.
More sophisticated PLL circuits such as those with a quadrature phase detector
[5] can extend the operating frequency range beyond the bandwidth of the loop
filter while maintaining the same jitter transfer bandwidth for low jitter output.
Another electronic clock recovery technique is the injection locking of
an electrical oscillator [6], which has a simpler configuration compared to the
PLL type clock recovery. Without external injection, the oscillator oscillates at
its free-running frequency. However, in the presence of an external injection, as
long as the frequency and strength of the injection satisfies a certain condition,
the oscillator can be locked to the injection with a constant phase difference.
CHAPTER 4: Clock Recovery Techniques
96
The advantage of this approach compared to using only a band-pass filter is that
the spectral purity is better and the output clock amplitude is stable. When
compared to the PLL approach, the input signal interacts directly with the
oscillator instead of through a phase comparator and a feedback loop. However,
while this simplifies the configuration, flexibility is somewhat reduced.
In recently years, the injection locking technique has been extensively
studied for clock recovery in fiber-optic communication, which can be roughly
categorized as either optoelectronic or all-optical.
In the optoelectronic category, photonic components are incorporated in
the oscillator so that an optical input signal can directly lock the oscillator. In
Ref. [7], a phototransistor is used as an optical to electrical converter while
providing gain for a ring oscillator. In Ref. [8], an optoelectronic ring oscillator
consisting of a photodetector, RF amplifier, E-O modulator, and optical fiber
was demonstrated to lock with either an electrical or an optical input signal and
generated both electrical and optical clocks. The long fiber link in this
optoelectronic oscillator increases the Q-value of the oscillator and can result in
a very low-jitter clock.
On the other hand, all-optical injection locking clock recovery provides
an alternative solution. The all-optical oscillator is usually an active and self-
sustained optical pulse source such as a two-section gain-coupled DFB laser [9],
a self-pulsating DFB laser [10], and a monolithic mode-locked laser diode [11]-
CHAPTER 4: Clock Recovery Techniques
97
[12]. There are several advantages in these all-optical approaches. First, these
devices are usually monolithically integrated and very compact. Secondly, the
operating frequency can be much higher than those achievable with other
solutions where the speed is limited by electronics. For example, in Ref. [11]-
[12], the mode-locked laser diodes can be locked by 160-Gb/s RZ data streams
and generate 160-GHz optical clock pulses. As a result, these all-optical clock
recovery techniques are attractive for all-optical signal processing in high-speed
optical communication systems where synchronized optical clock pulses are
highly required. However, the main issues for all-optical clock recovery are the
locking stability, input and output wavelength dependency, polarization
dependency, and pattern length dependency.
With the advent of optical time-division multiplexing (OTDM), the line-
rate of the optical signal is significantly increased beyond the reach of current
electronics, as discussed in Chapter 1. The highest reported speed of the
conventional electronic PLL clock recovery is 100 Gb/s, where a commercially
available harmonic mixer was used [13]. The VCO in this demonstration is
operating at 10 GHz which means that a sub-rate clock is obtained. For line-rate
speeds over 100 Gb/s, conventional electronic PLL is limited by the speed of
electronics. The all-optical clock recovery approaches are ideal for high-speed
(such as 160-Gb/s) clock recovery at the line-rate, which is useful for line-rate
all-optical 3-R regeneration [14]. Nevertheless, it is difficult for these all-optical
CHAPTER 4: Clock Recovery Techniques
98
approaches to lock at sub-rates since sub-harmonic tones in the OTDM signal is
usually very low. Sub-harmonic clocks at the base-rates are particularly essential
for OTDM demultiplexing and add-drop multiplexing. Assistive techniques such
as a regeneration loop [15] or a two-stage configuration [16] are required for the
all-optical approaches to lock with sub-harmonics but at the expense of
complexity.
In terms of sub-harmonic clock recovery for OTDM systems, the most
widely accepted approach is the combination of an optical demultiplexer and an
electronic clock recovery circuit operating at the sub-rate. This approach is
called scaled OTDM clock recovery and the concept is depicted in Figure 4.1.
Figure 4.1 Concept of scaled OTDM clock recovery
In the scaled OTDM clock recovery, the optical demultiplexer brings the
high-speed line-rate signal down to a sub-rate at which the electronic clock
recovery circuit operates. However, the timing of the optical demultiplexer is
controlled by the recovered clock from the electronic clock recovery circuit. As
a result, the clock acquisition process is more complicated than that of the
CHAPTER 4: Clock Recovery Techniques
99
original sub-rate operation. A detailed model will be presented in Section 4.3 to
study the locking dynamics.
Several demonstrations using this scaled approach have been reported in
the literature for OTDM systems. There are two major kinds of electronic clock
recovery technique used: standard PLL and balanced PLL. The use of a
standard PLL clock recovery circuit is a straightforward implementation. Up to
160-Gb/s operation has been demonstrated [17], including field experiments
[18]. The advantage of this approach is that standard PLL is a mature technology
and commercially available.
The balanced PLL approach originates from the idea that the average
output power of the optical demuliplexer can be used as a gauge of timing [20]:
when the demultiplexer is perfectly aligned with one of the tributaries in the
OTDM signal, the output power is maximal and it decreases as the timing shift
increases. However, there is a polarity ambiguity that arises from a DC offset
proportional to the average input power level. Techniques like dithering [19] and
bi-directional configuration [20] have been proposed to solve this problem, but
at the expense of complexity. A better solution is to use a balanced
photodetector to remove the DC offset [21], where one input to the balanced
photodetector is the demultiplexed signal and the other is the line-rate signal
with a properly adjusted power level. The advantages of this solution are two-
fold: (1) only a low-speed balanced photodetector with a bandwidth close to that
CHAPTER 4: Clock Recovery Techniques
100
of the error signal is required (typically 1 MHz); (2) the input power fluctuations
or even the relative intensity noise in the OTDM signal can be rejected by the
balanced detection. A small-signal analysis was presented in Ref. [22] to study
the minimization of locking-time. In addition, there are several variations of this
approach reported in the literature using a single electroabsorption modulator
(EAM) and operate at a 160-Gb/s line-rate [23]-[24].
Another PLL-based approach utilizes the cross-correlation obtained by
four-wave mixing between the optical input signal and an optical pulse train in a
traveling-wave laser-diode amplifier and demonstrated up to 400 Gb/s line-rate
operation [25].
In this chapter, two novel clock recovery subsystems based on traveling-
wave electroabsorption modulators (TW-EAMs) are proposed, demonstrated,
and analyzed. The first clock recovery subsystem is for high-seed OTDM. It
contains a standard PLL but is capable of simultaneous clock recovery and
demultiplexing using only a single TW-EAM. In Section 4.2, this concept is first
demonstrated at 40 Gb/s and then extended to 160 Gb/s using the scaled
approach. Detailed modeling and simulation of the scaled OTDM clock
recovery with a standard PLL is presented in Section 4.3. In Section 4.4, a novel
injection locking clock recovery subsystem using the TW-EAM as part of a ring
oscillator is proposed and its properties characterized at 10 Gb/s. This novel
clock recovery approach will be applied in chapter 6 to realize a compact TW-
CHAPTER 4: Clock Recovery Techniques
101
EAM-based 3-R regenerator (PAW-regenerator). The characterization presented
in this chapter helps to understand the performance limitations of the compact 3-
R regenerator. Both clock recovery subsystems proposed in this chapter take
advantage of the traveling-wave electrodes and the photo-detection capability of
the TW-EAM.
4.2 Scaled OTDM clock recovery with simultaneous
demultiplexing
In this section, a multi-functional clock recovery subsystem with
simultaneous demultiplexing capability will be proposed and demonstrated. A
TW-EAM is used in the subsystem to serve three simultaneous functions: a
demultiplexer, a photodetector, and a pulse generator for optical demultiplexing,
electrical clock recovery, and optical clock generation at a line-rate of 40 Gb/s.
Incorporating the concept of scaled OTDM clock recovery, this subsystem is
scaled to operate at a line-rate of 160 Gb/s. The key to simultaneous operation is
the utilization of harmonic frequencies and independent wavelengths in the TW-
EAM. Part of these results was published in [26].
CHAPTER 4: Clock Recovery Techniques
102
4.2.1 Concept
Figure 4.2 shows the concept of the proposed approach. It consists of a
TW-EAM and an electronic clock recovery circuit which takes an electrical
input at the line-rate in order to recover a clock at the base-rate. The electrical
line-rate input is provided by the TW-EAM, which is acting as a photodetector
to this purpose. The 3-dB detection bandwidth of the TW-EAM does not need to
be as high as the line-rate speed. The only requirement is that the detected line-
rate clock tone in the photocurrent signal must be high enough for the electronic
clock recovery circuit to operate properly.
Figure 4.2 Concept of simultaneous demultiplexing and clock recovery using a single TW-EAM
The recovered electrical clock at the base-rate (usually sinusoidal) is then
fed back to the TW-EAM at another electrical port with a proper phase and
power. This drives the TW-EAM as an optical gate with a base-rate repetition.
CHAPTER 4: Clock Recovery Techniques
103
When the timing of the optical gate is aligned with one of the time-division
tributary in the optical line-rate signal, demultiplexing of that particular channel
is obtained. The applied electrical base-rate clock then goes through the TW-
EAM and terminates at the electronic clock recovery circuit. This configuration
is unique to TW-EAM and does not require a power divider which is needed if a
lumped EAM is used instead. The use of a power divider would impose a 6-dB
power loss. On the other hand, a continuous wave (CW) at another wavelength
(λ2) can be coupled into the TW-EAM and an optical clock pulse train
synchronized with the input signal can be obtained after filtering. As a result,
simultaneous electrical and optical clock recovery as well as OTDM
demultiplexing is realized with a single TW-EAM.
4.2.2 40-Gb/s Operation
The proposed approach is first demonstrated with a 40-Gb/s line-rate and
a 10-Gb/s base-rate. The experimental configuration is shown in Figure 4.3. The
40-Gb/s OTDM RZ signal at 1554.5 nm is multiplexed from 10 Gb/s with
passive delay lines and couplers. The 10 Gb/s RZ signal is generated by
modulating a 10 GHz, 5 ps optical pulse train from a mode-locked fiber ring
laser using a LiNbO3 modulator with 231-1 PRBS. The CW light for optical
clock generation is 6 dBm at 1560.5 nm. A 2.4 nm optical band-pass filter is
used to separate the demultiplexed 10 Gb/s signal at 1554.5 nm and the
CHAPTER 4: Clock Recovery Techniques
104
generated 10 GHz optical clock at 1560.5 nm. The 10 GHz electrical clock
signal is amplified to 6 Vpp in order to drive the TW-EAM for demultiplexing
and optical clock generation.
Figure 4.3 Simultaneous clock recovery and demultiplexing setup with a 40-Gb/s line-rate
Figure 4.4 Optical to electrical small-signal response of the TW-EAM (50-Ω terminated)
CHAPTER 4: Clock Recovery Techniques
105
The TW-EAM used in the experiment has a 3-dB optical to electrical (O-
E) response close to 12 GHz, as shown in Figure 4.4. Because of the limited
bandwidth, the detected 40 Gb/s eye is closed (Figure 4.5(c)).
Figure 4.5 Electrical eyes detected by the TW-EAM (50-Ω terminated) with RZ input at
(a) 10 Gb/s; (b) 20 Gb/s; (c) 40 Gb/s
However, the electronic clock recovery circuit only requires the 40-GHz
clock tone from the input signal to operate, as long as the strength of the tone is
larger than the minimal required power, which is – 67 dBm for this particular
setup. Figure 4.6(a) shows that this would correspond to a minimal optical input
power of – 12 dBm. Nevertheless, a stronger clock tone would lead to a larger
locking range as shown in Figure 4.6(b), which yields higher tolerance to system
fluctuations. It is clearly seen from Figure 4.6(c) that the clock tone at the line
rate is strong in a RZ signal. At 5 dBm of input power, which is the power level
for all the experiments, the locking range is 0.764 MHz (centered around 40
GHz).
CHAPTER 4: Clock Recovery Techniques
106
(a) (b)
(c)
Figure 4.6 (a) Power of the 40-GHz tone detected by the TW-EAM as function of optical input power; (b) Locking range of the PLL clock recovery circuit as function of optical input power.
The gray area indicates the lockable frequency; (c) RF spectrum of the photocurrent signal detected by the TW-EAM with a 5-dBm, 40-Gb/s OTDM RZ input signal
The root-mean-square (RMS) timing jitter of the clock can be obtained
by integrating the single side-band (SSB) phase noise spectrum [27]:
RMS timing jitter = o
f
ffdffL π2)(
2
1 ⋅
where L( f ) is the SSB phase noise defined as the ratio of noise power in 1 Hz
bandwidth at an offset f from the clock to the clock power; fo is the center
CHAPTER 4: Clock Recovery Techniques
107
frequency of the clock. The measured SSB noise spectrum of the transmitter
clock and the recovered optical / electrical clocks are shown in Figure 4.7.
Figure 4.7 SSB noise spectrum of the transmitter clock, the recovered electrical clock, and the generated optical clock for 40-Gb/s line-rate. Insert: traces of the recovered clocks
By choosing an integration range of 1 kHz (f1) to 10 MHz (f2), the RMS
timing jitter of the transmitter clock, the recovered electrical clock, and the
generated optical clock (with EDFA amplification) are 223 fs, 231 fs, and 232
fs, respectively. The pulsewidth of the generated optical clock is 14 ps, which is
also the width of the gating window for demultiplexing. A shorter pulsewidth
can be expected by increasing the amplitude of the clock beyond 6 Vpp to the
TW-EAM.
Figure 4.8 shows the bit error rate (BER) curves and the eye diagrams of
the demultiplexed 10-Gb/s channel. All four OTDM channels have similar
performance and only the results for one channel are shown. BER comparison is
CHAPTER 4: Clock Recovery Techniques
108
done by switching the 10 GHz electrical clock supplied to the TW-EAM and the
BER tester from the transmitter clock to the recovered electrical clock. The
power penalty is less than 0.2 dB for using the recovered clock, which is pretty
low and close to the measurement accuracy (~ 0.1 dB). The above results show
the successful operation of the proposed concept for simultaneous
demultiplexing, electrical clock recovery, and optical clock generation at a 40-
Gb/s line-rate.
Figure 4.8 BER curves and eye diagrams measured with the transmitter clock and
the recovered electrical clock with a 40-Gb/s line-rate
4.2.3 160-Gb/s Operation
The extension of the proposed simultaneous clock recovery and
demultiplexing subsystem to higher OTDM line-rates is of high interest,
especially at 160-Gb/s. The setup presented in the previous section utilizes the
40-GHz tone in the photocurrent signal to extract the timing information.
CHAPTER 4: Clock Recovery Techniques
109
However, in a well multiplexed OTDM signal (equal power and precise timing
for each individual channel), the base-rate tone (40-GHz in this case) can be
very low or completely gone, as shown in Figure 4.9. This is the very problem
all-optical clock recovery approaches face when attempting to recover a sub-
harmonic clock, as discussed in Section 4.1.
Figure 4.9 RF spectrum of the photocurrent signal detected by the TW-EAM with a 160-Gb/s OTDM RZ input signal. Note that there is no perceivable 40-GHz tone.
To solve this problem, the scaled OTDM clock recovery technique
introduced in Section 4.1 is employed. Figure 4.10 shows the experimental
setup, where a standing-wave enhanced TW-EAM (EAM 2) is used as a 160- to
40-Gb/s demultiplexer and driven by a 6 Vpp, 40-GHz sinusoidal wave
multiplexed from the 10-GHz recovered clock. The gating window width is 5 ps
for the TE polarization. The 160 Gb/s RZ data signal (5 dBm) is multiplexed
passively from 10 Gb/s. The pulses from the mode-locked fiber ring laser are
compressed nonlinearly with 5 km of dispersion-shifted-fiber (DSF) to 2 ps.
CHAPTER 4: Clock Recovery Techniques
110
However, the tails of the compressed pulses are relatively long so that
intersymbol interference (ISI) occurs. To minimize the impact of ISI, the 160
Gb/s signal is multiplexed with alternating polarization since the loss of the
EAM is higher for the TM polarization, which reduces the interference from
adjacent channels.
Figure 4.10 Setup of scaled simultaneous clock recovery and demultiplexing
with a 160-Gb/s line-rate
Figure 4.11 shows the RF spectrum of the clocks. The RMS timing jitters
are 216 fs, 224 fs, and 229 fs for the transmitter clock, the recovered electrical
clock and the generated optical clock. These results are very close to that in the
40-Gb/s line-rate case, meaning that the scaled approach does not degrade jitter
performance.
CHAPTER 4: Clock Recovery Techniques
111
Figure 4.11 SSB noise spectrum of the transmitter clock, the recovered electrical clock, and the generated optical clock for 160-Gb/s line-rate. Insert: traces of the recovered clocks
Figure 4.12 BER curves and eye diagrams measured with the transmitter clock and the recovered electrical clock with a 160-Gb/s line-rate
The BER curves and the eye diagrams are shown in Figure 4.12. The
power penalty for using the recovered clock is less than 0.5 dB and can be
attributed to the slight difference in power (~ 0.5 dB) between the transmitter
CHAPTER 4: Clock Recovery Techniques
112
clock and the recovered electrical clock that are fed into EAM 2. The fiber-to-
fiber loss of the EAM, which is critical to the signal-to-noise ratio, is very
sensitive to the RF driving power. Though not demonstrated in the experiment,
EAM 2 can also be used to generate a synchronized optical clock at 40 GHz, just
like EAM 1 does for a 10-GHz clock.
4.3 Modeling of scaled clock recovery
The successful operation of the scaled OTDM clock recovery has been
demonstrated in section 4.2, where an electronic clock recovery circuit operating
at 40-Gb/s is scaled to work at a line-rate of 160-Gb/s with an additional
demultiplexer. It is then of interest to build a model in order to study the
performance of this scaled technique. The questions to be answered by the
modeling would include how the locking dynamic changes after adding the
demultiplexer and how it scales with the line-rate speed.
4.3.1 Model
Even though the base-rate electronic clock recovery circuit can be of the
PLL type or the injection locking type, the standard PLL type is considered in
the model since it is the technique used in the experiments presented in Section
CHAPTER 4: Clock Recovery Techniques
113
4.2. Both full numerical simulation and small-signal analysis of the differential
equation developed from the model will be presented.
Figure 4.13 Model of the clock recovery schemes (a) standard PLL clock recovery at the
base-rate; (b) scaled OTDM clock recovery with a demultiplexer and a standard PLL.
Figure 4.13(a) describes the standard PLL working at the base-rate [4].
ωo is the angular frequency of the base-rate clock. φi and φo are the phases of the
clock tone in the input signal and the VCO output. N is the number of
multiplexed channels. A1 and A2 are the power gain of the erbium-doped fiber
amplifier (EDFA) and the RF amplifier. km and ko are the gain of the RF mixer
and the VCO. RPD is the responsivity of the photodector at the base-rate
frequency. F1 and F2 are the losses of the band-pass filter (centered at the base-
rate) and the low-pass filter. For simplicity, the low-pass filter is assumed to
have a flat response in the frequency range of interest. This means that a first-
CHAPTER 4: Clock Recovery Techniques
114
order PLL is considered. The actual response of the low-pass filter does
influence the locking dynamics as indicated in [4] and [22] but the overall
property does not change dramatically. pi (t) is the waveform of the optical input
signal, which is assumed to be a pulse train at the respective rates to simulate a
RZ signal. This is a reasonable simplification since the difference between the
encoded RZ signal and the pulse train is that the clock tone is reduced by a
factor and a broad-band data spectrum is added [1], which is be filtered out by
F1.
The scaled OTDM clock recovery is modeled as in Figure 4.13(b). The
only difference from the base-rate case in Figure 4.13(a) is that an optical
demultiplexer is added, whose timing is controlled by the clock output from the
VCO. M represents the loss of the demultiplexer while f (t) is the gating window
shape in time domain.
The first step to derive a governing equation for the scaled clock
recovery is to expand pi (t) and f (t) by the complex Fourier series:
∞
−∞=
⋅⋅⋅=m
tjmNtjmnii
oi eeDPtp ωφ )()(
(4.1)
∞
−∞=
⋅⋅=n
tjntjnn
oo eeCtf ωφ )()(
(4.2)
CHAPTER 4: Clock Recovery Techniques
115
Note that pi (t) is expanded with a fundamental frequency of Nωo, which
is the line-rate frequency. On the other hand, f(t) is expanded with a fundamental
frequency of ωo, the base-rate frequency. The output of the demultiplexer, ps (t),
is a product of the two functions and the loss of the demultiplexer:
)()()( tftpMtp is ⋅⋅=
(4.3)
The significance of this process is that the demultiplexer is in fact an
optical mixer which generates numerous sub-harmonics that were absent in the
original OTDM signal. This optical signal is then detected by the photodetector
and converted into an electrical signal. The band-pass filter F1 picks only the
clock tone at the base-rate and rejects all other frequencies. After amplification,
the amplitude of the resulting electrical signal can be express as:
⋅⋅⋅⋅⋅⋅⋅= 211)( AFZRAMPtv PDPDir
( ) ( )( )∞
−∞=
−++ ⋅⋅+⋅⋅m
tjsmjsm
tjqmjqm
ooiooi eeCDeeCD ωφφωφφ
(4.4) where q = 1 - mN and s = -1 – mN.
At the RF mixer, νr (t) mixes with the output from the VCO, νo (t), and
only the low-frequency component is picked by the low-pass filter F2. The
CHAPTER 4: Clock Recovery Techniques
116
resulting signal, ν2 (t), which is used to control the frequency of the VCO, can be
express as:
)cos()()(2 oorm ttvkofcomponentfrequencylowtv φω +⋅⋅=
(4.5) As mentioned earlier in this section, the detailed response of the low-
pass filter is not considered. The derivative of the VCO output phase (i.e.
frequency) can be described by the following relation:
)()( 2 tvkdt
dt oc
oo ⋅+=+=Ψ
•ωφω
(4.6) The first part of (4.6) means that the frequency reference is set at ωo,
which is the frequency of the clock tone in the input signal. The difference
between the instantaneous VCO output frequency and the reference frequency is
attributed to the derivative of φo. When stable phase locking is achieved, φo does
not change in time and the VCO output frequency equals that of the input signal.
The second part of (4.6) indicates that the output frequency of the VCO is varied
according to the signal ν2 (t) with a gain of ko and the free-running frequency is
ωc. By substituting ν2 (t) from (4.5) into (4.6), a differential equation describing
the scaled clock recovery can be obtained:
CHAPTER 4: Clock Recovery Techniques
117
[ ] [ ]∞
=−+ −⋅⋅+⋅⋅+−⋅⋅+=
1111 )(cos)()(
miomNmNmooc
o NmaadRadRdt
d φφωωφ
(4.7)
where
omPDPDi kkFAFZRAMPR ⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅= 221141
− ⋅⋅=Ny
Ny in dttnNtpyN
d )cos()( 0ω , − ⋅⋅=y
yn dttntfy
a )cos()(1
0ω , y = π / ωo
The variables dn and an are the n-th order Fourier coefficient of the input
signal and the demultiplexer gating window, counted from their respective
fundamental frequency. ZPD is the impedance of the photodetector. Similarly,
the differential equation governing the original base-rate operation of the
standard PLL is:
[ ] [ ]iooco Rd
dtd φφωωφ −⋅⋅⋅+−= cos2 1
(4.8) The difference between the base-rate case and the scaled OTDM case in
Figure 4.13 can be studied by comparing equations (4.7) and (4.8), which will
be presented in the following sections.
4.3.2 Lock-in Range
First, the lock-in range of the two cases is studied. The lock-in range is
defined as the frequency difference between the clock tone in the input signal
CHAPTER 4: Clock Recovery Techniques
118
and the free-running frequency, for which after closing the loop the phase
difference between φi and φo converges to a steady-state value. Since there is
always some uncertainty in the actual frequency of the input signal, a reasonable
lock-in range is desired to allow system tolerance. For the first-order PLL, the
lock-in range equals the coefficient of the cosine function in the governing
equations (4.7) and (4.8). This can be easily understood from (4.8) that it is a
necessary condition for dφo / dt = 0.
The change in lock-in range after applying the scaled OTDM clock
recovery technique is of particular interest. This would indicate how the system
tolerance changes by the scaling process. The ratio of the lock-in ranges
measured at the base-rate frequency with and without scaling can be express as:
1
11
2
)(
)()(
dR
aadR
RateBaseOTDM
LRR mNmNm
L
Lratebase ⋅⋅
+⋅⋅=
−∆∆= −+
− ωω
where LRRbase-rate stands for “lock-in range ratio measured at the base-
rate”. By assuming that the full-width-at-half-maxima (FWHM) of the input
pulse train (pi (t)) to be 50% of the line-rate bit slot and a base-rate of 10 Gb/s,
the dependence of LRRbase-rate on the FWHM of the gating window of the
demultiplexer is numerically calculated and plotted in Figure 4.14 for several
multiplexing numbers. The summation goes up to m = 5, which essentially
converges.
CHAPTER 4: Clock Recovery Techniques
119
Figure 4.14 Locking range ratio seen at the base-rate as function of demultiplexer gating window width for multiplication numbers of 4, 8, 16, and 64
As the number of multiplexed channels increases, the lock-in range
decreases. However, for each multiplexing number (N), there exists an optimal
demultiplexer (Demux) window width that maximizes LRRbase-rate. With a 50%
FWHM input pulse train, the optimal Demux window FWHM is about 37% of
the line-rate bit slot for all N.
An interesting result would be obtained if the lock-in range of the scaled
case is measured at the line-rate. Define LRRline-rate as “lock-in range ratio
measured at the line-rate”, which has an expression of:
ratebaseL
Lrateline LRRN
RateBaseOTDMN
LRR −− ⋅=−∆
∆⋅=)()(
ωω
The dependence of LRRline-rate on Demux window FWHM is shown in
Figure 4.15. From this figure, the maximal lock-in range obtainable at the line-
CHAPTER 4: Clock Recovery Techniques
120
rate is about 50% of the un-scaled case and is independent of the total number of
channels multiplexed. This indicates that as long as the Demux window width is
properly chosen, the maximal lock-in range remains the same for all N when
measured at the line-rate but scales down with N when measured at the base-
rate. In other words, the tolerance on input frequency variation at the base-rate
decreases when the line-rate speed increases. To scale to a very high line-rate,
the stability of the base-rate VCO is critical since the lock-in range is
considerably reduced and very little drift in oscillation frequency can be
allowed.
Figure 4.15 Locking range ratio seen at the line-rate as function of demultiplexer gating window width for multiplication numbers of 4, 8, 16, and 64
4.3.3 Shift of free-running frequency
Comparing the first terms on the right-hand side of (4.7) and (4.8), there
is an extra term, ωshift = R*do*a1, for the scaled case in (4.7). This term acts like
CHAPTER 4: Clock Recovery Techniques
121
a shift to the free-running frequency ωo by applying a bias voltage of ωshift / ko to
the VCO. The physical origin of this shift is the modulation of the DC
component in the input pulse train (do) by the fundamental frequency of the
Demux gating window (a1). This modulation at the base-rate is converted into a
DC signal in ν2 (t) by the RF mixer, which can be regarded as a self-biasing
effect.
(a) (b)
Figure 4.16 (a) Normalized ωshift as function of Demux window FWHM; (b) Normalized ωshift at optimal Demux window width
The ratio of ωshift to the lock-in range is plotted in Figure 4.16(a) as
function of the Demux window width. Since the lock-in range is an indicator of
system tolerance, normalization of ωshift to the lock-in range is informative.
Figure 4.16(a) shows that when the multiplexing number increases, the shift of
free-running frequency is more sensitive to the Demux window FWHM. This is
another factor demanding that when the line-rate speed increases, the stability
and accuracy of the VCO should also increase.
CHAPTER 4: Clock Recovery Techniques
122
When the Demux window width is set at the optimal value to maximize
the lock-in range, the corresponding ωshift is about twice the lock-in range, as
shown in Figure 4.16(b). The relationship is depicted in Figure 4.17 at this
condition. It is evident from this figure that because of ωshift, the free-running
frequency of the VCO, ωc, must be detuned from the clock frequency of the
input signal, ωo, by 1 to 3 lock-in range.
Figure 4.17 Relationship of the free-running frequency, the shift of free-running frequency, and the lock-in range at the optimal Demux window width condition
4.3.4 Phase locking transient
Next, the time evolution of the locking process is studied and the
property of interest is the time it takes to acquire phase locking, i.e. the lock-in
time. To simulate the time evolution, parameter values listed in Table 4.1 are
used, with which the plain gain of the scaled PLL is calculated:
sradkkFAFZRAMPR omPDPDi
722114
1 1016.2 ×=⋅⋅⋅⋅⋅⋅⋅⋅⋅⋅=
ωωωωshift = R*do*a1
2 * locking range
ωωωωc Frequency
CHAPTER 4: Clock Recovery Techniques
123
Referring to equation (4.7), R represents the combined effect of varies
components in the scaled clock recovery subsystem that influences the time
evolution of the VCO phase, φo. However, the effects of the width of the input
pulse and the width of the gating window are not represented by R but by dn and
an instead.
Parameter Symbol Value
Optical peak power of the input pulse train Pi 10 mW
Loss of the demultiplexer M -14 dB
Power gain of EDFA A1 20 dB
Responsivity of the photodetector at base-rate RPD 0.5 A/W
Impedance of photodetector ZPD 50 Ω
Loss of the band-pass filter F1 3 dB
Loss of the low-pass filter F2 3 dB
Power gain of RF amplifier A2 26 dB
Conversion gain of RF mixer km -6 dB
Gain of VCO ko 3.45e+10 rad/s/V
Table 4.1 Parameters of the scaled OTDM clock recovery subsystem
Without loss of generality, the evolution of phase-locking in the N=4
case is studied. It is assumed that the FWHM of the input pulse is 50% of the
line-rate bit slot and the Demux window width is optimal (37%). In this case, the
lock-in range is about 364 kHz, which is calculated by considering only the first
CHAPTER 4: Clock Recovery Techniques
124
set of harmonic terms in the summation, i.e. R*d1*(aN+1+aN-1). Higher order
terms are much smaller and can be ignored. In this case, the shift of free-running
frequency is 717 kHz.
Figure 4.18 Time evolution of the phase and the angular frequency of the VCO for input frequency (a) inside the lock-in range; (b) outside the lock-in range.
The time evolution of VCO output phase φo and its time derivative are
numerically simulated by solving the differential equation (4.7) and plotted in
Figure 4.18. The initial phase can have all kinds of possible value between +π
and –π. Therefore, 200 evenly distributed initial conditions are simulated and all
the evolutions are plotted. Figure 4.18(a) shows a lockable condition, where the
CHAPTER 4: Clock Recovery Techniques
125
detuning frequency, defined as [(ωc+ωshift) - ωo] / 2π, is zero. Since N=4 there
are 4 distinctive stable solutions and the phase evolves from the initial condition
to the closest stable solution. Physically, this means that the base-rate clock can
be locked to the closest OTDM tributary, depending on the initial phase of
VCO. Figure 4.18(b) shows the case when the input frequency is outside the
locking range and the detuning frequency is – 400 kHz. The phase doe not
evolve into a stable steady value and the derivative never goes to zero.
Figure 4.19 Lock-in time as function of initial phase
Since the initial phase can take any possible value, the lock-in time can
also vary. Figure 4.19 shows the dependence of the lock-in time with initial
condition (360 initial phases are simulated). Depending on the difference
between the initial value and the stable solution, the lock-in time can vary in a
wide range. Therefore, an average value is taken as a representative lock-in
time. It is generally true that the time it takes to acquire synchronization using
CHAPTER 4: Clock Recovery Techniques
126
PLL and the injection locking techniques has some uncertainty since the initial
phase of the oscillator is unknown.
Figure 4.20 Average lock-in time for N=4 and N=16 as function of detuning in (a) absolute detuning scale; (b) normalized detuning scale
Next, the effects of detuning on the average lock-in time are studied.
Optimal Demux window width is assumed. Figure 4.20 shows the simulation
results for N=16 and N=4. From Figure 4.20(a), it is observed that the average
lock-in time is more sensitive to detuning at higher N. This makes sense because
the lock-in range is reduced when N increases. However, if the detuning is
normalized to the lock-in range, the same locking behavior is obtained for N=4
and N=16, as indicated by Figure 4.20(b), which is true for all N under the same
condition. The minimal average lock-in time, occurs when there is no detuning,
is the same for all N (with optimal Demux windows). This is an interesting
result since the lock-in range, which also determines the strength to change the
CHAPTER 4: Clock Recovery Techniques
127
phase as easily understood in equation (4.7), is reduced as N increases and one
would expect a slower transient. The reason for this interesting behavior can be
found in Figure 4.21: even though the phase changes slower when N increases,
there are more sable solutions available for the phase to evolve in. As a result,
the average lock-in time is invariant of N when there is no detuning.
Figure 4.21 Phase evolution without detuning for (a) N=4; (b) N=16
Figure 4.22 Average lock-in time as function of input power without detuning
CHAPTER 4: Clock Recovery Techniques
128
With no detuning, the variation of lock-in time with respect to the
change in input peak power (Pi) is plotted in Figure 4.22. As expected, the lock-
in time decreases with higher input power but is independent of the number of
channels multiplexed (N).
Figure 4.23 Phase evolution traces for N=4 with (a) no detuning; (b) -200 kHz detuning; (c) -360 kHz detuning
Detuning not only increases the average lock-in time but also has
adverse effects on the stability of clock recovery. Figure 4.23 shows the phase
evolution traces for three detuning levels. The unstable solutions marked in the
figures satisfy the condition dφo / dt = 0 just like the stable solutions do, but
when there is any perturbation, the phase will evolve to the nearest stable
solutions. On the other hand, perturbation on the stable solution will dissipate
and the phase is stable. As the detuning increases, the unstable solutions will
move towards the stable solutions, as shown in Figure 4.23. The problem with
this is that when the perturbation in phase is large enough, the phase can jump
CHAPTER 4: Clock Recovery Techniques
129
from a stable solution to or beyond an unstable solution. As a consequence, the
phase will evolve to the next stable solution and causes hopping in the output
phase. Therefore, it is advantageous to reduce the detuning or increase the lock-
in range for a better stability and lock-in time.
4.3.5 Small-signal analysis
Conventional analysis of PLL is usually carried out in the Laplace or the
frequency domain after linearizing the differential equation that describes the
loop [4]. This is particularly useful for analytical analysis and for studying the
small-signal response around a steady-state solution. The inclusion of a detailed
filter response is also much easier in this approach when compared to the
treatment in time domain.
The small-signal analysis for the scaled OTDM clock recovery starts
from simplifying equation (4.7). Keeping only the first term in the summation,
which dominates higher order terms, the equation can be re-written as:
)cos( ioLRo NR
dtd φφωφ −⋅+∆=
(4.9)
where ooc adR ωωω −⋅⋅+=∆ )( 1 and )( 111 −+ +⋅⋅= NNLR aadRR .
CHAPTER 4: Clock Recovery Techniques
130
Consider a set of perturbed solutions:
iii δφφ +Φ= and ooo δφφ +Φ=
where Φi and Φο are the steady-state solutions that yield dφo / dt = 0.
Substituting the perturbed solutions into (4.9) and assuming that the
perturbations are small-signal, a new differential equation for the perturbations
can be obtained:
( )ioLR
LRo N
RR
dtd δφδφωδφ −⋅⋅
∆−⋅±=2
1
(4.10)
Respectively, the plus / minus sign corresponds to the unstable / stable
solutions. A general solution for δφo with the minus sign is:
iLR
LRo Nt
RRNC δφωδφ ⋅+
⋅
∆−⋅⋅−⋅= 11exp
2
(4.11) where C is an integration constant.
From (4.11), it is shown that the phase perturbation at the base-rate will
equal to 1/N of the phase perturbation at the line-rate. Other time independent
perturbations such as a sudden phase jump will decay exponentially and the
VCO output phase will return to the stable solution.
CHAPTER 4: Clock Recovery Techniques
131
The small-signal solution actually provides several insights that agree
with the large-signal simulations presented in earlier sections. The dependence
of N×RLR in the exponent of (4.11) on the Demux window FWHM is already the
plot in Figure 4.15. At the optimal Demux window width, the maximal value of
N×RLR is achieved, which is independent of N. Therefore, when the detuning,
∆ω, equals to zero, the decay time of a sudden phase perturbation is also
independent of N, in analogy to the lock-in time in the large-signal simulation
presented in Section 4.3.3. In addition, (4.11) also indicates that when the
Demux window is optimized (the same N×RLR for all N), the decay time (lock-
in time) is determined by the detuning, ∆ω.
The jitter transfer function, H(ω), can be obtained by re-arranging (4.10):
ωω
ω
ωδφωδφω
⋅−
∆−⋅⋅
∆−⋅==
jR
RN
RR
H
LRLR
LRLR
i
o
2
2
1
1
)()(
)(
(4.12) and
2
3
2
2
1
11)(
+
⋅=
−dB
NH
ωω
ω
(4.13)
where τ
ωω 11
2
3 =
∆−⋅⋅=−LR
LRdB RRN
CHAPTER 4: Clock Recovery Techniques
132
ω3-dB is the 3-dB jitter transfer bandwidth of the scaled OTDM clock
recovery, which is related to the lock-in range RLR and the detuning ∆ω. The
inverse of ω3-dB is the decay time of a sudden phase perturbation around a stable
solution.
4.4 Clock recovery by injection locking a ring oscillator
In this section, the injection locking technique is used for clock recovery,
as an alternative to PLL. In general, fewer components are required in the
injection locking approach, which is an advantage over PLL. One important
feature of this approach is that a TW-EAM can be incorporated into the ring
oscillator, resulting in a very compact and versatile clock recovery subsystem. In
Chapter 6, this TW-EAM-based ring oscillator will be combined with the
photocurrent-assisted wavelength converter (PAW-converter, to be introduced in
Chapter 5) and the result is a very compact 3-R PAW-regenerator using only a
single TW-EAM. Characterization of 10-Gb/s clock recovery using a ring
oscillator is presented in this chapter, which is essential to the analysis of the
compact 3-R PAW-regenerator. Operation and characterization of clock
recovery at 40-Gb/s using the TW-EAM-based ring oscillator were presented in
[28]-[30].
CHAPTER 4: Clock Recovery Techniques
133
4.4.1 Concept
The oscillator for clock recovery can be constructed in many different
ways. It can be purely electronic with lumped-element components [6],
optoelectronic with both electronic and electro-optic components [8], or purely
optical without electronics components [9]. However, no matter how it is
constructed, an important issue is to ensure the stability of the oscillation
frequency subject to environmental changes such as temperature drift.
Figure 4.24 Various constructions of a ring oscillator for clock recovery (a) purely electrical;
(b) optoelectronic with a photodetector; (c) optoelectronic with a TW-EAM in the ring
CHAPTER 4: Clock Recovery Techniques
134
The ring oscillator architecture is chosen due to the possibility of
incorporating a TW-EAM directly into the loop. Figure 4.24(a) shows a purely
electronic version of the ring oscillator, which consists of a closed loop, a RF
amplifier, and a RF band-pass filter. The basic requirements for a sustained
oscillation are pretty much the same as those required by a laser. First, the
wavelength (and hence frequency) of the oscillating electrical signal in the ring
must satisfy the condition that the signal reproduces itself after each round trip.
This means that for any sustainable oscillation mode, the round trip phase
change must equal to multiples of 2π. The spacing in frequency between these
oscillation modes is the free-spectral range (FSR). Any change in the effective
length or phase of the ring will result in a shift of the mode frequencies.
Secondly, to compensate for the loss encountered along the ring, gain must be
provided by one or more RF amplifiers. The small-signal gain of the amplifier
must be larger than the loss of the loop so that oscillation can build up from
noises in the ring such as thermal noise and shot noise. However, when a steady-
state oscillation is reached, the total loop gain must be equal to unity so as to
satisfy the oscillation condition. This means that the gain of the amplifier must
saturate (and be clamped) at the oscillating power level. Depending on the gain
bandwidth of the amplifier, there can be several modes oscillating at the same
time. The band-pass filter in the loop is to pick only one oscillating mode around
CHAPTER 4: Clock Recovery Techniques
135
the frequency of interest. As a result, the bandwidth of the band-pass filter
should be smaller than the FSR of the ring oscillator.
When an electrical signal is injected into the oscillator, it will add up
with the oscillating signals in the ring and changes the oscillation according to
the strength and frequency of the injection. A detailed modeling is presented in
[6] and is widely applied in the literature to study injection locking of
oscillators. Following [31], the governing equation of an oscillator can be
written as:
)(2
)sin(2
tBQA
AQdt
dn
oio
o
ioo
o ωφφωωφ −−⋅⋅−=
(4.14)
where φo, ωo, and Ao are the phase, the angular frequency, and the amplitude of
the oscillating mode; φi and Ai are the phase and the amplitude of injected signal;
Q is the Q factor of the oscillator; Bn(t) is a time-varying noise susceptance for
modeling noise. Note the high resemblance of (4.14), (4.7) and (4.8). In PLL the
locking strength is determined by the gain of the loop while in injection locking
the locking strength is determined by the injection power normalized to the
oscillation power.
To recover the clock from an optical signal, a photodetector is required
to convert the optical signal into an electrical signal, which can then be used to
injection-lock the ring oscillator, as shown in Figure 4.24(b). However, a TW-
CHAPTER 4: Clock Recovery Techniques
136
EAM can be used instead to this purpose and acts as part of the ring oscillator
(Figure 4.24(c)). The advantage of using the TW-EAM to replace a
photodetector is that it can also function as an optical intensity modulator at the
same time. The oscillating signal in the ring goes through the TW-EAM and
modulates it to be a pulse source or an optical gate. If a CW at another
wavelength is couple into the TW-EAM, an optical clock can be obtained after
filtering out the injection wavelength. This adds an extra functionality to the
setup. It has been demonstrated in [28] that by a combining a wavelength
converter afterwards, the data in the input signal can be imprinted to the
recovered optical clock and optical 3-R regeneration can be achieved in a
simplified configuration. Again, this approach shows that the TW-EAM can act
as a photodetector and a modulator simultaneously just like in Section 4.2. A
lumped EAM can also be used to replace the photodetector in Figure 4.24(b) and
be capable of both detection and modulation. However, due to the use of a
power divider to link the lumped EAM with the ring, at least a 6-dB reduction in
modulation power will occur. This would also increase the round trip loss of the
ring oscillator by 6 dB.
4.4.2 Characteristics measured at 10-Gb/s
The clock recovery of an injection locked ring oscillator with a
photodetector (Figure 4.24(b)) is characterized at 10-Gb/s in this section. This
CHAPTER 4: Clock Recovery Techniques
137
oscillator is exactly the ring oscillator that will be used in Section 6.4 for an
integrated optoelectronic 3-R regenerator. Even though this is not a TW-EAM-
based ring oscillator as used in Section 6.2, the properties and behaviors are
essentially the same and the conclusions drawn in this section do apply to the
TW-EAM-based oscillator.
The photodetector used in the experiment has a responsivity of 0.7 A/W
and the band-pass filter has a flat transmission bandwidth of 100 MHz centered
at 9.953 GHz. The physical length of the ring is about 0.75 m, which have a FSR
of 533 MHz. The small signal net gain of the ring is estimated to be 6 dB. The
input optical signal is a 10-Gb/s RZ signal with a 40-ps pulse width and a PRBS
pattern length of 231-1.
First, the lock-in range of the ring oscillator is characterized. From
(4.14), the lock-in range can be derived as in (4.15). Note that the A’s are the
electrical amplitudes of the injection signal and the oscillation signal.
QAA o
o
iLR 2
ωω =∆
(4.15)
The optical signal-to-noise ratio (OSNR) is varied in the experiment by
changing the input signal to an EDFA. The OSNR is measured with an optical
spectrum analyzer with a 0.1-nm resolution. The tolerance to input OSNR is of
high interest for 3-R regeneration applications, where the OSNR of the input
CHAPTER 4: Clock Recovery Techniques
138
signal is generally degraded. It is then critical that the recovered clock is not
sensitive to OSNR degradation.
Figure 4.25 (a) Lock-in range as function of optical input power for different input OSNR leverls. (b) 10-GHz clock tone as function of OSNR
It is shown in Figure 4.25(a) that the lock-in range have low dependence
on the input OSNR. If the optical input power is plotted in linear scale, a linear
dependence is observed as described by (4.15). Note that amplitude of the
detected photocurrent is in proportion to the optical input power. At 4 dBm of
optical injection, the variation in lock-in range between 16-dB and 43-dB OSNR
is 40 kHz or 10%. This variation arises from the fact that at the same optical
input power, the clock tone in the low OSNR signal is reduced (Figure 4.25(b))
so that the lock-in range is smaller.
The timing jitter of the recovered clock can be measured by integrating
the SSB noise spectrum as described in Section 4.2.2. From Figure 4.26, it is
CHAPTER 4: Clock Recovery Techniques
139
clear that the recovered clocks have lower jitters compared to the input signal,
showing the re-timing capability of this clock recovery approach. When the
input OSNR degrades, the jitter of the input signal does increase but not rapidly.
The dependence of the recovered clock on input OSNR is very low, which
means that this injection locking clock recovery approach is suitable for 3-R
regeneration.
Figure 4.26 Jitter of the input optical signal and the recovered electrical clock with varied input OSNR
The reason why the jitter of the recovered clock can be lower than that of
the input signal is that there is a finite jitter transfer bandwidth. From the small-
signal analysis of (4.14), the power spectral density of the phase fluctuation in
the recovered clock can be written as [28], [31]-[32]:
( ) )()()(/)()(222
0fSfHfSfffHfS
ijLRj δφδφδφ ⋅+⋅∆⋅=
(4.16)
CHAPTER 4: Clock Recovery Techniques
140
where the jitter transfer function
LR
j
ff
ifH
∆+
=1
1)( and ∆fLR = ∆ωLR / 2π .
In (4.16), )(
0fSδφ and )( fS
iδφ are the power spectral density functions of
the phase fluctuation in the free-running clock and the injected signal,
respectively. It can be understood from this equation that the noise of the
recovered clock is dominated by: (1) the free-running clock when the frequency
is higher than the lock-in range; (2) the injected signal when the frequency is
lower than the lock-in range. These properties are well demonstrated by the
measured results in Figure 4.27(a).
As mentioned in Section 4.1, these properties are desirable because the
recovered clock should be able to track the low-frequency drift in the injected
signal while the high-frequency noise being suppressed. Figure 4.27(b) shows
the measured and the theoretical jitter transfer function defined in (4.16). The
lock-in range is 87 kHz with a – 2dBm injection power. Perfect match is
obtained, showing the effectiveness of this model. The flat and broadband nose
in the SSB spectrum of the injected optical signal should come from the relative
intensity noise (RIN) of the laser source.
CHAPTER 4: Clock Recovery Techniques
141
(a)
(b)
Figure 4.27 (a) SSB spectrum of the optical injection, free-running clock, and the recovered clock; (b) the measured and calculated jitter transfer functions
Since the locking range also determines the jitter transfer bandwidth, a
proper value should be chosen in order to balance between frequency tolerance
and jitter. Figure 4.28 shows that when the injection power increases, the jitter
transfer bandwidth also increases. More noise can thus be transferred from the
injection signal to the recovered clock. As a result, the RMS jitter of the
recovered clock increases with injection power.
CHAPTER 4: Clock Recovery Techniques
142
Figure 4.28 (a) SSB spectrum of the recovered clock with different injection power;
(b) RMS jitter of the recovered clock as function of injection power
The above characterizations are made with no detuning. When detuning
exists, the quality of recovered clock may degrade in the injection locking
approach. Figure 4.29 shows the spectrum of the oscillating signal with different
levels of detuning. When the detuning is larger than the lock-in range (343 kHz
at 4 dBm of injection), there are tones at the oscillation and the injection
frequencies as well as weaker tones spaced by the detuning frequency (Figure
4.29(a)). When the detuning is closer to the lock-in range, the number of equally
spaced tones increases and becomes semi-continuous (Figure 4.29(b)). Within
the lock-in range, only one tone would exist. However, the side bands are more
pronounced when there is detuning (Figure 4.29(c)), compared to the no
detuning case (Figure 4.29(d)).
CHAPTER 4: Clock Recovery Techniques
143
Figure 4.29 Spectrum of the oscillator output with 4-dBm optical injection at different detuning levels. (a), (b) over lock-in range; (c), (d) within lock-in range
The RMS timing jitter is measured as function of detuning in Figure
4.30, which shows that jitter does increase with increased detuning. Fortunately,
the jitter increases slowly until it is close to the lock-in range. This is why the
lock-in range should be sufficiently large to allow tolerance to system
fluctuations, within the limit of acceptable jitter transfer bandwidth.
CHAPTER 4: Clock Recovery Techniques
144
Figure 4.30 RMS timing jitter as function of detuning. The lock-in range is 343 kHz.
An interesting side effect of detuning is that the phase difference
between the optical injection and the recovered clock varies according to the
amount of detuning, ∆fdetuning. From equation (4.14) it can be derived that the
phase difference, ∆θ, satisfies the following condition:
i
o
o
uning
AA
Qf
fdet)sin(∆
−=∆θ
(4.17)
The phase shift (timing shift) can be observed and measured from the
oscilloscope traces in Figure 4.31. In Figure 4.32 the timing shift is plotted as
function of detuning for two injection levels. Perfect match with the theoretical
curves given by (4.17) is obtained. This detuning induced timing shift will play
an important roll in explaining one of the performance limitations of the
compact 3-R regenerator proposed in Chapter 6.
CHAPTER 4: Clock Recovery Techniques
145
Figure 4.31 Scope traces of the recovered clock under different detuning levels. The lock-in range is 343 kHz.
Figure 4.32 Measured and calculated timing (phase) shift of the recovered clock
CHAPTER 4: Clock Recovery Techniques
146
4.5 Summary
In summary, two clock recovery techniques, PLL-based and injection
locking-based, utilizing the unique properties of the TW-EAM are studied in this
chapter. In both cases, the TW-EAM simultaneously works as a modulator and a
photodetector, providing multiple functionalities with the same device. The
traveling-wave electrode was utilized to simplify the configurations of the clock
recovery subsystems without using power dividers that could lead to extra
power loss.
In the PLL-based clock recovery, three simultaneous functions (OTDM
demultiplexing, electrical clock recovery, and optical clock generation) were
realized by utilizing microwave harmonic frequencies and independent optical
wavelengths in a TW-EAM. It is demonstrated that the recovered electrical and
optical clocks have timing jitters as low as that of the transmitter clock. This
concept was demonstrated at 40Gb/s and extended to 160 Gb/s by adding an
EAM demultiplexer to scale the operating speed.
The scaled OTDM clock recovery technique implemented with a
standard PLL was modeled in detain to study the change in performance with
the increase of line-rate speed. Both large-signal simulation and small-signal
analysis were presented. It was found that choosing an optimal demultiplexing
window width is critical to obtain the maximal lock-in range and the minimal
lock-in time.
CHAPTER 4: Clock Recovery Techniques
147
Injection-locking clock recovery using a TW-EAM-based ring oscillator
was characterized in detain at 10-Gb/s. Many of the results will be essential to
the understanding of performance limitations of the 3-R regenerators to be
introduced in Chapter 6.
References
[1] P. R. Trischitta and E. L. Varma, Jitter in Digital Transmission Systems, Chapter 2, Artech House, Inc, Norwood, MA, 1989
[2] E. Shumakher, A. Hayat, M. Nazarathy, and G. Eisenstein, “Timing Extraction of a 10-Gb/s NRZ Signal Using an Electrooptic Multiplication Scheme”, IEEE Photon. Technol. Lett., vol. 16, no. 10, pp. 2353-2355, 2004
[3] A. Bilenca, D. Dahan, J. Lasri, G. Eisenstein, and D. Ritter, “High Bit Rate Clock Recovery of NRZ Data by All-Optical Processing in a Semiconductor Optical Amplifier and Direct Optical Injection Locking of a Self-Oscillatinf Phtotransister”, IEEE Photon. Technol. Lett., vol. 14, no. 3, pp. 399-401, 2002
[4] V. F. Kroupa, Phase Lock Loops and Frequency Synthesis, John Wiley & Sons Ltd, West Sussex, England, 2003
[5] S. Vehovc, M. Vidmar, and A. Paoletti, “80 Gbit/s optical clock recovery with automatic lock acquisition using electrical phase-locked loop”, Electron. Lett., vol. 39, no. 8, pp. 673-674, 2003
[6] K. Kurokawa, “Injection Locking of Microwave Solid-State oscillators”, Proceedings of the IEEE, vol. 61, no. 10, pp. 1386-1410, 1973
[7] J. Lasri, D. Dahan, A. Bilenca, G. Eisenstein, and D. Ritter, “Clock Recovery at Multiple Bit Rates Using Direct Optical Injection Locking of a Self-Oscillatinf InGaAs-InP Heterojunction Bipolar Phototransistor”, IEEE Photon. Technol. Lett., vol. 13, no. 12, pp. 1355-1357, 2001
[8] X. S. Yao and G. Lutes, “A High-Speed Photonic Clock and Carrier Recovery Device”, IEEE Photon. Technol. Lett., vol. 8, no. 5, pp. 688-690, 1996
[9] Y. Li, C. Kim, G. Li, Y. Kaneko, R. L. Jungerman, and O. Buccafusca, “Wavelength and Polarization Insensitive All-Optical Clock Recivery From 96-Gb/s Data by Using a Two-Section Gain-Coupled DFB Laser”, IEEE Photon. Technol. Lett., vol. 15, no. 4, pp. 590-592, 2003
[10] C. Bornholdt, B. Satorius, S. Schelhase, M. Mohrle, and S. Bauer, “Self-pulsating DFB laser for all-optical clock recovery at 40Gbit/s”, Electron. Lett., vol. 36, no. 4, pp. 327-328, 2000
CHAPTER 4: Clock Recovery Techniques
148
[11] T. Ohno, K. Sato, R. Iga, Y. Kondo, T. Ito, T. Furuta, K. Yoshino, and H. Ito, “Recovery of 160 GHz optical clock from 160 Gbit/s data stream using modelocked laser diode”, Electron. Lett., vol. 40, no. 4, 2004
[12] S. Arahira, S. Sasaki, K. Tachibana, and Y. Ogawa, “All-Optical 160-Gb/s Clock Extraction With a Mode-Locked Laser Diode Module”, IEEE Photon. Technol. Lett., vol. 16, no. 6, pp. 1558-1560, 2004
[13] I. D. Phillips, A. D. Ellis, T. Widdowson, D. Nesset, A. E. Kelly, and D. Trommer, “100Gbit/s optical clock recovery using electrical phaselocked loop consisting of commercially available components”, Electron. Lett., vol. 36, no. 7, pp. 650-652, 2000
[14] S. Arahira and Y. Ogawa, “160-Gb/s OTDM Signal Source With 3R Function Utilizing Ultrafast Mode-Locked Laser Diodes and Modified NOLM”, IEEE Photon. Technol. Lett., vol. 17, no. 5, pp. 992-994, 2005
[15] T. Ohno, K. Sato, T. Shimizu, T. Furuta, and H. Ito, “Recovery of 40 GHz optical clock from 160 Gbit/s data using regeneratively modelocked semiconductor lasaer”, Electron. Lett., vol. 39, no. 5, pp. 453-455, 2003
[16] H. Yokoyama, Y. Hashimoto, H. Kurita, and I. Ogura, “Two-stage all-optical subharmonic clock recovery using modelocked semiconductor lasers”, Electron. Lett., vol. 36, no. 18, pp. 1577-1578, 2000
[17] D. T. K. Tong, K.-L. Deng, B. Mikkelsen, G. Raybon, K. F. Dreyer and J. E. Johnson, “160Gbit/s clock recovery using electroabsorption modulator-based phase-locked loop”, Electron. Lett., vol. 36, no. 23, pp. 1951-1952, 2000
[18] J. P. Turkiewicz, E. Tangdiongga, G .D .Khoe, and H. de Waardt, “Clock Recovery and Demultiplexing Performance of 160-Gb/s OTDM Field Experiments”, IEEE Photon. Technol. Lett., vol. 17, no. 5, pp. 992-994, 2005
[19] H. Bülow, “Optoelectronic synchronization scheme for ultrahigh-speed optical demultiplexing”, Electron. Lett., vol. 31, no. 22, pp. 1937-1938, 1995
[20] I. D. Philips, A. Gloag, D. G. Moodie, N.J. Doran, I. Bennion, A. D. Ellis, “Simultaneous demultiplexing and clock recovery using a single electrical modulator in a novel bi-direction”, Optics Communications, vol. 150, pp. 101-105, 1998
[21] D. T. K. Tong, B. Mikkelsen, G. Raybon, T. N. Nielsen, K. F. Dreyer, and J. E. Johnson, “Optoelectronic Phase-Locked Loop with Balanced Photodetection for Clock Recovery in High-Speed Optical Time-Division-Multiplexed Systems”, IEEE Photon. Technol. Lett., vol. 12, no. 8, pp. 1064-1066, 2000
[22] D. Zibar, L. K. Oxenløve, A. T. Clausen, and J. Mørk, “Theoretical and Experimental Investigation of a Balanced Phase-Locked Loop Based Clock Recovery at a Bit Rate of 160 Gb/s”, Annual Meeting of the IEEE Lasers and Electro-Optics Society, vol. 1, pp. 27-28, 2003
[23] E. S. Awad, P. S. Cho, N. Moulton, and J. Goldhar, “Subharmonic Optical Clock Reocvery From 160 Gb/s Using Time-Dependent Loss Saturation Inside a Single Electroabsorption Modulator”, IEEE Photon. Technol. Lett., vol. 15, no. 12, pp. 1764-1766, 2003
[24] C. Boerner, C. Schubert, C.Schimidt, E. Hilliger, V. Marembert, J. Berger, S. Ferber, E. Dietrich, R. Ludwig, B. Schmauss, H. G. Weber, “160 Gbit/s clock recovery with electro-
CHAPTER 4: Clock Recovery Techniques
149
optical PLL using bidirectionally operated electroabsorption modulator as a phase comparator”, Electron. Lett., vol. 39, no. 14, pp. 1071-1073, 2003
[25] O. Kamatani and S. Kawanishi, “Prescaled Timing Extraction from 400 Gb/s Optical Signal Using a Phase Lock Loop Based on Four-Wave-Mixing in a Laser Diode Amplifier”, IEEE Photon. Technol. Lett., vol. 8, no. 8, pp. 1094-1096, 1996
[26] H.-F. Chou, Z. Hu, J. E. Bowers, D. J. Blumenthal, K. Nishimura, R. Inohara, and M. Usami, “Simultaneous 160-Gb/s Demultiplexing and Clock Recovery by Utilizing Microwave Harmonic Frequencies in a Traveling-Wave Electroabsorption Modulator”, IEEE Photon. Technol. Lett, vol. 16, no. 2, pp. 608-610, 2004
[27] M. Shimanouchi, “An Approach to Consistent Jitter Modeling for Various Jitter Aspects and Measurement Methods”, International Test Conference, paper 30.3, pp. 848-857, November 2001
[28] Z. Hu, H.-F. Chou, K. Nishimura, M. Usami, J. E. Bowers, and D. J. Blumenthal, “Optical Clock Recovery Circuits Using Traveling-Wave Electroabsorption Modulator-Based Ring Oscillators for 3R Regeneration”, IEEE J. Sel. Topics Quantum Electron., vol. 11, no. 2, pp. 329-337, 2005
[29] Z. Hu, H.-F. Chou, J. E. Bowers, and D. J. Blumenthal, “40-Gb/s Optical Clock Recovery Using a Traveling-Wave Electroabsorption Modulator-Based Ring Oscillator”, IEEE Photon. Technol. Lett., vol. 16, no. 5, pp. 1376-1378, 2004
[30] Z. Hu, K. Nishimura, H.-F. Chou, L. Rau, M. Usami, J. E. Bowers, and D. J. Blumenthal, “40-Gb/s Optical Packet Clock Recovery Using a Traveling-Wave Electroabsorption Modulator-Based Ring Oscillator”, IEEE Photon. Technol. Lett., vol. 16, no. 12, pp. 2640-2642, 2004
[31] H.-C. Chang, X. Cao, M. J. Vaughan, U. K. Mishra, and R. A. York, “Phase Noise in Externally Injection-Locked oscillator Arrays”, IEEE Trans. Microw. Theory Tech., vol. 45, no. 11, pp. 2035-2042, 1997
[32] J. Lasri, and G. Eisenstein, “Phase Dynamics of a Timing Extraction System Based on an Optically Injection-Locked Self-Oscillating Bipolar Heterojection Phototransistor”, J. Lightwave Technol., vol. 20, no. 11, pp. 1924-1932, 2002
CHAPTER 4: Clock Recovery Techniques
150
This page intentionally left blank
151
Chapter 5 Wavelength Conversion 5.1 Introduction
The need for wavelength converters has grown significantly as the
wavelength division multiplexing (WDM) network evolves. The first generation
of WDM network multiplied the point-to-point transmission capacity of a single
fiber by tens or even hundreds of times by the efficient use of available
bandwidth in the fiber. In addition, wavelength can be utilized to perform
functions like routing and switching in order to manage the flow of signal
through the network. However, when the network grows larger, there are not
enough wavelengths to address all network nodes. Therefore, wavelength
converters are required to solve this problem [1]. With the availability of
wavelength conversion, the signal can be converted into different wavelengths
along its path to the final destination so that the same wavelength can be used in
CHAPTER 5: Wavelength Conversion
152
different parts of the network. As a result, the efficiency of wavelength
utilization is improved. As the network becomes more heavily loaded,
wavelength conversion is also critical to reduce blocking and increase
throughput.
The most straightforward wavelength converter is an O-E-O wavelength
converter that converts the input optical signal at one wavelength into an
electrical signal and then back into an optical signal at a new wavelength. The
technologies involved are similar to those of the transmitter and the receiver and
can be considered very mature. When regeneration of signal is necessary,
electronic signal processing such as clock and data recovery (CDR) can be used,
adding flexibility to this approach. As a result, O-E-O wavelength converters are
the choice for practical deployment.
However, the reasons that triggered enormous amount of research for
alternative wavelength conversion approaches are the high power consumption
and high cost of O-E-O wavelength converters. Limited format and protocol
transparencies are also concerns for future all-optical networks. The power and
cost issues are particularly severe when the number of wavelength counts and
the number of network nodes increase. Therefore, many new technologies are
being developed in the hope to provide lower cost, lower power consumption,
and higher transparency. Higher operation speed is also desired for WDM
networks with high-speed ETDM or OTDM. Many of these new technologies
CHAPTER 5: Wavelength Conversion
153
can even have regenerative capabilities that reduce the noise or increase the
extinction ratio of the signal after wavelength conversion. The scope of this
chapter is mainly confined to wavelength conversion. Regeneration of signal
will be discussed and studied in more detail in Chapter 6.
The common feature of the new wavelength conversion technologies is
to eliminate or reduce the use of electronic circuits which causes most of the
problems relating to power consumption, excessive noise, and limited speed.
These wavelength converters can be roughly categorized as either all-optical or
optoelectronic with integration. There are sub-classes under each category and
will be discussed in the rest of this section:
• All-Optical
• Passive Material
Nonlinear Wave-Mixing
Second-order nonlinearity: SHG and DFG
Third-order nonlinearity: FWM and XPM
• Active Material
Nonlinear Wave-Mixing
Third-order nonlinearity: FWM and XPM
Optical Gating
XAM and XPM with an interferometer
• Optoelectronic with Integration
• Parallel Integration
• Serial Integration
CHAPTER 5: Wavelength Conversion
154
All-Optical — In all-optical wavelength converters, the signal does not
go through any electronic circuit as the name suggests. Many of them utilize the
same principles as the all-optical gates introduced in Chapter 3. Under this
category, wavelength converters can be further classified by the type of material
used: passive or active.
Passive all-optical wavelength converters use nonlinear wave-mixing
effects in dielectric materials such as LiNbO3 crystals or silicon fibers to
generate new wavelengths. The nonlinearity can be second-order or third-order.
Second-order nonlinearity involves three interacting waves while third-order
nonlinearity involves four. The major advantages of passive all-optical
wavelength converters are high speed, low noise, and high transparency to data
format and phase information.
Second-order nonlinearity is in general much stronger than third-order
nonlinearity, which implies that a much shorter interaction length and/or lower
pumping power is required. However, only materials having asymmetric crystal
structure possess second-order nonlinearity. As a result, only third-order
nonlinearity can be utilized in fibers since silica is symmetric.
Despite of their higher efficiency, second-order nonlinear interactions
such as sum-frequency generation (SHG) and difference-frequency generation
(DFG) require tight phase matching between the interacting waves. This restricts
the operating wavelength and temperature range. The quasi-phase-matching
CHAPTER 5: Wavelength Conversion
155
(QPM) technique can be used to relax this restriction and also provide many
additional design flexibilities. For example, cascaded SHG and DFG can be
realized in a single periodically poled LiNbO3 chip so that the pump and the
input/output wavelengths can all reside in the C-band [2]. The advantage of this
type of wavelength converter is simultaneous conversion of several wavelengths
but the interacting wavelengths must satisfy energy conservation. The
photorefractive effect in LiNbO3 is an issue for long term stability.
Third-order nonlinear effects such as four-wave mixing (FWM) [3] and
cross-phase modulation (XPM) [4] in fibers can also be used for wavelength
conversion. Because of the weaker nonlinearity, a long interaction length on the
order of several of kilometers is generally required unless highly nonlinear
fibers are used which reduces the fiber length to less than 1 km [4]. Walk-off
between interacting waves is one of the main limiting factors together with the
need for high peak-power pump signals. Similar to the second-order case, it is
not possible for FWM to convert an unknown wavelength to a predetermined
wavelength. Therefore, they are more useful for large bandwidth conversions
with predetermined wavelengths. For example, a 640-Gb/s OTDM signal can be
converted between the C-band and the L-band by FWM in a highly nonlinear
fiber [5].
On the other hand, in XPM all-optical wavelength converter the output
wavelength is determined by the probe, which relieves the restriction mentioned
CHAPTER 5: Wavelength Conversion
156
above. The spectrum of the probe is broadened by a strong pump signal through
XPM. An offset optical filter is used to filter out part of the side-band and a non-
inverting signal very close to the wavelength of the probe can be obtained. Phase
matching is less a problem in XPM and the performance is mainly limited by
walk-off. Up to 160-Gb/s wavelength conversion has been demonstrated with
Raman enhanced XPM in highly nonlinear fibers [6].
The other class of all-optical wavelength converter uses active devices
made of semiconductor materials such as the semiconductor optical amplifier
(SOA) and the electroabsorption modulator (EAM). These devices require
current injection or bias voltage to operate. The wave-mixing techniques, FWM
and XPM, can also be implemented with these active devices [7]-[8]. However,
the physical mechanisms involved in the active material are more diverse and
complicated. The conversion performance and speed are generally limited by the
carrier dynamics. Nevertheless, wave-mixing type XPM wavelength conversion
up to 40-Gb/s was reported using the SOA [8].
Other than serving as a medium for nonlinear wave-mixing, an active
device can also be configured as an optical gate to switch the probe. The
mechanisms that can be utilized include cross-gain modulation (XGM) in SOAs,
cross-absorption modulation (XAM) in EAMs, and XPM in both types of
devices. However, in these gating operations, the phase information of the pump
CHAPTER 5: Wavelength Conversion
157
signal is not transferred to the output signal, which means that these wavelength
converters are only suitable for converting amplitude-shift keying (ASK) signal.
XGM in SOAs happens when a strong pump signal saturates the gain of
the SOA and the output power of the probe varies according the data carried by
the pump but the polarity inverted. Despite its simplicity, the operating speed is
limit by the carrier dynamics to about 20 Gb/s [9].
XAM in EAMs is analogous to XGM in SOAs. The absorption of probe
is modulated by the pump signal through absorption saturation. The speed of
operation is limited by the absorption recovery time which is directly related to
the carrier escape/swept-out time from the active region of the material. Due to
the high peak power required to saturate absorption, RZ data format is generally
required. As will be discussed in more detail in Ssection 5.2.1, there is a trade-
off between low pump power and high speed in the design of the active material.
Nevertheless, the speed can be better than XGM in SOAs and 8-channel 40-Gb/s
broadcasting wavelength conversion was reported [10].
On the other hand, with the addition of an interferometer, XPM can also
be used to realize an optical gate for wavelength conversion. Gating-type XPM
wavelength conversion using SOAs attracted intensive attention in recent years
[11]-[14]. The main reasons are: (1) XPM can be optimized to be faster than
XGM in SOAs; (2) output power is generally satisfactory since SOAs can
provide gain. Compared to XPM in passive materials, XPM in SOAs is not as
CHAPTER 5: Wavelength Conversion
158
fast because of the limitations set by carrier dynamics. However, the
compactness, high efficiency, and integration potential of semiconductor devices
still make this approach very attractive.
XPM in SOAs mainly arises from the change of carrier density induced
by the pump signal. The phase modulation must be converted to amplitude
modulation through an interferometer, which makes the configuration more
complicated than XGM. The speed limitation set by the carrier dynamics of the
SOA is manifested as a long recovery time, which influences both XPM and
XAM and limits XPM to the same speed as XGM. However, by using
differential techniques to take advantage of the faster rise time [11]-[13], the
long falling tail caused by carrier dynamics can largely be cancelled out by
destructive interference. Wavelength conversions up to 160-Gb/s were
demonstrated [12]-[13]. In addition, high-level monolithic integration is also
report recently [14], where a tunable laser and several SOAs are integrated on
the same chip, resulting in a very compact and efficient active wavelength
converter. Related XPM-based schemes include sophisticated optical filtering
[15] and a monolithically integrated Sagnac Interferometer [16].
Similarly, XPM in EAMs can also be utilized for higher speed
wavelength conversion. 100-Gb/s wavelength conversion was demonstrated
with delayed-interferometer [17], where the device is cooled to zero degree
CHAPTER 5: Wavelength Conversion
159
Celsius to blue-shift the bandgap so that a higher reverse bias is allowed to
increase the absorption recovery time.
The above mentioned all-optical wavelength converters do possess many
desirable properties such as high speed and low noise. However, long term
stability, wavelength and polarization sensitivity, and power efficiency can be
issues for many of these all-optical approaches. In addition, tunable or even
special filters are often required.
Optoelectronic with Integration — With the advance of integration
technology, integrated optoelectronic wavelength converters attract more and
more attention because of many of their inherent advantages over all-optical
approaches. For example, the receiver can be easily designed to be wavelength
and polarization insensitive. In most cases, tunable filters are not required since
the input and output optical waves are not mixed optically. Long term stability is
generally better in the optoelectronic approach since all-optical approaches are
more sensitive to environmental variations.
High level integration is the key to revive this approach, which provides
the potential to overcome the power consumption and cost issues of traditional
O-E-O wavelength converters. Some of the proposed approaches do not required
electrical amplifiers and signal processing circuits, which further reduces the
power consumption and excess noise caused by electronics.
CHAPTER 5: Wavelength Conversion
160
The operation speed of integrated optoelectronic wavelength converters
may not be as high as some of the all-optical ones, for example, up to 160 Gb/s.
However, in current WDM networks, the bit-rate for each wavelength is either
2.5 Gb/s or 10 Gb/s, with 40 Gb/s being considered for the next generation.
Given the recent developments in high-speed photodetectors and modulators,
there is a great possibility for integrated optoelectronic wavelength converters to
operate up to 40 Gb/s. In other words, unless high-speed OTDM is incorporated
with the WDM network, optoelectronic wavelength converters should be able to
catch up with the bit-rate of each wavelength channel.
The recent development of large-scale photonic integrated circuits
boosted the revival of the optoelectronic approach. In Ref. [18], 10×10 Gb/s
WDM transmitters were integrated on a single chip with 50 individual
functionalities. A matching 10-channel receiver array is also integrated
monolithically on a single ship. This type of integration is parallel since it
integrates different channels together. Significant saving in space and coupling
loss can be obtained. However, yield is a critical issue to keep the cost
competitive. In principle, this technology can enable a large-scale wavelength
conversion hub for wavelength routing.
The work presented in Ref. [18] still requires electronic amplification
and processing of signals between the integrated receivers and transmitters. The
other line of recent progress in photonic integration is serial, where the receiver
CHAPTER 5: Wavelength Conversion
161
and the transmitter are fabricated on a single chip, in contrast to integrating an
array of respective components. The main idea behind this approach is that the
photocurrent signal from the photodetector can be strong enough to drive a high-
efficiency modulator directly without going through electronic amplification.
This would result in a very compact optoelectronic wavelength converter but the
success of this approach would strongly rely on the gain provided by the on-chip
SOA and the high-power handling capability of the photodetector.
This serial integration concept was first demonstrated with discrete
components in Ref. [19] for 10-Gb/s NRZ data. Later on, a uni-traveling carrier
photodiode (UTC-PD) was monolithically integrated with a traveling-wave
electroabsorption (TW-EAM) and 100 Gb/s wavelength conversion was
demonstrated [20]. Ref. [21] integrates a 2×2 array of PD-EAM pairs as a
crossbar switch and 1.25 Gb/s operation is reported. Further integration with the
light source is also realized, where a tunable sampled grating distributed Bragg
reflector (SGDBR) laser is monolithically integrated with SOAs and a pair of
PD and EAM. 10-Gb/s NRZ wavelength conversion was demonstrated [22].
In this chapter, a novel wavelength converter based on a TW-EAM is
proposed and studied, which can be regarded as the merging of an all-optical
XAM wavelength converter and an integrated optoelectronic wavelength
converter. This unique combination makes it simple and versatile. This concept
is first verified at 2.5 Gb/s with NRZ data format. Then, considerable work is
CHAPTER 5: Wavelength Conversion
162
done to increase the speed up to 10 Gb/s. An alternative way for improving RZ
format conversion is also proposed and demonstrated, which provides additional
regenerative capabilities and will be utilized further in Chapter 6 for 3-R
regeneration. A first-order theoretical model is presented and verified to
simulate cascaded performance of the proposed wavelength converter. Part of
the research presented in this chapter is published in Ref. [23]-[24].
5.2 Photocurrent-Assisted Wavelength Conversion (PAW-Conversion)
In this section, a new mechanism for XAM-based wavelength conversion
using a TW-EAM is proposed and studied. This new mechanism utilizes the
photocurrent signal detected by the TW-EAM to re-modulate the device itself,
which is independent of the traditional XAM mechanism based on absorption
saturation. The associated wavelength conversion process is thus called
photocurrent-assisted wavelength conversion (PAW-Conversion). This concept
is first verified by converting a NRZ data at 2.5-Gb/s and then extended to 10-
Gb/s by improving the material design and device configurations.
5.2.1 Concept
The configuration of PAW-Conversion using a TW-EAM is shown
schematically in Figure 5.1(a). When the pump signal at λ1 is coupled into a
CHAPTER 5: Wavelength Conversion
163
reverse-biased TW-EAM, it gets absorbed and generates electron-hole pairs in
the active region. These carriers are swept to the electrodes by the external field
(due to reverse bias) and excite microwave propagation modes of the traveling-
wave electrodes. In this regard, the TW-EAM acts just like a high-speed
photodetector. In general, the higher the external field (reverse bias), the shorter
the carrier sweep-out time. The excited microwave signal, which will be called
“photocurrent signal” afterwards, propagates in both directions along the
traveling-wave electrodes. Since the TW-EAM is highly absorptive under
reverse bias, most of the pump is absorbed near the input facet of the active
waveguide. Half of the photocurrent signal propagates to the left (counter-
propagating) and finally goes to the termination at the lower electrical port. The
other half of photocurrent signal propagates to the right (co-propagating),
heading towards the upper electrical port. When the photocurrent signal is
passing through the rest of the active waveguide, it modulates the local voltage
around the DC bias and, as a result, changes the absorption experienced by the
probe at λ2. Therefore, the intensity information carried by the pump
wavelength is first transferred to photocurrent through photodetection and then
to the probe through electroabsorption, all in the same waveguide. Since
photocurrent plays an intermediate role in this process, this mechanism is called
photocurrent-assisted. The photocurrent signal can subsequently be received by
CHAPTER 5: Wavelength Conversion
164
outside electronics to provide signal monitoring capability, which adds an extra
functionality to PAW-Conversion and increases the overall power utilization.
(a)
(b) Figure 5.1 (a) Configuration of PAW-Conversion using a TW-EAM; (b) first-order model of PAW-Conversion. PD: photodetector; MOD: modulator; CW: continuous wave; Gray: optical
link; Black: electrical link.
In reality, the absorption and re-modulation processes happen gradually
and simultaneously along the active waveguide of the TW-EAM. A full
description would require a distributed model similar to the one presented in
Chapter 2. However, since the TW-EAM is reverse-biased, the pump signal can
CHAPTER 5: Wavelength Conversion
165
be absorbed within a short distance. For the TW-EAM used in the experiment,
over 90% of the coupled power can be absorbed within less than 100 µm.
Therefore, to the first-order of approximation, PAW-Conversion can be modeled
as a photodetector serially followed by an intensity modulator, as shown in
Figure 5.1(b). Numerical simulations based on this model will be presented in
Section 5.3.
PAW-Conversion is conceptually close to the integrated optoelectronic
wavelength converters mentioned in Section 5.1, where a modulator is directly
driven by a photodetector without electrical amplification [20]-[22]. In these
optoelectronic wavelength converters, a well-defined photodetector and a well-
defined modulator are connected electrically. However, In PAW-Conversion,
both are merged into one waveguide with the same material, reducing problems
such as microwave loss and impedance mismatch between the photodetector and
the modulator.
In the saturation mechanism of XAM, a large number of carriers must be
generated to produce sufficient field screening in order to get an appropriate
extinction ratio. This imposes a requirement for high pumping power. On the
other hand, to enable high-speed operation, these carriers must be swept out of
the active region as fast as possible to recover the absorption. Unfortunately, the
sweep-out time increases with the number of carriers, which limits the operation
speed [25]. From another point of view, if the active material of the EAM is
CHAPTER 5: Wavelength Conversion
166
designed to saturate easily (such as those having deep QWs and valance band
discontinuities) so that the pumping power required to reach saturation is
lowered, the carrier escape time of these material structures tends to be long and
the operation speed is limited. As a result, EAM-based wavelength converters
using the saturation mechanism have an inherent trade-off between speed and
pumping power.
However, in the proposed photocurrent-assisted mechanism, absorption
saturation due to field screening is not required and the material design is
completely opposite to those required by the saturation mechanism. For the
photocurrent-assisted mechanism, the material design is closer to that of a high-
saturation power EAM, where the photo-generated carriers in the active region
should be swept out as fast as possible so that saturation does not occur to distort
the detected photocurrent. This means that there is no design trade-off between
speed and pumping power in this new XAM mechanism. In general, the
pumping power can be lower than that required by the saturation mechanism,
and NRZ format conversion is possible at reasonable power levels.
Nevertheless, the extinction ratio still depends on the pumping power,
which determines the strength of photocurrent signal, and the efficiency of the
modulator. For a given material, by keeping the pumping power at an acceptable
level, the extinction ratio can be improved by factors such as the length of active
waveguide and the value of microwave impedance. The former can extend the
CHAPTER 5: Wavelength Conversion
167
interaction between the photocurrent signal and the probe while the later can
increase the amplitude of voltage modulation at a given photocurrent strength.
5.2.2 Proof of concept (2.5 Gb/s NRZ operation)
The operation of PAW-Conversion is first demonstrated at 2.5Gb/s with
NRZ data format as a proof of concept. To verify the existence of the
photocurrent-assisted mechanism, two kinds of termination are used at the lower
electrical port (Figure 5.1(a)). One is 50 Ohm and the other is open. The
photocurrent signal will be terminated without reflection by a 50 Ohm
termination. On the other hand, an open termination will reflect the photocurrent
signal back to the active waveguide and adds up with the co-propagating
tributary of photocurrent signal. If wavelength conversion is realized only by the
saturation mechanism, there should be no difference between the two
configurations.
The pump signal is 13 dBm at 1545.8 nm with transverse magnetic (TM)
polarization and the probe is 1 dBm CW at 1555.2 nm with transverse electric
(TE) polarization. TM polarization generates more photocurrent and TE
polarization has less loss for the device used in the experiment. The reverse bias
voltage is 0.8 V. A 2.4 nm optical band-pass filter is used to block the pump
after wavelength conversion, although a polarizer could also be used.
CHAPTER 5: Wavelength Conversion
168
Figure 5.2 Electrical and optical eye diagrams of 2.5Gb/s NRZ PAW-Conversion with 50 Ohm and open terminations.
Figure 5.3 Bit-error-rate curves of the (a) optical signals; (b) detected electrical signals
Figure 5.2 shows the electrical and the optical eyes for the two
terminations. The electrical eye amplitude for the 50 Ohm termination is 417
mVpp and 829 mVpp for the open termination, which is almost doubled because
of the reflection. The optical eye for open termination is larger in amplitude and
has a better extinction ratio than that of the 50 Ohm termination at the same bias,
CHAPTER 5: Wavelength Conversion
169
indicating that the photocurrent-assisted mechanism does exist and is capable of
influencing the wavelength conversion.
Bit-error-rate (BER) curves for the back to back and the wavelength
converted signals are shown in Figure 5.3(a). The optical power penalty for the
open termination is 0.5 dB and 1.5 dB for the 50 Ohm termination. The
difference in power penalty is caused by the difference in optical eye opening
and extinction ratio.
The photocurrent signal is fed into the BER tester without a matched
filter. At 13 dBm of pump power, no error was detected with both terminations
in a time period over 15 minutes. Therefore, the pump power is lowered to
measure the BER. The results are shown in Figure 5.3(b). The sensitivity at 10-9
BER is about 2 dB better for the open termination, mainly due to its larger
electrical eye amplitude. The slight decrease in slope should be caused by the
reflection from the open termination, which increases jitter and multiple
reflections along the electrical link. The traveling-wave electrode leading to the
open termination can be cut shorter to minimize the round trip time to the
termination, reducing intersymbol interference due to the reflection especially at
higher bit rates. Also, the reflections along the electrical link should also be
minimized for best performance.
For both terminations, the fall time of the detected electrical signal is
relatively long (about 300 ps to 400 ps) compared to the bit slot (Figure 5.2).
CHAPTER 5: Wavelength Conversion
170
This is due to the fact that the carrier sweep out time of the current TW-EAM is
not short enough and the speed as a photodetector is marginal for 2.5-Gb/s
operation. Fortunately, the E-O transfer function of the TW-EAM as a
modulator is highly nonlinear. The optical output of PAW-Conversion is re-
shaped by the nonlinearity and the resulting optical fall time is significantly
improved to 100 ps. In Chapter 6, this nonlinear property will be explored
further to provide strong re-shaping of ASK digital signals in a regenerator.
Figure 5.4 Power penalty and extinction of the wavelength converted signal at 2.5 Gb/s
The conversion is demonstrated in the entire C-band, as shown in Figure
5.4. The pump wavelength is fixed at 1545.8 nm. The bias voltage is adjusted to
optimize the BER for each wavelength. The power penalty of the converted
signal is less then 3 dB and can be as low as 0.5 dB. The increase of power
penalty near the pump wavelength is due to a finite bandwidth of the optical
CHAPTER 5: Wavelength Conversion
171
filter, which fails to block the pump sufficiently. By choosing a narrower filter
(< 2.4 nm) the smallest possible wavelength difference can be reduced. This is a
general restriction for all-optical wavelength converters requiring an optical
filter to separate the pump and the probe signals. This restriction is
disadvantageous to system level applications. Some of the proposed solutions
include the adopting of a counter-propagating configuration or the use of an
extra stage of wavelength converter.
The limit of performance is different at the short and the long
wavelength ends of the operating range. The short wavelength end is limited by
the increased loss of the TW-EAM which degrades the OSNR. The long
wavelength end is limited by the reduced modulation efficiency which decreases
the extinction ratio of the converted signal at the same pumping power. The
choice of the pump wavelength is not critical as the photocurrent changes by
less than 15% in the entire C-band.
At high pumping levels, the saturation mechanism may also contribute to
the wavelength conversion in a TW-EAM. To identify the contribution from
each mechanism, the pump signal is changed from NRZ data to a CW with
twice the average power. In terms of power level, this CW pump can simulate
the “1” bits when it is turned on, and when it is turned off, it simulates the “0”
bits. The difference in transmitted probe power between these two states is an
estimate of the contribution to the extinction ratio from the saturation
CHAPTER 5: Wavelength Conversion
172
mechanism because there is no microwave component in the CW photocurrent
to produce significant voltage modulation along the active waveguide. The DC
photocurrent generated by the CW pump does lead to a small amount of bias
change because of the series resistance along the traveling-wave electrode and
the external electrical connection. Any voltage drop across these resistances can
change the bias voltage seen by the TW-EAM. However, the total resistance is
most likely smaller than 1 Ohm and the change in bias can be neglected. In
contrast, the high frequency photocurrent generated by a data at the GHz range
sees a microwave impedance of around 25 Ohm along the traveling-wave
electrodes and generates a much larger voltage modulation around the bias
point.
The total extinction ratio can be estimated from the optical eye diagrams
measured on the sampling scope, which is limited to slightly over 10 dB of
resolution. The extinction ratio contribution from the photocurrent-assisted
mechanism can be obtained by subtracting the contribution due to saturation
from the total extinction ratio. Figure 5.5 shows the results for the two
terminations at different pump powers. It is observed that the extinction ratios
contributed by both mechanisms increase with pump power but the photocurrent
contribution increases slower, which can be explained by the fact that a larger
portion of the active waveguide is saturated and becomes less sensitive to
photocurrent modulation. However, the contribution from photocurrent is
CHAPTER 5: Wavelength Conversion
173
always higher than that from saturation in the open termination case. This
indicates that the photocurrent-assisted mechanism can be more effective than
the saturation mechanism in a TW-EAM. The photocurrent contribution with an
open termination is about twice as large as that with a 50 Ohm termination, in
close agreement with the difference in electrical eye amplitude. This is another
evidence of the photocurrent-assisted mechanism.
Figure 5.5 Measured extinction ratio compositions with (a) 50 Ohm termination; (b) open termination.
The fact that the saturation mechanism is also active in the current
generation of TW-EAM is not preferable from the point of view of the
photocurrent-assisted mechanism. When saturation occurs, the detection speed
CHAPTER 5: Wavelength Conversion
174
decreases. Therefore, to make the best use of the photocurrent-assisted
mechanism, saturation effects should be minimized.
5.2.3 Speed limitations
In the previous section, PAW-Conversion is successfully demonstrated
at 2.5 Gb/s. It is of great interest to increase the operation speed to higher bit
rates such as 10 Gb/s.
Figure 5.6 Electrical and optical eye diagrams measured at 10 Gb/s using the first generation TW-EAM. Upper: with NRZ data; Lower: with RZ data
Figure 5.6 shows the eye diagrams of the detected electrical and the
converted optical signals at 10 Gb/s using the current generation TW-EAM for
both the NRZ and the RZ data formats. In both cases, it is clear that the
detection speed of the TW-EAM is not fast enough to operate at 10 Gb/s. The
CHAPTER 5: Wavelength Conversion
175
electrical eye opening is small in the NRZ case and the eye of the converted
optical signal is almost closed. For the RZ case, the electrical eye opening is
larger than in the NRZ case because the power is more concentrated in time and
there is more room between symbols for a long falling tail. The converted
optical signal is a little bit refined due to the nonlinearity of TW-EAM but the
eye is still far from perfect.
To improve the performance, the detection speed of TW-EAM should be
increased. For a photodetector, the speed is determined by two factors: one is the
transit time in the active region and the other is the RC limited electrical
charging time. For a TW-EAM the later is less a problem because of the
traveling-wave design. The former is the bottleneck for high-speed operation.
Since the TW-EAM has a multiple QW design, the carriers must first escape
from the wells and then drift to the electrodes by the externally applied field.
The carrier escape probability depends exponentially on the barrier height of
QWs. Therefore, to reduce the carrier escape time, shallower QWs should be
used. This is particularly important for heavy holes because of a larger effective
mass. Also, valance band discontinuities should also be minimized. On the other
hand, an increased applied field also helps to sweep out the carriers by distorting
the shape of QWs (which lowers the effective barrier height) and by increasing
the drift velocity. Nevertheless, the reverse bias cannot be unconditionally high
because the total optical loss would be too much. Therefore, the turn-off voltage
CHAPTER 5: Wavelength Conversion
176
of the TW-EAM should be shifted to a higher bias point. This can be achieved
by lowering the photoluminescent wavelength of the material. In Ref. [17], the
EAM was even cooled to lower temperature to this purpose.
For the RZ data format, there is an assistive approach to improve the
performance, where the long falling tail in the converted optical signal is
removed by an externally applied electrical signal. This approach also provides
re-timing capability while improving the performance, which will be presented
in detail in Section 5.4. In the next section, the focus will be on 10-Gb/s NRZ
conversion using a new generation of TW-EAM with an improved material
design.
5.2.4 10-Gb/s NRZ operation using shallow QWs
In this section, the speed of NRZ PAW-Conversion is increased from 2.5
Gb/s to 10 Gb/s by optimizing and improving several device properties, which
include material design, microwave termination, and device length.
As mentioned in the previous section, the most critical bottleneck of
speed is the carrier escape time from the QWs. To reduce the carrier escape
time, shallower QWs are used in the new generation of TW-EAM. The material
structure is compatible with the integration platform reported in Ref. [14] and
[22], which means that a high-level integration with a tunable SGDBR laser and
SOAs are possible for ultra compact PAW-Conversion on a single chip.
CHAPTER 5: Wavelength Conversion
177
Figure 5.7 Electrical 10-Gb/s NRZ eye diagrams detected by (a) a previous generation 300-µm
TW-EAM with a 50-Ω termination, a 50-Ω probe, and Vb= – 0.7V; (b) a new generation 350-µm TW-EAM with an open termination, a 50-Ω probe, and Vb= – 2.1V
Figure 5.8 Normalized optical transmission curves of the previous and the new
Generations of TW-EAM with different lengths (TE polarization, 1555 nm)
CHAPTER 5: Wavelength Conversion
178
Figure 5.7 shows the 10-Gb/s NRZ electrical eye diagrams detected by
the previous and the new generation of TW-EAM, along with the barrier heights
of electrons and holes. It is clear that the new generation TW-EAM has a more
open eye with less pattern dependency. All the QW barrier heights are reduced
in the new generation, which accounts for the speed improvement.
Besides shallow QWs, the new generation TW-EAM also has additional
properties that are advantageous for improved performance. Figure 5.8 shows
the normalized transmission curves of the previous and the new generation TW-
EAMs with different lengths. The turn-off voltage of the new generation is
shifted towards higher reverse bias, which is desirable for increased speed. The
modulation efficiency (in terms of dB/V per unit length) of the new generation
is reduced compared to the previous generation. This could be a side effect of
shallower QWs. However, for the 350-µm long new generation TW-EAM, a 7
dB/V modulation efficiency can still be obtained.
Another advantageous property of the new generation TW-EAM is that
there is a relatively flat transmission region (flat top) before the optical
transmission drops substantially, as circled in Figure 5.8. The shorter the device,
the longer the flat top extends. The significance of the flat top is that it provides
re-shaping capability to PAW-Conversion. As illustrated in Figure 5.9, the flat
top of the transmission curve can clamp the mark level of the converted optical
signal and reduces the noises and fluctuations therein. This property is easier to
CHAPTER 5: Wavelength Conversion
179
obtain in TW-EAMs with a higher turn-off voltage. For the previous generation
TW-EAM (Figure 5.8), the transmission drops rapidly with reverse bias and
there is very little flat top under reverse bias.
Figure 5.9 Illustration of the re-shaping capability of PAW-Conversion through the utilization of nonlinear E-O transformation
The microwave property of the TW-EAM is also crucial for improving
the operation speed. In Section 5.2.2, it is pointed out that with an open
termination on one end of the traveling-wave electrode the photocurrent
amplitude in the active waveguide can be doubled which leads to a better eye
opening. However, the position of the open termination should be close to the
active waveguide so that the reflected signal does not cause intersymbol
CHAPTER 5: Wavelength Conversion
180
interference. Therefore, in the new generation of TW-EAM, one end of the
active waveguide is itself an open termination, as shown in Figure 5.10.
Figure 5.10 Configuration and measured results for a 500-µm TW-EAM (a) with a 50-Ω output
termination; (b) with a 25-Ω output termination
The impedance seen at the other end (output end) of the traveling-wave
electrode is of high importance. The impedance of the active waveguide is
CHAPTER 5: Wavelength Conversion
181
generally below 25 Ohm. If a standard 50-Ohm microwave probe is used to
collect the photocurrent, which also acts like a termination at the output end, an
impedance discontinuity occurs. This discontinuity will reflect a certain amount
of photocurrent signal back into the active waveguide. Unfortunately, the other
end of the active waveguide is terminated with an open, which reflects 100% of
the signal. As a result, multiple reflections take place inside the TW-EAM,
which leads to a long signal life-time and a low detection speed. To mitigate this
problem, the impedance of the probe should be reduced for impedance
matching. As shown in Figure 5.10(b), a parallel 50-Ohm resistor is added to
this purpose and the effective impedance of the probe seen by the TW-EAM is
reduced to 25 Ohm. Experimentally this is achieved by using a custom-made
microwave probe with a tiny 50-Ohm parallel resistor on the tip.
The effect of this approach is significant. Figure 5.10 shows the impulse
response traces of the TW-EAM when excited by a 0.5 ps, 20 MHz optical pulse
train. The device length is 500 µm for both cases. The bandwidth of the
sampling scope and the electrical cables combined is around 40 GHz. It is
observed from the impulse response traces that the width of the detected pulse is
shortened from 64 ps to 40 ps, indicating that the detection speed in the 25-Ohm
terminated case is improved. The improvement in detection speed is also evident
in the detected electrical eyes, also shown in Figure 5.10. The eye opening is
much better with 25-Ohm termination.
CHAPTER 5: Wavelength Conversion
182
Figure 5.11 Detected electrical and converted optical eyes of a 200-µm TW-EAM with a 50-Ω output termination at various bias points. The pump is 18 dBm, TE polarized.
Figure 5.12 Measured results for a 200-µm shallow QW TW-EAM with 18-dBm pump power. (a) optical extinction ratio and electrical eye amplitude; (b) receiver sensitivity of the converted
optical signal
CHAPTER 5: Wavelength Conversion
183
The choice of bias point for a given device and pump power involves the
balance between the clamping effect and the extinction ratio. As shown in
Figure 5.10(b), even with the improved material and microwave termination, the
eye of the photocurrent signal is still degraded by a limited rise / fall time and
the eye opening is only about 73%. To obtain a better eye opening in the
converted optical signal, the flat top of the nonlinear E-O transfer function can
be utilized.
Figure 5.11 shows the eye diagrams at three different bias voltages. At a
smaller reverse bias (Vb= – 2.0 V), the mark of the photocurrent signal sees the
flat top of the transfer function and the mark of the converted optical signal is
well clamped. However, the extinction ratio is only about 7 dB, which will result
in significant power penalty. When the reverse bias is higher (Vb= – 3.2 V), the
extinction ratio is improved, but the clamping effect is gone. Even worse, the
fluctuations on the mark of the converted signal are amplified because now the
mark of the photocurrent signal lies in a portion of the E-O transfer function
where the conversion slope is steep. Although the amplitude of the photocurrent
signal increases with reverse bias, it does not fully compensate the increase in
reverse bias and keep the mark within the flat top. Therefore, a proper bias
voltage is crucial to balance between the clamping effect and the extinction
ratio. Figure 5.12(b) shows that the best receiver sensitivity of the converted
signal can be obtained at – 3V of reverse bias using a 200-µm long shallow QW
CHAPTER 5: Wavelength Conversion
184
TW-EAM with 18 dBm of pump power. It can be inferred from this figure that
the clamping effect is more important to the current device since the receiver
sensitivity degrades more rapidly when the reverse bias increases beyond the
optimal point.
The last optimization process for 10-Gb/s operation is to choose the
length of the electroabsorption waveguide, which determines the final optical
transmission characteristics (Figure 5.8). The longer the device length is, the
steeper the transfer function and the lower the driving voltage for a given
extinction ratio. However, the flat top region will shrink in a longer device.
Therefore, there is a trade-off between efficiency and re-shaping capability in
choosing the device length.
From the point of view of a lumped circuit model, a longer device also
means a larger capacitance and hence a reduced RC-limited bandwidth.
Fortunately, the TW-EAM acts like a high-speed traveling-wave photodetector
and the bandwidth is not RC limited. Instead, it is impedance-mismatch limited,
as already discussed earlier in this section. As long as there is no impedance
discontinuity, the operation speed is only limited by the response of the material
and the loss and dispersion of the traveling-wave electrode.
BER test is use to determine the optimal device length, which is a
composite gauge of all the parameters. First, a 200-µm device is tested and the
results are shown in Figure 5.13. The CW probe is 3.6 dBm at 1555 nm, TE
CHAPTER 5: Wavelength Conversion
185
polarized while the NRZ pump is 18-dBm at 1545nm, also TE polarized. Both
the standard 50-Ohm and the reduced 25-Ohm output terminations are tested.
The bias voltage is adjusted in each case so that the power penalty is minimized.
Figure 5.13 Eye diagrams and BER curves of a 200-µm shallow QW TW-EAM. The output
wavelength is 1555 nm.
The detected electrical eye amplitude is almost halved in the 25-Ohm
case because the parallel 50-Ohm resistor splits the voltage. However, other than
amplitude, little difference in shape of the detected electrical eyes is observed.
This can be attributed to the fact that the 200-µm device is short (3.7 ps for
single pass) so that the life-time of the electrical signal in the TW-EAM “cavity”
is also short even with a reflective 50-Ohm termination.
CHAPTER 5: Wavelength Conversion
186
The power penalties are low in both cases. However, the power penalty
for the 50-Ohm terminated case is 0.2 dB lower because the optical extinction
ratio is a little bit higher. This is due that fact that the effective voltage swing in
the TW-EAM is larger in the 50-Ohm terminated case, which is a result of the
multiple reflections. The performance of this 200-µm TW-EAM is good but not
perfected yet because the extinction ratio is not high enough at this pump level,
which results in 0.5 to 0.7 dB power penalty.
Next, a 500-µm device is tested and the results are shown in Figure 5.14.
It is clear that the choice of output termination makes a significant difference. In
the 50-Ohm case, the electrical eye is distorted because of the multiple
reflections inside the TW-EAM. Even with the nonlinearity of the E-O transfer
function, the eye with the lowest possible power penalty is highly distorted and a
5.6 dB power penalty occurs. With a 25-Ohm output termination, the detected
electrical eye is almost the same as that of a 200-µm device, indicating that a
matched impedance is important for the TW-EAM to overcome the “RC-
limited” bandwidth as a photodetector.
Since the modulation efficiency is better in the 500-µm device
(compared to the 200-µm device), the extinction ratio of the converted optical
signal is improved. However, clamping effect is reduced in the 500-µm case
because the E-O transfer function becomes steeper and the flat top is reduced.
CHAPTER 5: Wavelength Conversion
187
Therefore, the same power penalty of 0.7 dB is obtained as in the 200-µm case
with a 25-Ohm output termination.
Figure 5.14 Eye diagrams and BER curves of a 500-µm shallow QW TW-EAM
Figure 5.15 Eye diagrams and BER curves of a 350-µm shallow QW TW-EAM
CHAPTER 5: Wavelength Conversion
188
It is then obvious that a length between 200 µm and 500 µm should be
able to balance between the extinction ratio and the clamping effect. Therefore,
a 350 µm device is prepared and tested. The results are shown in Figure 5.15.
The converted optical eye is very close to that of the input signal with an
extinction ratio over 10 dB and sufficient clamping on the mark level. The
power penalty is as low as 0.1 dB.
The operation wavelength range of the optimized PAW-Converter is
tested in the entire C-band. First, the pump wavelength is fixed at 1545 nm and
the output wavelength is varied in the C-band. Figure 5.16(a) shows the receiver
sensitivity as function of output wavelength. The back to back receiver
sensitivity at 1545 nm is – 28.8 dBm. The bias voltage of the TW-EAM is
adjusted in the range of – 2 V to – 3.3 V for each wavelength to optimize the
receiver sensitivity. For several output wavelengths around 1560 nm, receiver
sensitivities better than the back to back value are obtained. This is mainly
caused by the gain variation of the EDFA in the receiver. The variation of
EDFA gain is also plotted in Figure 5.16(a). In Figure 5.16(b), the power
penalty of the converted optical signal after offsetting the EDFA gain variation
is shown. In the entire C-band, the power penalty is below 0.8 dB and averaged
at 0.45 dB.
CHAPTER 5: Wavelength Conversion
189
Figure 5.16 Output wavelength dependence in the C-band. (a) receiver sensitivity (circle) and
EDFA gain (square) as function of wavelength; (b) EDFA gain adjusted power penalty
Figure 5.17 Pump wavelength dependence in the C-band
Next, the output wavelength is fixed at 1555 nm and the pump
wavelength is varied in the C-band, without changing the bias voltage. The
results are shown in Figure 5.17. For pump wavelengths in the range of 1535 nm
to 1560 nm, the receiver sensitivity of the converted signal does not change
CHAPTER 5: Wavelength Conversion
190
significantly. However, it degrades by 2 dB when the pump wavelength is
further increased to 1565 nm. This is due to the fact that the absorption at longer
wavelengths is reduced, which results in a smaller photocurrent and, therefore, a
lower output extinction ratio. If the bias voltage is increased to – 3.1 V, the
absorption of the pump can be increased and receiver sensitivity similar to those
with shorter pump wavelengths can be obtained.
In summary, the high-performance 10-Gb/s NRZ PAW-Conversion is
achieved by first improving the QW structure of the TW-EAM in order to
reduce the carrier escape time. Then the microwave terminations on both ends of
the traveling-wave electrode are properly arranged to optimize the electrical
bandwidth. Next, the flat top of the nonlinear E-O transfer function is utilized to
clamp the mark level in order to compensate the distortion in the photocurrent
signal caused by a marginal detection bandwidth. Finally, the device length and
the bias voltage are properly tuned to balance the clamping effect and the
extinction ratio. The result is an ultra-low power penalty PAW-Converter for 10-
Gb/s NRZ data format. Operation in the entire C-band is also demonstrated with
low power penalties in the range of 0.2 to 0.8 dB.
5.3 Modeling of PAW-Conversion
A straight forward model is presented in this section for PAW-
Conversion. Despite its simplicity, simulations based on this model match the
CHAPTER 5: Wavelength Conversion
191
experimental results very well. This model is then applied to simulate cascaded
performance of PAW-Conversion, which is an important issue in WDM
networks.
5.3.1 Model
The first-order approximation of PAW-Conversion using a TW-EAM is
illustrated in Figure 5.18(a). As discussed in Section 5.2.1, the first-order model
treats the detection and the re-modulation processes as sequential events by
separating the TW-EAM into two discrete components: a photodetector and an
intensity modulator.
Figure 5.18 Model of PAW-Conversion (a) first-order model; (b) sequence of simulation. PD: photodetector; MOD: modulator; Gray: optical link; Black: electrical link
CHAPTER 5: Wavelength Conversion
192
The distributed effects are not considered and both devices are modeled
as lumped elements. These assumptions are validated since (1) the pump signal
is absorbed quickly within a short distance; (2) for lower operation speeds where
the microwave wavelength is much longer than the device dimension, the
modulator can be treated as a lumped element, as concluded in Chapter 2. For
example, the microwave wavelength at 10 GHz in the active waveguide is about
6000 µm, which is much longer than the typical TW-EAM length of 200 µm ~
500 µm.
The logic flow of the simulation is summarized in Figure 5.18(b), which
follows the first-order model. First, the input optical signal is generated. Pseudo-
random-bit-sequence (PRBS) is assumed and the pattern length is set at 27-1.
Longer pattern lengths would require more computation time and computer
memory. Noise is not included into the simulation. The main purpose of this
modeling work is to study the dependence of the converted waveform on several
critical device parameters and to see how the waveform evolves after cascaded
conversions. A detailed model with complete noise and BER calculations is
beyond the scope of this dissertation. The simulation is restricted to NRZ format
in this section, although the model applies to any amplitude-shift keying signal.
In Chapter 6, this model will be extended and applied to simulate 3-R
regeneration of RZ signals.
CHAPTER 5: Wavelength Conversion
193
Figure 5.19 Experimentally measured and numerically simulated 10-Gb/s NRZ pump signal
First, the experimental results of 10-Gb/s NRZ PAW-Conversion are
simulated to verify the effectiveness of this first-order model. Figure 5.19 shows
the experimentally measured and the simulated 10-Gb/s NRZ pump signals. The
simulated pattern is obtained by convolving a perfectly sharp NRZ 27-1 PRBS
with an impulse response p(t):
( )ca tbttp ⋅−⋅= exp)( (5.1)
where a = 2, b = 0.154, c = 1. In the frequency domain, this impulse response
has a 3-dB bandwidth of 12.5 GHz. The Fourier transform of p(t) can be
regarded as the E-O transfer function of the transmitter.
The simulated 10-Gb/s NRZ pump signal is then convolved with the
impulse response of the photodetector to obtain a simulated photocurrent signal
(detected signal). Instead of using a measured impulse response, an impulse
response in the form of equation (5.1) is assumed. The parameters, a, b, and c
are varied so that the simulated eyes of the photocurrent signal fit the measured
CHAPTER 5: Wavelength Conversion
194
ones. Figure 5.20 shows the measured and the simulated eyes. Very close
matching is obtained for three different cases. The fitted parameters are listed in
Table 5.1.
(a)
(b)
Figure 5.20 (a) Measured; (b) simulated 10-Gb/s NRZ eye diagrams for three TW-EAMs
CHAPTER 5: Wavelength Conversion
195
500 µm, 50 Ω 500 µm, 25 Ω 350 µm, 25 Ω
a 0.2 0.2 0.2 b 0.0256 0.0314 0.0467 c 1.0 1.0 1.0
Table 5.1 Fitted parameters of the photodetector impulse response
Next, the simulated photocurrent signal is used to drive the modulator.
The bandwidth of the modulator is assumed to be infinite since it is integrated in
the same waveguide with the photodetector. This is generally valid for 10-Gb/s
or lower speed. The E-O transfer function measured with CW input (plotted in
Figure 5.8) is used to describe the modulator. The bias voltage and the peak to
peak amplitude of the photocurrent signal are varied to fit the measured
converted optical signal and the fitted parameters are listed in Table 5.2.
500 µm, 50 Ω 500 µm, 25 Ω 350 µm, 25 Ω
Bias voltage – 3.0 V – 3.0 V – 3.8 V
Vpp 2.15 V 2.3 V 3.2 V
Table 5.2 Fitted parameters of the modulator
Figure 5.20 shows that the simulated outputs are almost the same as the
measured results, indicating that this simple first-order model is very effective in
describing the PAW-Conversion. The fitted impulse responses and the
corresponding power spectrum for the three simulated cases are plotted in
CHAPTER 5: Wavelength Conversion
196
Figure 5.21. The 3-dB photodetection bandwidth of the best case is only 6.6
GHz but thanks to the nonlinear E-O transfer function of the TW-EAM, the
converted optical signal is well shaped and has a very low power penalty as
experimentally demonstrated in the previous section.
Figure 5.21 Plots of (a) the photodetector impulse response and (b) the corresponding power
spectrum of the three simulated cases
5.3.2 Simulation of cascaded conversions
In the WDM network, the transmitted data may undergo several times of
wavelength conversion for routing or network management purposes before
reaching the final destination. Therefore, it is crucial to study the cascaded
performance of wavelength converters. Experimentally, it can be done in the
laboratory by constructing a re-circulating loop. Theoretically, numerical
simulations can be used as an alternative way to study cascaded performance.
Since the effectiveness of the first-order model is verified in the previous
CHAPTER 5: Wavelength Conversion
197
section, it can be applied to study the cascaded performance of PAW-
Conversion.
Figure 5.22 Simulated optical eye diagrams of cascaded 10-Gb/s NRZ PAW-Conversions (a) original input signal; (b)-(f) after 1, 2, 4, 6, and 8 PAW-conversions. The gray blocks
in (a) and (d) show how the eye height is measured
First, the best experimental case is simulated: the 350-µm long TW-
EAM with a 25-Ohn output termination. In the simulation, transmission effects
such as the dispersion of fiber are not considered between the PAW-Converters.
This should reflect the best scenario of cascaded conversions. Inclusion of fiber
transmission effects is technically feasible, which has been studied extensively.
The PAW-Converters are assumed to be connected back to back with the same
CHAPTER 5: Wavelength Conversion
198
optical input power. The bias voltages as well as E-O transfer functions are fixed
for each PAW-Converter.
Figure 5.22 shows the simulated optical eye diagrams for the 350-µm
TW-EAM. The pump power for each stage is the same as in the experiment
presented in the previous section (18 dBm). It is observed that the eye opening
shrinks after each conversion because of the finite bandwidth (rise and fall
times) of the converter. The distortion of eye is significant after 4 conversions.
Note that the experimentally measured power penalty after a single PAW-
Conversion is as low as 0.2 dB (EDFA gain calibrated). Even so, the conversion
can only be cascaded less than 6 times in the back to back scenario.
Since noise is not included in the simulation, BER cannot be calculated
as a quantitative gauge of signal quality. Instead, eye height penalty is used. The
eye height of the n-th conversion, Hn, is defined as the maximum height of a
rectangular block that can fit into a normalized eye. Examples are shown in
Figure 5.22(a) and (d) as gray blocks. The width of the block is 0.2 unit interval
(UI), which equals 20 ps in this case. The eye height penalty is defined as:
⋅=
n
original
H
HPenaltyHeightEye log10
(5.2) where Horiginal, is the eye height of the original input signal.
CHAPTER 5: Wavelength Conversion
199
The calculated eye height penalties for the three cases studied in the
previous section are plotted in Figure 5.23. If arbitrarily a 2-dB limit is set, the
maximum number of conversions are 1, 2, and 5 for the (500 µm, 50 Ohm), the
(500 µm, 25 Ohm), and the (350 µm, 25 Ohm) cases. The biggest difference
between these three cases is the difference in detection bandwidth. The (350 µm,
25 Ohm) case has the highest bandwidth and hence the highest number of
acceptable cascades.
Figure 5.23 Eye height penalty as function of conversions for the three cases
For the (350 µm, 25 Ohm) case, if the detection bandwidth can be further
improved, the acceptable number of cascades can be increased. Figure 5.24(a)
shows that with the same 2-dB limit, the acceptable number of cascades can be
increased from 5 to 10 if the detection bandwidth can be raised from 6.6 GHz to
10 GHz. On the contrary, if the detection bandwidth is reduced to 5 GHz the
number is reduced to 3.
CHAPTER 5: Wavelength Conversion
200
Figure 5.24 Eye height penalties of the 350-µm TW-EAM with a 25-Ω output termination. (a) dependence on detection bandwidth; (b) dependence on photocurrent signal amplitude. The
circle represents the case that matches the experimental conditions in Section 5.2
The optical pump power is also an influential factor on the acceptable
number of cascades. The higher the pump power is, the larger the photocurrent
amplitude. Surely the pump power must be within the limits set by the power
handling capability of the TW-EAM and the available gain and OSNR of the
wavelength converter subsystem. The advantage of increasing the photocurrent
amplitude is that more clamping effect due to the flat top of the nonlinear E-O
transfer function can be utilized while keeping an acceptable extinction ratio.
This effectively results in better rise and fall times. Figure 5.24(b) shows that
when the peak to peak amplitude is increased from the fitted value of 3.2 V to
4.4 V, which corresponds to a 1.38 dB increase in pump power, the acceptable
number of cascades can increase from 5 to 7.
CHAPTER 5: Wavelength Conversion
201
It can be concluded that for NRZ data format, the acceptable number of
cascades is ultimately limited by the finite bandwidth of the wavelength
converter. In the conventional optoelectronic approach, a D-flip-flop circuit can
be used so that the transition edge of each bit can be regenerated and the
acceptable number of cascades is limited by other factors such as noise. In
Chapter 6, several 3-R regenerators capable of wavelength conversion are
introduced and simulations show that for RZ data format the limitation due to a
finite conversion bandwidth can be removed.
5.4 RF-Driven PAW-Conversion (PAW-Regeneration)
Fro the RZ data format, the speed of PAW-Conversion not only can be
increased by the improvements introduced in the previous sections, but also by
an assistive technique which involves feeding a synchronize electrical clock to
the TW-EAM in order to re-shape and re-time the converted output. Since the
TW-EAM is driven by an electrical signal, this approach is termed as RF-driven
PAW-Conversion. In addition, because this approach has regenerative
capabilities, it is also abbreviated as PAW-Regeneration, which stands for
“Photocurrent-Assisted Wavelength Conversion-based Regeneration”.
CHAPTER 5: Wavelength Conversion
202
5.4.1 Concept
As pointed out in Section 5.2.3, the problem with RZ PAW-Conversion
is that the response of the TW-EAM as a photodetector may not be fast enough
to resolve the pulse shape of the input signal and causes distortions such as a
long falling tail in the photocurrent signal. As shown in Figure 5.25(a), the
converted signal carries essentially the same distortion, resulting in excess
power penalty. This problem is solved in PAW-Regeneration by supplying a
synchronized electrical clock to the TW-EAM. As shown in Figure 5.25(b), the
photocurrent signal adds with a stronger sinusoidal clock with a properly
adjusted phase and the long tails are pulled down. This process also suppresses
the noise between bits. The shape and timing of the converted signal are thus
mainly determined by the sinusoidal clock while information is transferred from
the input by the photocurrent signal. As a result, re-timing and lateral re-
shaping can be obtained.
Vertical re-shaping can be realized by utilizing the nonlinear E-O
transfer function of the TW-EAM. Figure 5.25(b) shows that by setting the bias
voltage properly, the mark level can fall in the flat-top region of the transfer
function and noise and fluctuation in the mark can be compressed after E-O
transformation. On the other hand, as long as the electrical eye opening, ∆V, is
large enough so that the extinction ratio of the converted signal, ∆T, is better
than that of a degraded input signal, noise in the space level can be suppressed in
CHAPTER 5: Wavelength Conversion
203
the converted signal. In other words, vertical re-shaping is achieved by
redistributing noise through the non-linear E-O transformation.
Figure 5.25 E-O transformation of PAW-Conversion (a) without RF-drive; (b) with RF-drive
This RF-driven approach was originally proposed in Ref. [26] to re-
shape the converted RZ signal in the conventional EAM-based wavelength
converter that utilizes the saturation mechanism. According to Ref. [26], the
applied RF-drive can increase the bias voltage momentarily to reduce the carrier
sweep-out time. Consequently, the tails of the converted RZ signal can be
trimmed. When this RF-driven approach is applied to PAW-Conversion, the RF-
drive not only increases the bias voltage momentarily to assist carrier sweep-out
but also adds with the photocurrent signal and re-shapes it. Therefore, the
CHAPTER 5: Wavelength Conversion
204
trimming effect is further enhanced through the nonlinear E-O transformation
and the shape of the converted signal is mainly determined by the strong RF-
drive.
5.4.2 10-Gb/s RZ operation
The RF-driven PAW-Conversion of RZ data is demonstrated at 10-Gb/s
using the original generation of TW-EAM (without shallow QWs). The
configuration is plotted in Figure 5.26. A synchronized 10-GHz sinusoidal
electrical clock is coupled into the TW-EAM with a proper phase through a
power divider from the upper electrical port. The third end of the power divider
is terminated by 50 Ohm, which is connected to the input port of the sampling
scope during the experiment. The lower electrical port of the TW-EAM can be
terminated with open or 50 Ohm.
In the experiment, the pump is a 16-ps, 13-dBm RZ signal at 1545 nm
with 231-1 PRBS. The CW probe is 1-dBm at 1555nm. Figure 5.27 shows the
electrical and the optical eyes without the RF-drive. The pulse shape of the
photocurrent signal is distorted by the long falling tail and so is the converted
optical signal. With an open termination, the detected photocurrent signal is
larger in amplitude compared to that with a 50-Ohm termination. However, the
fall time is much worse because of the multiple reflections. In the end, the
CHAPTER 5: Wavelength Conversion
205
effective electrical eye openings are similar for both terminations, which lead to
the same amount of power penalty, as high as 7.5 dB.
Figure 5.26 Configuration of RF-driven PAW-Conversion (PAW-Regeneration)
Figure 5.27 Electrical and optical eye diagrams of 10-Gb/s RZ PAW-Conversion without RF-drive. Pump: 13 dBm at 1545 nm; CW probe: 1-dBm at 1555 nm
CHAPTER 5: Wavelength Conversion
206
Next, the RF-drive is turned on, which is a 6 Vpp, 10-GHz sinusoidal
electrical clock from the transmitter. The phase is adjusted so that the peak of
the clock signal coincides with the peak of the detected electrical signal. Figure
5.28 shows the measured electrical and optical eye diagrams. The electrical eye
is a combination of the photocurrent signal and the applied RF-drive. The small
splitting of the sinusoidal waves is the eye of the photocurrent signal. This
shows that the applied clock is much stronger than the photocurrent signal. The
relative strengths of the applied clock and the photocurrent signal measured on
the scope are different for the two terminations because the measure ports and
the electrical configurations are not the same. BER measurement of these
electrical signals is possible if a matched low-pass filter is provided to remove
the strong 10 GHz tone.
Figure 5.28 Electrical and optical eye diagrams of 10-Gb/s RF-driven PAW-Conversions
CHAPTER 5: Wavelength Conversion
207
The converted optical eyes are well shaped without long tails, showing
the effectiveness of this RF-driven approach. The converted pulsewidth is 16 ps.
The bias voltages were adjusted to minimize the power penalty. The extinction
ratios are better than 10 dB for both terminations. Figure 5.29 shows the BER
results. Significant improvement from the original PAW-Conversion is evident.
The power penalty is 1.0 dB for the 50-Ohm case and 1.8 dB for the open case.
The higher penalty for the open case is caused by the multiple reflections
between the open termination and the microwave circuits on the other end (such
as the amplifier). This can be identified from the converted eyes in Figure 5.28,
where there is more noise on the optical eye with open termination. The 1.0 dB
power penalty for the 50-Ohm case can be attributed to (1) a finite extinction
ratio; (2) the intersymbol interference caused by a marginal detection
bandwidth; (3) the residue reflections from the amplifier.
Figure 5.29 BER curves of 10-Gb/s RZ PAW-Conversions with and without RF-drive
CHAPTER 5: Wavelength Conversion
208
It should be pointed out that this RF-driven approach can improve the
performance of RZ PAW-Conversion only when the detection speed of the TW-
EAM is at least marginal. If the detection speed is way too slow, the eye
opening will be reduced significantly and the intersymbol interference will be
serious. Even though these problems can be mitigated by the nonlinear E-O
transformation, they should be kept as low as possible for better performance.
5.4.3 Regenerative properties
The applied electrical clock not only pulls down the long falling tail of
the converted signal but also re-times it. To evaluate the re-timing capability of
the RF-driven PAW-Conversion, the pulse train in the transmitter is frequency
modulated (FM) to simulate a jittered input signal. The FM modulation is 1-
MHz with a modulation index of 0.3, which corresponds to ± 4.8 ps peak to peak
timing jitter. The sinusoidal clock supplied to the TW-EAM remains un-
modulated.
Figure 5.30 shows the eyes of the FM modulated pump signal and the
converted signal. Because of the FM modulation, the trace of the pulse is
broadened. The standard deviation of the pump signal measured with the
sampling scope is 3.5 ps. After the conversion, the standard deviation is reduced
to 1.8 ps, showing effective re-timing capability. However, there is still a 1.2 dB
of power penalty caused by effects such as a limited extinction ratio. A higher
CHAPTER 5: Wavelength Conversion
209
pumping power will increase the extinction ratio and reduce the power penalty,
which will be demonstrated in Chapter 6 as a compact 3-R PAW-Regenerator.
RF-driven PAW-Converter can consistently reduce the standard deviation of
trace fluctuations by half up to a FM modulation index of 0.6, as shown in
Figure 5.31(a).
Figure 5.30 Input and output eye diagrams showing re-timing (upper) and re-shaping (lower) capabilities of RF-driven PAW-Conversion
In the electrical spectrum of the original and the converted optical
signals, the FM modulation generates side-modes spaced by 1 MHz around the
10 GHz clock tone. The relative strength of the first-side mode to the carrier
(clock tone) can be regarded as a measure of timing fluctuation in the frequency
domain. Figure 5.31(b) shows that the first side-mode suppression ratio can be
increased by 10 dB after conversion, which is another indication of the re-timing
CHAPTER 5: Wavelength Conversion
210
capability of RF-driven PAW-Conversion. Re-shaping capability is also
demonstrated in Figure 5.30(b), where the pump pulse is intentionally broadened
to 30 ps but the average power is still 13 dBm. The converted signal has a
reduced pulsewidth of 16 ps, which is determined by the RF-drive. This
confirms the lateral re-shaping capability of RF-driven PAW-Conversion.
Figure 5.31 Measurement of re-timing capabilities of RF-driven PAW-Conversion. (a) standard deviation of the pulse traces; (b) First side-mode suppression ratio
5.4 Summary
In this chapter, PAW-Conversion using a TW-EAM was proposed and
studied. The configuration of PAW-Conversion resembles the traditional all-
CHAPTER 5: Wavelength Conversion
211
optical wavelength conversion based on absorption saturation of an EAM but
the underlying mechanism is closer to the integrated optoelectronic wavelength
converter. The pump power required for NRZ conversion using the current
generation of TW-EAM is comparable to that required for RZ conversion with a
10% pulsewidth using a lumped-EAM [27]. This would suggest a 10 dB higher
pumping power if the NRZ conversion were carried out with the lumped-EAM.
This indicates the advantage of the newly proposed photocurrent-assisted XAM
mechanism.
To increase the operating speed from 2.5 Gb/s to 10 Gb/s, improvements
in the material design as well as optimization works in microwave configuration,
device length, and bias voltages were presented. Very low power penalty
wavelength conversion of 10-Gb/s NRZ signal was demonstrated in the entire
C-band.
A first-order model was proposed and its validity was verified by
matching with the experimental results. This model was then applied to simulate
cascaded performance of 10-Gb/s NRZ PAW-Conversion. Simulation results
suggested that with the current experimental condition, the 10-Gb/s NRZ PAW-
Conversion can be cascaded less than 6 times. For a better cascaded
performance, the detection speed of the TW-EAM should be further improved.
To improve the performance of RZ signal conversion, the RF-driven
PAW-Conversion was proposed, which successfully reduced the power penalty
CHAPTER 5: Wavelength Conversion
212
by up to 6.5 dB for a bandwidth limited TW-EAM. Regenerative capabilities
such as re-timing and re-shaping were also demonstrated with this technique. In
Chapter 6, the RF-driven PAW-Converter will be combined with the injection-
locking clock recovery technique introduced in Chapter 4 to realize a very
compact optical 3-R regenerator using only a single TW-EAM.
References
[1] S. J. B. Yoo, “Wavelength Conversion Technologies for WDM Network Applications”, J. Lightwave Technol., vol .14, no. 6, pp. 955-966, 1996
[2] M. H. Chou, I. Brener, M. M. Fejer, E. E. Chaban, and S. B. Christman, “1.5-mm-Band Wavelegnth Conversion Based on Cascaded Second-Order Nonlinearity in LiNbO3 Waveguides”, IEEE Photon. Technol. Lett., vol. 11, no. 6, pp. 653-655, 1999
[3] O. Aso and S. Nsmiki, “Recent advances of ultra-broadband fiberoptics wavelength converters”, presented at the Eur. Conf. Optical Communication (ECOC), Munich, Germany, Sep. 3-7, 2000. Paper ThA3
[4] W. Wang, H. K. Poulsen, L .Rau, H.-F. Chou, J. E. Bowers, and D. J. Blumenthal, “Raman-Enhanced Regenerative Ultrafast All-Optical Fiber XPM Wavelength Converter”, J. Lightwave Technol., vol .23, no. 3, pp. 1105-1115, 2005
[5] H. Sotobayashi, W. Chujo, T. Ozeki, “Inter-wavelegnth-band conversions and demultiplexings of 640 Gbit/s OTDM signal”, Optical Fiber Communication Conference (OFC), paper WM2, pp. 261-262, 2002
[6] W. Wang, L. Rau, D. J. Blumenthal, “160 Gb/s Variable Length Packet/10 Gb/s-Label All-Optical Label Switching With Wavelength Conversion and Unicast/Multicast operation”, ”, J. Lightwave Technol., vol .23, no. 1, pp. 211-218, 2005
[7] J. Zhou, N. Park, K. J. Vahala, M. Newkirk, and B. I. Miller, “Four-wave mixing wavelength conversion efficiency in semiconductor travelin-wave amplifiers measured to 65 nm of wavelength shift”, IEEE Photon. Technol. Lett., vol. 6, no. 8, pp. 984-987, 1994
[8] M. L. Nielsen, B. Lavigne, and B. Dagens, “Polarity-preserving SOA-based wavelength conversion at 40 Gbit/s using bandpass filtering”, Electron. Lett, vol. 39, no. 18, pp. 720-21, 2003
[9] J. M. Wiesenfeld, J. S. Perino, A. H. Gnauck, and B. Glance, “Bit error rate performance for wavelength conversion at 20 Gbit/s”, Electron. Lett, vol. 30, no. 9, pp. 720-721, 1994
CHAPTER 5: Wavelength Conversion
213
[10] L. Xu, N. Chi, K. Yvind, L. J. Christiansen, L. K. Oxenløwe, J. Mørk, P. Jeppesen, and J. Hanberg, “8×40 Gb/s RZ all-optical broadcasting utilizing an electroabsorption modulator”, Optical Fiber Communication Conference (OFC), paper MF71, vol. 1, 2004
[11] D. Wolfson, A. Kloch, T. Fjelde, C. Janz, B. Dagens, and M. Renaud, “40-Gb/s All-Optical Wavelegnth Conversion, Regeneration, and Demultiplexing in an SOA-Based All-Active Mach-Zehnder Interferometer”, IEEE Photon. Technol. Lett., vol. 12, no. 3, pp. 332-334, 2000
[12] J. Leuthold, L. Möller, J. Jaques, S. Cabot, L. Zhang, P. Bernasconi, M. Cappuzzo, L. Gomez, E. Laskowski, E. Chen, A. Wong-Foy, and A. Griffin, “160 Gbit/s SOA all-optical wavelength converter and assessment of its regenerative properties”, Electron. Lett, vol. 40, no. 9, pp. 554-555, 2004
[13] S. Nakamura, Y. Ueno, and K. Tajima, “168-G/bs All-Optical Wavelength Conversion With a Symmetric-Mach-Zehnder-Type Switch”, IEEE Photon. Technol. Lett., vol. 13, no. 10, pp. 1091-1093, 2001
[14] M. L. Masanovic, V. Lal, J. A. Summers, J. S. Barton, E. J. Skogen, L. G. Rau, L. A. Coldren, and D. J. Blumenthal, “Widely Tunable Monolithically Integrated All-Optical Wavelegnth Converters in InP”, J. Lightwave Technol., vol .23, no. 1, pp. 211-218, 2005
[15] J. Leuthold, J. Jaques, and S. Cabot, “All-optical wavelength conversion and regeneration”, Optical Fiber Communication Conference (OFC), paper WN1, vol. 1, 2004
[16] V. M. Menon, W. Tong, C. Li, F. Xia, I. Glesk, P. R. Prucnal, and S. R. Forrest, “All-Optical Wavelegnth Conversion Using a Regrowth-Free Monolithically Integrated Sagnac Interferometer”, IEEE Photon. Technol. Lett., vol. 15, no. 2, pp. 254-256, 2003
[17] K. Nishimura, R. Inohara, and M. Usami, “100 Gbit/s wavelength conversion using MQW-EAM with blueshifted absorption edge”, Optical Fiber Communication Conference (OFC), paper FD2, vol. 2, 2004
[18] R. Nagarajan et. al, “Large-Scale Photonic Integrated Circuits”, IEEE J. Select. Topics Quantum Electron., vol. 11, no. 1, pp. 50-65, 2005
[19] M. Yoneyama, Y. Miyamoto, N. Shimizu, K. Hagimoto, and T. Ishibashi, “Simple wavelength converter using an optical modulator directly driven by a uni-traveling-carrier photodiode”, ”, Electron. Lett, vol. 34, no. 23, pp. 2244-2245, 1998
[20] T. Yoshimatsu, S. Kodama, K. Yoshino, and H. Ito, “100-Gbit/s Error-Free Wavelegnth Conversion Using PD-EAM Optical Gate”, presented at Eur. Conf. Optical Communication (ECOC), Stockholm, Sweden, Sep. 5-9, 2004. Paper Tu4.4.1
[21] H. V. Demir, V. A. Sabnis, J.-F. Zheng, O. Fidaner, J. S. Harris, Jr., D. A. B. Miller, “Scalable Wavelegnth-Converting Crossbar Switches”, IEEE Photon. Technol. Lett., vol. 16, no. 10, pp. 2305-2307, 2004
[22] M. N. Sysak, J. S. Barton, L. A. Johansson, J. W. Raring, E. J. Skogen, M. L. Masanovic, D. J. Blumenthal, and L. A. Coldren, “Single-Chip Wavelength Conversion Using a Photocurrent-Driven EAM Integrated With a Widely Tunable Sampled-Grating DBR Laser”, IEEE Photon. Technol. Lett., vol. 16, no. 9, pp. 2093-2095, 2004
[23] H.-F. Chou, Y.-J. Chiu, A. Keating, J. E. Bowers, and D. J. Blumenthal, “Photocurrent-Assisted Wavelegnth (PAW) Conversion With Electrical Monitoring Capability Using a
CHAPTER 5: Wavelength Conversion
214
Traveling-Wave Electroabsorption Modulator”, IEEE Photon. Technol. Lett., vol. 16, no. 2, pp. 530-532, 2004
[24] H.-F. Chou, J. E. Bowers, and D. J. Blumenthal, “Novel photocurrent-assisted wavelength (PAW) converter using a traveling-wave electroabsorption modulator with signal monitoring and regeneration capabilities”, Optical Fiber Communication Conference (OFC), paper FD4, 2004
[25] S. Højfeldt, F. Romstad, and J. Mørk, “Absorption Recovery in Strongly Saturated Quantum-Well Electroabsorption Modulators”, IEEE Photon. Technol. Lett., vol. 15, no. 5, pp. 676-678, 2003
[26] K. Nishimura, M. Tsurusawa, and M. Usami, “Novel All-Optical 3R Regeneration Using Cross-Absorption Modulation in RF-Driven Electroabsorption Waveguide”, Eur. Conf. Optical Communication (ECOC), Paper We.F.2.4, 2001
[27] T. Miyazaki, N. Edagawa, M. Suzuki, and S. Yamamoto, “Novel Optical-Regenerator using Electroabsorption Modulators”, Optical Fiber Communication Conference (OFC), paper WM53, pp. 350-352, 1999
215
Chapter 6 3-R Regeneration 6.1 Introduction
With proper dispersion management and optical amplification, point-to-
point fiber transmission can reach sufficiently long distances even in a WDM
environment. With the use of new modulation formats such as return-to-zero
differential phase-shift keying (RZ-DPSK), the transmission distance can be
further extended without using nonlinear (soliton) transmission. For example,
the transmission of ninety-six 10-Gb/s RZ-DPSK WDM channels can reach over
13,100 km of installed submarine fibers in a recent demonstration [1]. On the
other hand, for traditional modulation formats such as return-to-zero or non-
return-to-zero amplitude-shift keying (RZ-ASK or NRZ-ASK), the transmission
reach must be improved by O-E-O regenerators, which involve conversions of
signal between the optical and the electrical domains. During these conversions,
CHAPTER 6: 3-R Regeneration
216
the signal quality can be improved so that the regenerated optical signal can be
transmitted to longer distances compared to the non-regenerated signal.
However, even though O-E-O regenerators are effective and also provide many
additional capabilities, cost and power consumption are issues for their
widespread deployment. This is particularly problematic in WDM systems since
each channel would require an independent O-E-O regenerator.
On the other hand, the rapid growth of internet traffic pushes the optical
network to evolve into more advanced architectures such as optical switching
with IP over WDM [2]. In these new architectures, the optical signal will have to
pass through various components such as cross-connect switches, wavelength
converters, WDM multiplexers and demultiplexers before reaching its final
destination. These components will add extra noise, crosstalk, and distortion to
the optical signal. In addition, the exact transmission distance would be an
unknown in these dynamically switched networks. Optical regenerators are thus
needed to support these new network architectures. The traditional O-E-O
regenerator may not be able to scale cost effectively to this purpose and new
technologies are required for optical regeneration.
The degradation of optical signal can come from several sources:
amplified spontaneous emission (ASE), residue chromatic dispersion,
polarization mode dispersion (PMD), nonlinear effects in fibers, cross-talk in
CHAPTER 6: 3-R Regeneration
217
cross-connect switches, and imperfectness of wavelength converters. They are
discussed briefly as follows.
The ASE noise comes from the optical amplifiers such as the erbium-
doped fiber amplifier (EDFA) and the semiconductor optical amplifier (SOA).
The out of band ASE noise can be suppressed by optical filtering. However, the
in-band ASE noise cannot be filtered optically and will accumulate after each
stage of amplification, which poses one of the fundamental limitations on
transmission distance. The ASE noise beats with the original optical signal and
causes amplitude fluctuations that are particularly severe on the mark level of
the signal.
Residue chromatic dispersion is due to imperfect dispersion
compensation. This is much less a problem in point-to-point transmission.
However, in an optical switching network, the chirp and path of each channel /
packet is not well-known so that exact dispersion compensation is no longer
possible. The residue dispersion is manifested by broadened pulse shape and
reduced extinction ratio.
PMD is more problematic in high bit-rate systems, such as those over 40
Gb/s per wavelength. Although there are several proposed compensation
schemes, it remains a difficult problem. Recent researches showed that it may be
possible to restore the PMD degraded signal using a 3-R regenerator instead of
complicated compensation schemes [3].
CHAPTER 6: 3-R Regeneration
218
The nonlinear effects in fiber can appear in the form of self-phase
modulation (SPM), cross-phase modulation (XPM), and four-wave mixing
(FWM), all originating from the third-order nonlinearity of the fiber. SPM is less
a problem if the optical power is controlled below a certain level. FWM can be
minimized with non-zero dispersion fibers. Intrachannel XPM is more of a
problem in WDM systems since the phase of a particular wavelength channel
can be influence by other channels, which causes timing jitter. Nonlinear effects
also distort the pulse shape in general.
Cross-talk in optical cross-connect switches will result in beating noise
and extinction ratio degradation. The use of forward-error-correction (FEC) can
mitigate this problem in part [4] but if several switching nodes need to be passed
by the signal, other measures should also be taken to ensure the signal quality.
On the other hand, wavelength converters in the reconfigurable or the future
packet switched WDM networks can also impose signal degradations in many
ways.
No matter what the origins are, all the above degradations manifested
themselves in the time-domain in terms of amplitude noise, pulse deformation,
extinction ratio degradation, and increased timing jitter. The first three are
related to the shape while the last one is related to the timing of the signal. In the
spectral-domain, the ASE noise degrades the optical signal to noise ratio
(OSNR).
CHAPTER 6: 3-R Regeneration
219
Optical regenerators can be used to combat these signal degradations.
Depending on the number of regenerative functionalities included, optical
regenerators can be classified as 1-R, 2-R, and 3-R.
The three R’s in optical regeneration represent re-amplification, re-
shaping, and re-timing. Optical amplifiers like EDFA have re-amplification
capability and can be regarded as “1-R” regenerators, which have played a
critical role in WDM transmission. An alternative candidate for 1-R regenerator
is the SOA, which offers very compact re-amplification due to its small size and
integration potential. However, the pattern dependence of an SOA originating
from the gain recovery dynamics is a major challenge for practical application.
Since EDFAs and SOAs can be used for re-amplification, re-shaping and re-
timing are the two functions that need to be addressed by new generations of 3-
R regenerator. If only re-shaping and re-amplification are implemented, the
regenerator is denoted as “2-R”.
Since the conventional O-E-O regenerators have some concerns in cost
related issues, the research on new regenerators in the past decade mainly
concentrated on “all-optical” approaches, where the signal remains in the optical
domain during regeneration. However, recent developments in integration
technology have made it possible to put together many different components on
a single optoelectronic chip. This could drastically change the economics of O-
E-O regenerators because of the cost and power reduction claimed by this
CHAPTER 6: 3-R Regeneration
220
approach [5]. The regenerators that will be presented later in this chapter using
TW-EAMs conceptually follow this trend. These regenerators are referred to as
“integrated optoelectronic regenerators”. This type and the all-optical type of
regenerators are all called “optical regenerators” in the rest of this chapter in
order to distinguish from the traditional O-E-O regenerators.
The researches on optical regenerator reported so far mainly focused on
the regeneration of ASK signals because the mechanisms available for re-
shaping only apply to the amplitude of the signal. Regeneration mechanisms for
PSK signals are yet to be explored. The following discussions will be
constrained to the regeneration of ASK signals. This chapter will focus on 3-R
regeneration of RZ optical signals. It is worth noting that re-timing of NRZ
signal using all-optical approaches is difficult. Therefore, only 2-R regenerations
of NRZ signal were reported. In the traditional O-E-O regenerator, electronic
flip-flop circuits can be used to re-time the NRZ signal but there does not seem
to be an appropriate all-optical counterpart to this purpose.
Figure 6.1 Standard optical 3-R regeneration architecture for RZ-ASK data format
CHAPTER 6: 3-R Regeneration
221
Figure 6.1 shows the standard architecture of optical 3-R regeneration
for RZ signal. There can be variations to this architecture but the basic ideas are
the same. First, the degraded optical input signal is split into two parts in the
regenerator. On part is used to recover a synchronized electrical clock using
electronic clock recovery. The recovered clock should have a lower jitter
compared to that of the degraded input signal. The detailed requirements for
clock recovery have been discussed in Section 4.1. The recovered electrical
clock then drives an optical pulse source to generate a synchronized optical
pulse train. The pulse shape is well-defined and lateral re-shaping is obtained.
The regenerated pulse train is subsequently modulated by a nonlinear
optical gate that is controlled by the other branch of the input signal. The optical
gate should have a step-like transfer function so that digital decision can be
obtained and the noise on the “1” (mark) and the “0” (space) can be eliminated.
This provides the capability of vertical re-shaping. For RZ signal, it is necessary
that the width of the gating window is wider than that of the regenerated pulses.
In this way, the well-defined shape of the regenerated pulse train can be
preserved and the timing jitter of the input signal can be reduced after
regeneration (since the regenerated pulses do not sense the jitter in the original
signal if the gate is wide enough). This would impose restrictions on the
pulsewidths of the input and the output signals. When the input pulse is very
short compared to the bit slot, some measures must be taken to broaden the
CHAPTER 6: 3-R Regeneration
222
pulsewidth in order to obtain effective re-timing capability. On the contrary, for
NRZ signal regeneration (2-R), it would be advantageous that the nonlinear gate
has an impulse response as short as possible so that the pulse shape is not
broadened after regeneration. This is clearly demonstrated with numerical
simulation in Section 5.3.2.
The re-shaped and re-timed signal is then re-amplified by an optical
amplifier to a desired output power level. An optical amplifier can also be place
at the input of the regenerator so that the required input power level can be
lowered and the dynamic range increased.
The purpose of the optical regenerator is to condition the optical signal
so that after regeneration, the optical signal can be transmitted to a longer
distance. Even though the eye diagram of the optical signal looks cleaner and
well defined after regeneration, the errors that have occurred cannot be corrected
by the regenerator. Only FEC has the capability of correcting errors since
redundant information is contained in the signal. The regenerator only slows the
degradation of the optical signal. It is demonstrated numerically in Ref. [6] that
regenerators can only reduce bit-error-rate (BER) degradation but cannot
improve it, unless the regenerator can distinguish between the “1” bit and the
“0” bit. Put it in another way, the nonlinear gate cannot remove the amplitude
noise that passes over the decision threshold. The receiver sensitivity
improvement (usually in the form of negative power penalty) obtained in many
CHAPTER 6: 3-R Regeneration
223
reported experiments is generally obtained by transmitting the regenerated
signal through additional fiber links or by using a pre-amplified receiver with an
EDFA as the gain media. In both cases, a well regenerated signal may degrade
slower than the original (degraded) input signal so that a better BER can be
measured. One exception is the improvement in extinction ratio, which can be
reflected in receiver sensitivity improvement right after regeneration.
The position of regenerators in an optical transmission link should be
carefully designed so that optimal BER performance can be obtained with a
given number of regenerators [7] but this is more difficult to implement in a
dynamically reconfigured network.
The shape of the input to output transfer function of the nonlinear optical
gate is critical to the amplitude noise reduction (vertical re-shaping) capability.
The implemented optical gates usually do not possess a sharp step-like transfer
function as shown in Figure 6.1. The smoother the transfer function is, the
weaker the regeneration capability [8]. Most interferometer-based nonlinear
gates can only provide a sinusoidal transfer function and have a weaker vertical
reshaping strength. In many cases, the bias point of the nonlinear gate can be
adjusted in order to suppress more noise on the mark or on the space, depending
on the degradation of the input signal.
The reported optical regenerators mainly differ in the implementation of
the nonlinear gate. Most wavelength converters and optical demultiplexers can
CHAPTER 6: 3-R Regeneration
224
serve as the nonlinear decision gate so long as they possess the desired transfer
function. In fact, when the wavelength of the pulse train in Figure 6.1 is
different from that of the input signal, a regenerative wavelength converter is
achieved naturally. However, the problem with using an all-optical wavelength
converter as the nonlinear gate is that regeneration at the same wavelength is not
possible as it is. Two stages of wavelength converters are thus required to bring
the output signal back to the original wavelength, which complicates the
configuration. Optoelectronic regenerators have an intrinsic advantage on this
regard as will be discussed in more detail in Section 6.3.
Several demonstrated 3-R approaches are reviewed in Ref. [9], most of
them can operate up to 40-Gb/s or beyond. The reported nonlinear gates for 3-R
regeneration include SOA with Mach-Zenhder Interferometer (MZI) [10],
cascaded electroabsorption modulators (EAMs) [11], SOA with delayed-
interferometer [12], ultrafast nonlinear interferometer (UNI) [13], and SPM in
highly nonlinear (HNL) fiber [14]. These 3-R regenerators were all tested in re-
circulating loops. Using a Kerr-switch based on HNL fiber, effective RZ signal
regeneration has been demonstrated at 160 Gb/s [15].
By implementing the functionalities shown in Figure 6.1 individually,
effective 3R regenerations can be realized but many components are required.
Incorporating some of these functionalities together with fewer components
would be advantageous and critical for practical deployment. For example, an
CHAPTER 6: 3-R Regeneration
225
EAM can work both as a phase comparator for clock recovery and a non-linear
decision gate [16]. A self-pulsating laser can recover an optical sinusoidal pulse
train, combining the clock recovery and an optical pulse source together [17].
Three optical regenerators based on TW-EAMs are proposed and studied
in this chapter. In section 6.2, the compact 3-R PAW-Regeneration that
combines the RF-driven PAW-Conversion (Section 5.4) and the injection-
locking clock recovery technique (Section 4.4) is presented. This is the first
optical 3-R regenerator that can achieve clock recovery, pulse generation, and
nonlinear decision with a single optoelectronic device. Successful regeneration
of a degraded optical signal is demonstrated and the performance limitations are
studied in great deal with numerical simulations.
For an optical 3-R regenerator with an even stronger regeneration
capability, the nonlinear E-O transfer function of the EAM is utilized to enable a
simplified optoelectronic 3-R regenerator in Section 6.3. This new regenerator
has all the good features of the conventional O-E-O regenerator but with fewer
electronic components. The standard architecture is followed as it is and very
effective regeneration is demonstrated experimentally. Numerical simulations
are also presented to study the cascaded performance and the regeneration
capabilities regarding timing and amplitude noises.
After demonstrating the effectiveness of the simplified optoelectronic 3-
R regenerator in Section 6.3, an integrated version of it is presented in Section
CHAPTER 6: 3-R Regeneration
226
6.4. By using InP-based integration technology a tunable laser is integrated with
a SOA and two EAMs on a single chip in order to replace the transmitter part of
the optoelectronic regenerator. Great saving in space and coupling loss is
achieved with integration. Regeneration of ASE noise degraded signal is
demonstrated. This integrated version of simplified optoelectronic 3-R is very
promising and have the potential to scale cost effectively for practical use in
WDM networks as a generative wavelength converter or a 3-R regenerator.
Many of these 3-R regeneration results presented in this chapter are
published in Ref. [18]-[20].
6.2 Compact 3-R PAW-Regeneration
In Section 5.4, RF-driven PAW-Conversion is introduced to improve the
performance of wavelength conversion for RZ data format, where a
synchronized electrical clock is fed into the TW-EAM to remove the long falling
tail in the converted signal. In addition to this lateral re-shaping capability, it
was demonstrated that re-timing of a jittered signal can be achieved after the
wavelength conversion. Therefore, RF-driven PAW-Conversion is also called
PAW-Regeneration for its regenerative capabilities.
The synchronized electrical clock required by PAW-Regeneration can be
provided externally by a clock recovery unit. However, in order to obtain an
CHAPTER 6: 3-R Regeneration
227
even more compact 3-R regenerator, the injection locking clock recovery
technique introduced in Section 4.4 is merged with PAW-Regeneration in this
section. In the new configuration, three essential functionalities (re-shaping, re-
timing, and clock recovery) can be realized with a single TW-EAM
simultaneously. In Section 6.2.2, the successful operation of this concept is
demonstrated and verified at 10 Gb/s with negative power penalty. The
limitations of this approach are studied with numerical simulations in Section
6.2.3.
6.2.1 Concept
The configuration of the proposed compact 3-R PAW-Regeneration is
shown in Figure 6.2. It resembles closely the clock recovery setup using an
injection-locked ring oscillator introduced in Section 4.4 (Figure 4.24(c)). Both
electrical and optical clocks can be obtained in the original injection locking
clock recovery setup since the oscillating electrical clock modulates the TW-
EAM as an optical pulse generator. The input power required for injection-
locking is not very high. For example, it is demonstrated in Section 4.4.2 that an
implemented setup can acquire locking even at an input power level of –10
dBm. If the input power increases, the locking range can also be increased,
giving more tolerance to environmental fluctuations and frequency uncertainties.
CHAPTER 6: 3-R Regeneration
228
Figure 6.2 Configuration of the compact 3-R PAW-Regeneration. BW: bandwidth, LPF: low-pass filter, BPF: band-pass filter, amp: amplifier
As the input power increases beyond some point (typically over +10
dBm), the photocurrent signal can be strong enough to enable PAW-Conversion.
In fact, this PAW-Conversion is RF-driven since an oscillating electrical clock
runs through the TW-EAM and adds with the photocurrent signal. As a result,
RF-driven PAW-Conversion is achieved in a self-sustained fashion with clock
recovery embedded in the same setup. The result is a highly compact 3-R
regenerator using only a single TW-EAM. The operating principles of the RF-
driven PAW-Conversion and the injection locking clock recovery are described
in detail in Section 5.4 and Section 4.4, respectively. The main concern here is if
they can co-exist and work together properly as a compact optical regenerator.
Referring to Figure 6.2, PAW-Conversion happens along the active
waveguide inside the TW-EAM, where the electrical and the optical waves
CHAPTER 6: 3-R Regeneration
229
interact with each other through photodetection and electroabsorption. The
necessary condition for PAW-Conversion to take place is that the photocurrent
signal must be strong enough to modulate the CW probe with sufficient
extinction. Other than that, the setup complies very well with that of the
injection locking clock recovery using a TW-EAM-based ring oscillator, as
shown in Figure 4.24(c). Therefore, the two functions (PAW-Regeneration and
clock recovery) should be able to co-exist.
Nevertheless, at least three issues need to be addressed so that the two
functions can work together properly. First, the photocurrent signals coming out
of the TW-EAM (co-propagating and counter-propagating tributaries) must be
terminated properly outside the device to prevent multiple reflections which can
cause jitter and intersymbol interference. In the setup of injection locking clock
recovery shown in Figure 4.24(c), the electrical band-pass filter (BPF) passes
only the clock tone of the counter-propagating photocurrent signal and rejects
(reflects) all other spectral components (including those containing data). This
may not be a problem when the photocurrent signal is weak under low injection,
which is sufficient for clock recovery. However, for PAW-Regeneration, the
photocurrent signal is much stronger and the reflections would cause serious
signal degradation.
To solve this problem, an electrical amplifier (Amp2) is added between
the band-pass filter and the TW-EAM as shown in Figure 6.2. Electrical
CHAPTER 6: 3-R Regeneration
230
amplifiers generally have an output return loss (S22) better than 10 dB and can
serve as a termination for the photocurrent signal. The same principle applies to
the co-propagating tributary of the photocurrent signal, which is terminated at
the input of amp1. Note that the co-propagating tributary is taped to provide
signal monitoring capability. Care must also be taken to minimize any back
reflection from the tap. The 0.75-bandwidth low-pass filter (LPF) is used to
reject the strong oscillating clock and pass most of the data components in the
photocurrent signal so that a nearly NRZ electrical signal can be obtained after
the filter. The clock tone reflected by the LPF dose not impose a serious
problem since it only changes the effective phase of the ring oscillator seen at
the clock frequency and does not cause jitter or intersymbol interference.
However, there can be residue reflections of data components from the LPF.
Using an absorptive LPF or inserting an electrical amplifier as a buffer can
minimize residue reflections.
Another issue arises from adding Amp2 as a termination. At steady-state,
at least one of the amplifiers in the ring oscillator must saturate with its gain
reduced so that the net loop gain is decreased to unity. However, the output of a
saturated amplifier is very likely distorted and contains harmonic frequencies,
which would result in corresponding distortions in the regenerated optical
signal. This is not a problem in the clock recovery case since a BPF is present
after the amplifier. Unfortunately, the BPF is intentionally separated from the
CHAPTER 6: 3-R Regeneration
231
TW-EAM by Amp2 for the reason stated in the previous paragraph. It is not
possible to add another BPF after Amp2 or the problem would become recursive
(reflection would occur again). To solve this problem, it is arranged so that only
amp1 saturates at steady-state in order to bring the net loop gain down to unity
while keeping Amp2 unsaturated to avoid distortion.
The third issue is about the relative phase of the recovered electrical
clock and the photocurrent signal. It is necessary to align the peak of the applied
sinusoidal clock with the peak of the photocurrent signal to obtain the maximal
eye opening of the converted signal, as can be easily understood from Figure
5.25(b). In the original RF-driven PAW-Conversion, the phase of the applied
electrical clock can be adjusted manually to this purpose before feeding into the
TW-EAM. However, in the case of compact PAW-Regeneration, the phase
difference seems to be decided by the regenerator itself since it is self-sustained.
It is not obvious if the phase difference is optimal. This is a profound issue and
will be explored in more detail in Section 6.2.3. At this point it is assumed that
this is not a major problem and the phase difference is appropriate.
6.2.2 10-Gb/s RZ signal regeneration
The proposed concept is demonstrated at 10 Gb/s with RZ signals. The
actual experimental setup is shown in Figure 6.3. The electrical ring oscillator
has approximately 4.7 dB of small-signal gain (before saturation takes place).
CHAPTER 6: 3-R Regeneration
232
As discussed in the previous section, at steady-state it is important that only
Amp1 saturates, preventing waveform distortion due to saturation of Amp2,
which drives the TW-EAM. The 10-GHz input power versus output power
relationship of the two amplifiers is plotted in Figure 6.4. The loss and gain of
the ring is designed so that Amp2 gives about 25 dBm output power with very
little saturation while Amp1 is highly saturated with an input power level around
10 dBm.
Figure 6.3 Experimental setup of compact 3-R PAW-Regeneration for 10-Gb/s RZ signal
In order to maintain signal integrity, the reflection of photocurrent signal
from outside electronics is kept lower than – 20 dB up to 10 GHz by using
additional attenuators. The electrical BPF has a 100-MHz bandwidth, centered at
9.953 GHz. The loop of the oscillator is approximately 0.75 m long and the
corresponding free spectral range is 533 MHz. As a result, only one resonant
CHAPTER 6: 3-R Regeneration
233
mode is picked by the BPF. An electrical phase shifter (tunable delay) is inserted
in the loop to tune the resonance frequency. The RZ input signal is a 15-dBm,
20-ps, 231-1 pseudorandom binary sequence (PRBS) data at 1550.92 nm. The
CW is 4-dBm at 1557.50 nm. EDFA is used for re-amplification.
Figure 6.4 Gain and output power as function of input power for (a) Amp1; (b) Amp2
First, the regenerator is evaluated with a high-quality input signal in
order to access the back-to-back performance. The extinction ratio of the high
quality input signal is over 15 dB. The lock-in range is 1.15 MHz. The free-
running frequency of the ring oscillator can be changed by adjusting the
electrical phase shifter, which changes the effective length of the ring oscillator.
It is observed that, within the lock-in range, the relative phase of the oscillating
clock and the photocurrent signal can also be changed by adjusting the electrical
phase shifter. More detailed discussions of this effect will be given in the next
CHAPTER 6: 3-R Regeneration
234
section. Experimentally, the relative phase is adjusted so that the eye opening of
the regenerated optical signal is optimal, which gives the best BER performance.
Figure 6.5 Scope traces of (a) the input signal; (b) the photocurrent signal without the recovered clock; (c) the photocurrent signal with the recovered clock; (d) the output optical signal at
1557.50nm. These traces are measured at two bias voltages, - 2 V and - 3 V.
The bias voltage of the TW-EAM is critical to the regeneration
performance. Figure 6.5 shows traces of the input and output signals as well as
the detected photocurrent signals with and without the oscillating clock. The
input signal is well defined with little fluctuations on the “1” (mark) and the “0”
(space) levels. However, the detected photocurrent signal is degraded with long
falling tails and pattern dependence caused by a limited detection bandwidth.
Fortunately, by adding the recovered electrical clock (RF-driven), the long
CHAPTER 6: 3-R Regeneration
235
falling tails are removed in the converted signal, which is made possible through
the highly nonlinear E-O transformation and the high extinction of the TW-
EAM.
On the other hand, the pattern dependence in the photocurrent signal is
more difficult to get rid of. By setting a small reverse bias such as – 2 V in this
case, the fluctuations at the mark level can be clamped by taking advantage of
the flat top of the E-O transfer function, as already discussed in detail in Section
5.2.4. The left of Figure 6.5(d) clearly shows the clamping effect on the mark
level in the output signal. However, this is achieved by sacrificing the extinction
ratio of the output signal since the amplitude (eye opening) of the photocurrent
signal is finite. If a larger portion of the eye opening is used for clamping, the
portion left for producing the extinction is reduced. Therefore, it can be inferred
that a stronger input power (larger photocurrent signal) will lead to a better
output signal quality and also better vertical re-shaping capability. Surely, this is
limited by the optical gain available and the power handling capability of the
TW-EAM. As a result, the bias voltage of the TW-EAM is chosen to be – 3.0 V,
which balances the clamping effect and extinction at this input power level for
best BER performance (right of Figure 6.5(d)). The strength of the oscillating
electrical clock at the input of the TW-EAM is estimated to be 5.4 Vpp.
The eye diagrams with a high-quality input signal are shown in the left
column of Figure 6.6. The converted eye (Figure 6.6(b)) is clean and has a
CHAPTER 6: 3-R Regeneration
236
smaller width of 15 ps, which is an indication of lateral re-shaping. The
pulsewidth of the regenerated signal can be changed by adjusting the strength of
the oscillating clock to the TW-EAM. The extinction ratio of the converted
signal is higher than 10 dB. BER results in Figure 6.7 show that only 0.3 dB of
power penalty is imposed, indicating good performance as a self-sustained RZ
wavelength converter.
Figure 6.6 Eye diagrams of (a), (e) input signal; (b), (f) wavelength converted signal; (c), (g) monitored electrical signal without LPF; (d), (h) with LPF; for the cases of high-quality input signal (left column) and degraded input signal (right column). The arrows indicate zero levels.
The RMS timing jitter is measured by integrating the single-side-band
(SSB) noise from 1 kHz to 10 MHz with a 100 Hz resolution bandwidth. The
jitter of the input optical signal, the recovered electrical clock and the converted
CHAPTER 6: 3-R Regeneration
237
optical signal are 727 fs, 349 fs, and 590 fs, respectively, showing re-timing
capability of the regenerator.
Figure 6.7 Bit-error-rate (BER) results of the high-quality and the degraded input cases
Figure 6.6(c) shows the monitored electrical signal without LPF filtering,
which is a combination of the photocurrent signal and the oscillating clock. With
a 7.5 GHz LPF, the clock tone is suppressed and a nearly non-return-to-zero
signal is obtained as in Figure 6.6(d), which gives error-free BER measurement
and provides electrical signal monitoring.
Next, the input signal is intentionally degraded (Figure 6.6(e)) in order to
access the regeneration capability. The timing jitter of the input signal is
increased by applying a 2 MHz frequency modulation to the pulse source in the
transmitter with a 0.15 modulation index. The extinction ratio is decreased to
about 6 dB by lowering the driving voltage to the transmitter and a change in
modulator bias. Other parameters such as the input power and the TW-EAM
CHAPTER 6: 3-R Regeneration
238
bias are kept the same. After regeneration, the converted eye (Figure 6.6(f))
shows a reduced jitter, an increased extinction ratio, and a shortened pulse
width. A 1.0 dB negative power penalty is measured when compared to the
degraded signal, as shown in Figure 6.7. Although the eye opening, ∆V, in
Figure 6.6(g) is smaller than that in the previous case (Figure 6.6(c)) due to the
degraded extinction ratio of the input signal but after E-O transformation the
extinction ratio can still be improved to over 10 dB because of the high
modulation efficiency of TW-EAM. The jitter of the input optical signal, the
recovered electrical clock and the converted optical signal are 2.9 ps, 1.2 ps, and
1.5 ps, respectively, showing up to 50% timing jitter reduction. This is in good
agreement with the jitter characterization presented in Section 5.4.3 for the RF-
driven PAW-Conversion. These results verify successful operation of the
compact 3R PAW-Regeneration, which is capable of improving the signal
quality through wavelength conversion.
The required input power may depend on several factors such as the
degree of degradation, the response and the transfer function of the TW-EAM.
Hhigh modulation efficiency, expressed in dB/V, is needed to keep the required
∆V low, which implies a lower input power. From Figure 6.6(c), it is estimated
that ∆V is 0.4 V inside the TW-EAM, taking into account that the impedance of
the active waveguide is only 25 Ohm. This corresponds to 6 dB of extinction
ratio with 15-dB/V modulation efficiency. However, the measured extinction
CHAPTER 6: 3-R Regeneration
239
ratio of the converted signal is over 10 dB, indicating that the saturation
mechanism works in favor of PAW-Regeneration in the experiment. Although
the saturation mechanism seems to have a positive effect on extinction ratio, it
should be reduced as much as possible form the point of view of the
photocurrent-assisted mechanism and in terms of speed. The more the device
saturates, the slower it is as a photodetector, as discussed in Section 5.2.3.
6.2.3 Performance limitations
In this section, the performance limitations of the compact 3-R PAW-
Regeneration are studied, which can be classified as either technical or intrinsic.
The former is related to the implementation of the regenerator and the latter
comes from the regeneration mechanism itself, which is more difficult to
overcome.
Technical limitations — many technical limitations of the compact 3-R
PAW-Regeneration originate from a limited TW-EAM detection bandwidth.
The 3-dB O-E response bandwidth of the TW-EAM used in the previous
experiment is between 8 to 12 GHz, as shown in Figure 4.4 under small-signal
measurement. At higher input powers, the bandwidth can be further decreased
because of saturation. As shown in Section 5.2.3, the bandwidth is not large
enough to handle 10-Gb/s RZ PAW-Conversion without the aid of the RF-
CHAPTER 6: 3-R Regeneration
240
driven technique. This is mainly due to the fact that the bandwidth of the RZ
pulses is generally much larger than the bit-rate.
Figure 6.8 shows the scope trace of the detected RZ photocurrent signal
at – 3V of bias and 15 dBm of input power. In addition to the long falling tails,
which can be counter-acted by the RF-driven approach, a limited bandwidth can
cause pattern-dependent fluctuations on the mark and the space, leading to a
reduced ∆V. It can be observed from Figure 6.8 that ∆V, which generates the
eye opening in the converted signal, is less than 60% of the peak to peak
amplitude of the photocurrent signal. This means that a lot of input power is
wasted since 40% of the photocurrent signal is not useful to open the eye of the
converted signal. Even worse, the fluctuations on the mark and the space can be
transferred to the output and result in significant degradation of the converted
signal. This happens even with a high-quality input signal because these
fluctuations are caused by the regenerator itself. This is not desired since a
regenerator is supposed to improve the signal quality.
Figure 6.8 Bit pattern of a 15-dBm, 27-1 PRBS, high-quality 10-Gb/s RZ input signal detected by the TW-EAM at – 3.0 V bias. (200 ps/div)
CHAPTER 6: 3-R Regeneration
241
The nonlinear E-O transfer function of the TW-EAM can help to
minimize these adverse effects. For example, Figure 6.5 shows that if a smaller
bias is acceptable (while maintaining an adequate extinction ratio in the
converted signal), the flat-top of the E-O transfer function can help to clamp the
mark levels. The space level is easier to flatten out because only “high
extinction” but not “flat transmission” is required. This is due to the fact that the
difference in signal quality is not significant as long as the extinction ratio is
over 10 dB.
However, all these re-shaping capabilities provided by the nonlinear E-O
transfer function of the TW-EAM should be applied to combat the degradations
in the input signal but not wasted on compensating the problems caused by the
regenerator itself. Therefore, it is crucial that the detection bandwidth of the
TW-EAM is sufficient for the operating speed of interest. Many of the
approaches that can improve the speed of the TW-EAM as a photodetector are
discuss in Section 5.2.3 and 5.2.4, which include the use of shallow QWs in the
active material.
The next performance limitation is subtle and is implicitly caused by a
limited detection bandwidth. It arises from the simultaneous operation of PAW-
Regeneration and the injection locking clock recovery within the same setup. As
demonstrated in Section 4.4, the injection locking clock recovery is capable of
generating a pure and low-jitter clock signal. The noise in the input signal above
CHAPTER 6: 3-R Regeneration
242
the lock-in range is not transferred to the recovered clock. Figure 6.9 shows the
SSB spectrums of the input optical signal, the recovered electrical clock and the
output optical signal taken during a regeneration experiment similar to the ones
presented in the previous section. The calculated RMS jitters for the three
signals are 358 fs, 235 fs, and 312 fs, respectively. Compared to Figure 4.27(a),
which represents a typical case of injection locking clock recovery, there is a
significant increase of noise in Figure 6.9 above 100 kHz from the expected
behavior (dashed gray line) for the recovered electrical clock and the output
optical signal. Even though the calculated RMS jitters of these two signals are
still better than that of the input signal in this case, the deviation from the typical
behavior is worthy of exploration. When the regenerators are cascaded several
times, this problem can become more serious.
Figure 6.9 SSB spectrums of the input signal, the recovered electrical clock, and the output optical signal. The dashed line is the expected spectrum of the recovered electrical clock.
CHAPTER 6: 3-R Regeneration
243
As a matter of fact, this phenomenon is also caused by the limited
detection bandwidth of the TW-EAM. To explain for this, the distorted
photocurrent signal is modeled as a train of asymmetric triangular pulses as
shown in Figure 6.10(b). The symmetry parameter, ε, is used to describe the
degree of symmetry of the triangular pulses. The pulses are symmetric when
ε equals unity.
Figure 6.10 (a) A typical photocurrent pattern detected by the TW-EAM used in the regeneration experiment; (b) the shape of the modeled photocurrent signal
The clock tone that can be used to injection-lock the ring oscillator is the
fundamental frequency of the pulse train. By applying Fourier series expansion,
the fundamental component can be expressed as:
( )θ−+ twba ocos21
21
CHAPTER 6: 3-R Regeneration
244
where
( )
+−= 2)cos(111
21 επεπ
a , 21
)sin(επ
επ=b , and ( )
( )
−+= −
επεεπθcos21
sintan 1
Note that the parameters depend only on ε but not on frequency. It is θ
that determines the phase shift of the clock tone in the photocurrent signal. The
dependence of θ on ε is plotted in Figure 6.11, where the phase shift is
converted to timing shift assuming that the frequency is 10 GHz.
Figure 6.11 Timing shift of the clock tone in the photocurrent signal as function of the
symmetry parameter. The frequency is assumed to be 10 GHz.
Taking ε to be 0.5, which is a fair value for the experimentally observed
photocurrent signal, the corresponding timing shift is 7.4 ps. The relationship of
the modeled photocurrent signal and its fundamental clock tone at this condition
is shown in Figure 6.12. It can be seen from this figure that the timing shift
between the peak of the photocurrent signal and the peak of the fundamental
CHAPTER 6: 3-R Regeneration
245
clock tone is 7.4 ps. Also shown in the figure (dashed gray curve) is the optimal
timing of the applied clock for RF-driven PAW-Conversion. To obtain the
maximal eye opening in the converted signal, it is necessary that the peak of the
applied clock coincides with the peak of the photocurrent signal. This would
also lead to a wavelength converted signal with the lowest possible power
penalty at a given input power level.
Figure 6.12 Plots of the modeled asymmetric photocurrent signal (solid black), the fundamental
Fourier component of the photocurrent signal (solid gray), and the ideal timing of the applied electrical clock for maximal eye opening of the converted signal (dashed gray)
However, from equation 4.17 in Section 4.4.2, the phase difference
between the recovered clock and the injected signal (photocurrent signal)
satisfies the following relationship:
i
o
o
uning
AA
Qf
fdet)sin(∆
−=∆θ
CHAPTER 6: 3-R Regeneration
246
When there is no detuning (∆fdetuning=0), the phase difference, ∆θ, equals
zero. This means that the recovered clock is in phase with the clock tone in the
photocurrent signal, which has 7.4 ps of timing shift with respect to the peak of
the photocurrent signal. This would lead to a reduced eye opening in the
wavelength converted signal and consequently an increased power penalty (or a
waste of input power since the full useful ∆V is not utilized).
Figure 6.13 Timing shift between the recovered clock and the injected signal as function of
detuning for the injection locking clock recovery (re-plotted from Figure 4.32)
To obtain a recovered clock with – 7.4 ps of timing shift relative to the
clock tone in the photocurrent signal, it is necessary to impose detuning of the
ring oscillator. This can be done by adjusting the tunable delay (phase shifter)
shown in Figure 6.3. When the effective length of the ring oscillator is changed,
the frequency of the oscillating mode is shifted accordingly. As an example, the
curve shown in Figure 4.32 for an injection-locked ring oscillator with 4-dBm of
injection power is re-plotted in Figure 6.13. In order to obtain – 7.4 ps of timing
CHAPTER 6: 3-R Regeneration
247
`shift, 140 kHz of detuning is necessary. Unfortunately, as pointed out in
Section 4.4.2, when there is detuning, the jitter of the recovered clock can be
increased, which is experimentally confirmed in Figure 4.30. In the frequency
domain, this corresponds to an increase of SSB noise. For the same injection-
locked ring oscillator with 4-dBm of injection, the SSB spectrum with 150 kHz
of detuning is plotted in Figure 6.14. It can be easily observed that the SSB
noise is increased significantly beyond 100 kHz with detuning, which resembles
the behavior shown in Figure 6.9. This explains qualitatively the behavior
observed in Figure 6.9.
Figure 6.14 SSB spectrums of an injection-locked ring oscillator with 4-dBm injection power
From the above arguments, it can be summarized that the limited
detection bandwidth of the TW-EAM leads to an asymmetric photocurrent
signal, which has a timing shift between the peak of the signal and its clock
CHAPTER 6: 3-R Regeneration
248
tone. To optimize the eye opening of the converted signal, detuning of the ring
oscillator is necessary to shift the timing of the recovered clock so that it
coincides with the peak of the photocurrent signal. However, this increases the
SSB noise of the recovered clock as a side effect. This problem can be
minimized if the detection bandwidth of the TW-EAM can be improved and a
more symmetric RZ photocurrent signal is obtained. Fortunately, the increase of
jitter with respect to detuning is not rapid unless the detuning is large and close
to the edge of lock-in range, as indicated in Figure 4.30. This problem only
occurs when the injection locking clock recovery and PAW-Regeneration are
merged in one setup. If they are separated, this problem would not exist since
the phase of the applied clock can be adjusted independently with a delay line to
optimize the performance.
Intrinsic limitations — Intrinsic limitations on the performance of
compact 3-R PAW-Regeneration come from the regeneration mechanism itself.
In PAW-Regenerator, the optical pulse source and the nonlinear decision gate
are merged together to achieve a very compact architecture. In principle, it can
be a high-performance wavelength converter if the detection bandwidth is
appropriate for the bit-rate of interest. For example, power penalty as low as 0.3
dB at 10-Gb/s is demonstrated in Figure 6.7. However, the merging of the two
functions can limit the regeneration capability when the input signal is seriously
degraded. In the rest of this section, numerical simulations similar to the one
CHAPTER 6: 3-R Regeneration
249
presented in Section 5.3 are implemented to study these intrinsic limitations.
The sequence of simulation is shown in Figure 6.15. The major difference from
the PAW-Conversion case in Figure 5.18 is that a synchronized sinusoidal clock
is added to the photocurrent signal before modulating the TW-EAM.
Figure 6.15 Sequence of simulation for PAW-Regeneration
Figure 6.16 E-O transfer function of the TW-EAM used in PAW-Regeneration simulations
The E-O transfer function used in the simulation is plotted in Figure
6.16, which is artificially assigned and has flat transmission regions for effective
CHAPTER 6: 3-R Regeneration
250
clamping of fluctuations on the mark and the space levels. The purpose of using
an ideal curve is to explore the limits of regeneration capabilities.
First, PAW-Conversion without the RF-drive technique is simulated.
Relevant experimental results are shown in Figure 5.27. The detection
bandwidth is assumed to be 6.6 GHz, which is the best case achieved
experimentally in Chapter 5. The simulated eye diagrams are shown in Figure
6.17. The FWHM of the input signal is 15 ps. Instead of assigning the input
power, the peak to peak amplitude of the photocurrent signal is assigned, which
is 1.8 V in this case.
Due to the limited bandwidth, the photocurrent signal is distorted with
long falling tails, shown in Figure 6.17(b). Fortunately, the fluctuation on the
mark is not serious at this bandwidth. The wavelength converted output signal at
– 3.1 V of bias voltage is shown in Figure 6.17(c). This is approximately the
smallest bias that gives an adequate extinction ratio ( > 10 dB). It is clear that
the long falling tails are transferred from the photocurrent signal to the output
signal, even though the tails are cut shorter by the nonlinear E-O transfer
function. Increasing the reverse bias can further shorten the falling tails but at
the expense of increased loss (Figure 6.17(d)). To reduce the loss, the
photocurrent amplitude must be increased but this is constrained by the input
power available and the power handling capability of the TW-EAM.
CHAPTER 6: 3-R Regeneration
251
Figure 6.17 Simulated eye diagrams of RZ PAW-Conversion without RF-drive (a) input optical signal; (b) driving signal to the TW-EAM with 1.8 V peak to peak amplitude; (c) output signal at
Vb = - 3.1 V; (d) output signal at Vb = - 3.9 V.
Figure 6.18 Simulated eye diagrams of PAW-Regeneration (a) driving signal to the TW-EAM. The eye height is 1.8 V and the amplitude of the clock is 5 V; (b) output signal at Vb = - 5.6 V.
With the aid of the RF-driven technique, the converted signal is well-
shaped without the long falling tails, as shown in Figure 6.18(b). In this
simulation, the amplitude of the photocurrent signal is still 1.8 V and the applied
sinusoidal clock is 5 Vpp (Figure 6.18(a)). The peak of the composite driving
CHAPTER 6: 3-R Regeneration
252
signal can reach the low loss portion of the E-O transfer function in order to
obtain the maximal output power. These simulation results complement the
experimental results presented in Section 5.4.2.
The previous results show that PAW-Regeneration can be a high
performance wavelength converter when the input signal is clean. The following
simulations will explore its regenerative capabilities with respect to timing and
amplitude jitters in the input signal.
Figure 6.19 Simulated eye diagrams with timing jitter degraded input signal (a) input signal
with random 10% UI peak to peak timing jitter; (b) driving signal to the TW-EAM; (c) output signal at Vb = - 5.6 V
First, regeneration of timing jitter degraded signal is simulated, where
10% unit-interval (UI) random timing jitter is imposed on the input signal, as
shown in Figure 6.19(a). Because of the timing jitter, the trace of the eye
diagram is broadened. The photocurrent signal also carries the timing jitter but
CHAPTER 6: 3-R Regeneration
253
the tailing edge is “thinned” because the relatively slow detection of the TW-
EAM “stretches out” the jitter. The applied electrical clock pulls down both the
rising and the falling edges and the timing jitter is further reduced by this
processes. The converted signal is shown in Figure 6.19(c), where the timing
jitter is reduced when compared to the input signal. The reduction of timing
jitter is better on the falling edge and the eye becomes a little bit asymmetric.
This is also observed experimentally as shown in Figure 6.20, where the input
signal is FM-modulated at 1 MHz with a modulation index of 0.6. The measured
results in Figure 5.31 showed that 50% reduction of timing jitter reduction can
be obtained, which is supported by the simulation results in Figure 6.19. These
results indicate that PAW-Regeneration does have timing jitter reduction
capability but the strength is only moderate (~50%).
Figure 6.20 Experimentally measured eye diagrams (a) optical input signal with intentional
1-MHz FM-modulation and a modulation index of 0.5. (b) converted optical signal
Next, regeneration of amplitude jitter degraded signal is simulated. The
amplitude jitter can be random (caused by ASE nose) or deterministic (caused
CHAPTER 6: 3-R Regeneration
254
by dispersion or pattern dependence). In the simulation, 20% peak to peak
random amplitude jitter is imposed on the input signal (Figure 6.21(a)). The bias
voltage is set at – 5.6 V, which is suitable for converting an un-degraded input
signal just like the case in Figure 6.18. The amplitude jitter is not suppressed in
the converted signal (Figure 6.21(c)). Instead, it is increased to about 30%. This
is because the peak of the driving signal falls in a region of the E-O transfer
function where the slope is higher than unity.
Figure 6.21 Simulated eye diagrams with amplitude jitter degraded input signal (a) input signal with 20% peak to peak amplitude jitter; (b) driving signal to the TW-EAM; (c) output signal at
Vb = - 5.6 V; (d) output signal at Vb = - 5.2 V
Reducing the bias voltage can shift the driving signal to the flat
transmission region and the output can be clamped. Figure 6.21(d) shows the
output signal at – 5.2 V of bias, where the amplitude jitter is reduced. However,
CHAPTER 6: 3-R Regeneration
255
there are two side effects for reducing the reverse bias. First, the extinction ratio
of the converted signal can be reduced. The actual degree of reduction would
depend on the amplitude of the photocurrent signal (or more precisely the eye
opening ∆V) and the shape of the E-O transfer function. Secondly, the shape of
the converted RZ pulses is deformed because of the clamping effect. In contrast,
this is not a problem at all for NRZ signal since its mark level is supposed to be
flat, as demonstrated in Section 5.2.4.
In summary, the intrinsic limitations originating from the mechanism of
PAW-Regeneration impose a strong restriction on the regeneration of amplitude
jitter (or amplitude noise) degraded signal. By utilizing the flat transmission
region of the E-O transfer function, the amplitude fluctuations can be reduced
but the side effect would be pulse deformation in the regenerated signal. In
addition, ∆V of the photocurrent signal must be large enough to support an
adequate extinction ratio under this situation. Replacing the CW probe by a
regenerated pulse train can remove the pulse deformation problem but the
simplicity advantage of PAW-Regeneration is somewhat reduced. Monolithic
integration may be a good way to solve this dilemma.
The intrinsic limitation on timing jitter reduction is moderate. Both
experimental and numerical results show that a 50% reduction of timing jitter
can be obtained. However, pulse shape deformation can also happen. This can
also be solved by replacing the CW probe with a pulse train.
CHAPTER 6: 3-R Regeneration
256
The regeneration of extinction ratio degraded signal is much more
effective in PAW-Regeneration, as experimentally demonstrated in the previous
section (Figure 6.6 and Figure 6.7). The only requirement is that ∆V of the
photocurrent signal should be larger enough, which depends on the input power
level, power handling capability and the detection speed of TW-EAM. In
addition, a high modulation efficiency is also necessary since it reduces the
required ∆V for a given extinction ratio.
Technical limitations mainly come from the finite detection speed of the
TW-EAM. A deficient bandwidth can cause (1) long falling tails; (2) pattern
dependence or intersymbol interference or (3) excessive SSB noise. The long
falling tails can be solved by the RF-driven approach as studied in detail in
section 5.4. The pattern dependence can be mitigated by the nonlinear E-O
transfer function but this should be reserved for the degradation in the input
signal. This means that even though the detection bandwidth of the TW-EAM
may not be high enough to resolve the exact pulse shape of the RZ signal (that
may require a bandwidth at least two times the bit-rate), it should be high
enough to avoid pattern dependence and intersymbol interference. This would
suggest a minimal bandwidth about 0.75 times the bit-rate, which is a general
rule for filtered NRZ signal. The excessive SSB noise problem is subtle and is a
relatively minor one unless the required detuning is close to the lock-in range.
This extreme should not occur because that would imply that the detection
CHAPTER 6: 3-R Regeneration
257
bandwidth is way too low and causes serious pattern dependence. However, if
several cascades are required, the excessive SSB noise may accumulate and
result in a significant increase of timing jitter. To solve all the above problems at
once, the TW-EAM should have an adequate detection bandwidth of at least
75% of bit-rate and the RF-driven technique be applied to take care of the pulse
shape.
In short, the compact 3-R PAW-Regenerator proposed in this section can
be a high-performance and small-size RZ wavelength converter when the
detection bandwidth is adequate. As a 3-R regenerator, the requirement for
detection bandwidth is higher so that the regenerator itself does not impose
technical limitations. The regeneration capability is high for extinction ratio
degraded signal, moderate for timing jitter degraded signal, and weak for
amplitude jitter degraded signal. The ultimate limitations on regeneration
capability are intrinsic.
6.3 Simplified optoelectronic 3-R regeneration
To obtain a 3-R regenerator with stronger regeneration capabilities, it is
necessary to re-consider the standard architecture of 3-R regeneration shown in
Figure 6.1. PAW-Regeneration is compact but the merge of the pulse source and
the nonlinear decision gate into a single TW-EAM intrinsically limits the
CHAPTER 6: 3-R Regeneration
258
regeneration strength. Several all-optical approaches for RZ signal regeneration
have been reviewed in Section 6.1, most of them followed the standard 3-R
architecture. In most all-optical 3R regenerators, the input and output optical
waves are mixed in a non-linear medium at high power levels. Despite of their
high-speed potential, there can be several issues for the all-optical approaches:
(1) the non-linear medium may be sensitive to the polarization and wavelength
of the input signal; (2) high power optical amplifiers may be required and
tunable optical filters are needed to suppress the amplified spontaneous emission
(ASE) noise; (3) regeneration at the same wavelength would require cascading
two stages of regenerator with an intermediate wavelength, which complicates
the design.
In terms of these issues, optoelectronic 3R regenerators may have
advantages over their all-optical counterparts: (1) due to the use of a receiver,
the polarization and wavelength dependence can be minimized; (2) high-power
optical amplifiers and tunable optical filters are not required since electrical
amplifiers are used instead; (3) regeneration at the same wavelength without
cascade is inherently viable because the input and output signals are not mixed
optically. These advantages make optoelectronic 3R regeneration very
competitive when the bit-rate can be handled by electronics. Recently, the speed
of optoelectronic 3R regeneration has been increased to 40 Gb/s using electronic
flip-flop circuits [21]. Nevertheless, cost and power consumption are among the
CHAPTER 6: 3-R Regeneration
259
main concerns. It is then of great interest to simplify the configuration of
optoelectronic regenerators, which is the main focus of this section.
6.3.1 Concept
The conventional optoelectronic 3-R regeneration architecture for RZ
data format is illustrated in Figure 6.22(a), where the input optical signal is
converted into an electrical signal for clock recovery and signal processing. The
electronic signal processing unit usually contains flip-flop circuits for re-timing
and re-shaping. It is customary to demultiplex the input signal into lower speed
ETDM (electrical time-domain multiplexing) tributaries so that more mature
circuitry can be utilized to process the electrical signals. Electrical multiplexing
is then necessary to combine the processed tributaries back to the original bit-
rate. In the last stage, a regenerated optical pulse train (can be at a different
wavelength if wavelength conversion is necessary) is modulated by the
processed electrical signal to complete the optoelectronic regeneration. This
approach is proven to be effective and widely used in deployed systems.
However, it is relatively complicated and strongly dependent on the quality and
speed of the electronics circuits. It would be advantageous to simplify the
architecture while keeping the benefits of optoelectronic approaches mentioned
above.
CHAPTER 6: 3-R Regeneration
260
Figure 6.22 (a) Conventional optoelectronic 3R regenerator; (b) simplified optoelectronic 3R regenerator
In this section, it is proposed that the electronic signal processing unit in
an optoelectronic 3-R regenerator can be removed by utilizing the nonlinearity
of EAM, as shown in Figure 6.22(b). A similar idea has been proposed to
suppress amplitude noise in a 2-R regenerator for NRZ data format [22], where
an EAM is directly driven by a uni-traveling carrier photodiode (UTC-PD). No
active electronic circuits were used, which promises a simple construction and
high-speed potential. In this case, the input optical signal is boosted by an
optical amplifier such as EDFA to a high power level so that the photocurrent
signal generated by the UTC-PD is strong enough to drive the EAM (~2V).
Monolithically integrated versions were also reported [23]-[24], but strong
regeneration capabilities have not yet been demonstrated by these “direct-drive”
CHAPTER 6: 3-R Regeneration
261
approaches. As discussed in Chapter 5 and the previous sections in this chapter,
the strength of the photocurrent signal is extremely critical to the vertical re-
shaping capability. The regeneration performance of these direct-drive
approaches is ultimately limited by the maximal photocurrent amplitude
available form the photodiodes. Nevertheless, the realization of a compact and
efficient wavelength converter (with limited regeneration capabilities) is still
promising with this approach.
In this section, the concept of using the nonlinearity of EAM is extended
further to realize a simplified optoelectronic 3-R regenerator where the
electronic signal processing unit is not necessary. However, different from the
direct-drive approaches, an electrical amplifier is used to boost the photocurrent
signal to sufficient amplitudes so that the full potential of the nonlinear E-O
transfer function can be released. At high bit-rates, electrical amplifiers are
relatively mature compared to complex signal processing circuits. Amplifiers
with more than 40-GHz bandwidth are commercially available. The feature of
this approach is that it scales with the advance of electronics.
Referring to Figure 6.22(b), the detected and amplified photocurrent
signal drives the EAM to gate a regenerated (re-timed and well-shaped) pulse
train at the same or another wavelength. This can overcome many intrinsic
limitations of PAW-Regeneration discussed in the previous section. The
amplitude noise in the input signal is suppressed through the nonlinear gating
CHAPTER 6: 3-R Regeneration
262
process. Figure 6.23 shows the step-like E-O transfer function of the EAM,
which is ideal for compressing the noise on the mark and the space levels
(vertical re-shaping). In contrast, the sinusoidal transfer function of a LiNbO3
modulator (LN-MOD) does not have flat transmission regions and,
consequently, does not result in strong noise suppression. The proposed concept
is demonstrated at 10 Gb/s in the next section and the 3-R regeneration
capabilities with respect to timing tolerance, ASE noise degradation, and
dispersion tolerance are evaluated.
Figure 6.23 E-O transfer functions of the EAM used in the experiment and a typical LiNbO3 modulator.
6.3.2 10-Gb/s RZ signal regeneration
The experimental setup of is shown in Figure 6.24. The TW-EAM used
in the experiment requires 2.6 V to change the normalized transmission from 0.1
CHAPTER 6: 3-R Regeneration
263
to 0.9 (defined as the transition voltage). This means that the amplitude of the
electrical driving signal must be larger than 2.6 V in order to utilize the two flats
of the transfer function for effective noise suppression.
Figure 6.24 Experimental setup of 10-Gb/s simplified optoelectronic 3-R regeneration
The input optical signal is a 40-ps, 10-Gb/s RZ data with 231-1 pseudo-
random binary sequence (PRBS) at 1555 nm. The input power level is fixed at –
12 dBm for all experiments in this section. The input signal is first detected by a
10-Gb/s receiver and then boosted by a 10-Gb/s non-inverting broadband
amplifier to 9 V peak-to-peak in order to drive the TW-EAM from the lower
electrical port. The combined electrical gain is 56 dB. Due to lower impedance
(25-Ohm) of the TW-EAM, the amplitude of the electrical driving signal is
reduced to 6.8 V inside the device but this is still 2.6 times the transition voltage.
It is extremely difficult for a photodiode to generate such a high driving voltage
even sufficient input power can be provided by high-power optical amplifiers.
CHAPTER 6: 3-R Regeneration
264
The electrical driving signal then goes out of the upper electrical port of
the TW-EAM and is utilized by a commercial electronic clock recovery unit to
recover a synchronized 10-GHz electrical clock. The nominal timing jitter is 245
fs. A LN-MOD is driven by the recovered electrical clock to regenerate a 10-
GHz, 40-ps optical pulse train at the same wavelength, which is then gated by
the TW-EAM to complete the 3R regeneration.
Figure 6.25 (a) The amplified electrical driving signal; (b), (c), and (d) the corresponding gating
windows at different bias voltages
The finite bandwidth of the receiver and the amplifier can broaden the
pulsewidth of the electrical signal (Figure 6.25(a)), which leads to a wider gating
window. This allows improved timing tolerance without a pulse shaper as
required by all-optical approaches [15]. The optical outputs measured when a
continuous-wave (CW) is fed into the TW-EAM (instead of the pulse train) are
shown in Figure 6.25(b)-(d). They represent the gating window shape at the
CHAPTER 6: 3-R Regeneration
265
respective bias voltages. At –1.5 V of bias voltage (Vb), the mark is sufficiently
clamped and the output eye has a wide flat top, indicating a strong noise
suppression capability on the mark and a wide timing jitter tolerance. However,
the noise on the space is increased compared to that of the driving signal
because the space level of the driving signal now falls in a portion of the E-O
transfer function where the slope is high. The reverse bias has to be increased to
suppress the noise on the space (Figure 6.25(c)-(d)) but at the expense of
reducing the width of the flat top on the mark. Therefore, there should be an
optimal bias point to balance between noise suppressions on the mark and on the
space levels. Ideally, the larger the driving voltage (or the smaller the transition
voltage), the more flat transmission region can be utilized and, hence, better
noise suppression capability can be obtained.
This simplified optoelectronic 3-R regenerator is first evaluated with a
high-quality input signal. The bias voltage is set at – 2.0 V and remains the same
for all experiments in this section. As shown in Figure 6.26(a), the regenerated
eye is very clean and has a slightly reduced pulsewidth of 35 ps due to the shape
of the gating window. Bit-error-rate (BER) results in Figure 6.26(b) show that
no power penalty was imposed on the regenerated signal, indicating that the
regenerator itself does not impose signal degradation. This is necessary for a
high-performance optical regenerator.
CHAPTER 6: 3-R Regeneration
266
Figure 6.26 Results with a high-quality input signal (a) eye diagrams; (b) BER curves
The timing tolerance is evaluated by shifting the phase of the 10-GHz
electrical clock to the LN-MOD, which changes the relative timing of the pulse
train and the gating window generated by the TW-EAM. Figure 6.27 shows that
within 1-dB of power penalty, up to 36 ps of timing shift is allowed. This would
imply high tolerance to the timing jitter degraded input signal, which will be
studied with numerical simulation in the next section. Dependence on the
polarization of the input signal is not observed since the receiver has a
polarization dependent loss of only 0.06 dB, which is difficult to achieve with
all-optical regenerators. Combining polarization insensitivity with the high
timing tolerance, rectification capability of polarization mode dispersion (PMD)
degraded signal can be expected.
CHAPTER 6: 3-R Regeneration
267
Figure 6.27 Timing tolerance measurement
Next, the regeneration of an ASE degraded signal is evaluated. This is
the most common type of degradation in a real transmission system. The optical
signal to noise ratio (OSNR) of the input signal is degraded by adding ASE
noise from an Erbium-doped fiber amplifier (EDFA). The OSNR is degraded to
20 dB (measured with 0.1 nm resolution) and a 2.4 nm optical band-pass filter is
used to suppress out-of-band noise. Figure 6.28(a) shows that the added ASE
leads to observable beating noise on the mark and Figure 6.28(b) indicates that
1.2 dB of power penalty is imposed. After 3-R regeneration, the noise is
significantly reduced as verified by the eye diagram of the regenerated signal in
Figure 6.28(a). 0.8 dB of negative power penalty is obtained compared to the
degraded input signal. This confirms the effectiveness of the regenerator.
CHAPTER 6: 3-R Regeneration
268
Figure 6.28 Results with 20-dB OSNR input signal (a) eye diagrams; (b) BER curves
Figure 6.29 Spectrums of the input signal (before filter: solid black curve; after filter: dashed
black curve) and the regenerated signal (solid gray curve)
The spectrums are plotted in Figure 6.29. The OSNR of the regenerated
signal is improved to 56 dB, which is determined by the laser source in the
CHAPTER 6: 3-R Regeneration
269
regenerator. Note that the filtered input signal is the actual signal that is detected
by the receiver. The high OSNR improvement of 36 dB benefited from the fact
that the input and output waves are not mixed optically in the optoelectronic
regenerator.
The ASE noise suppression capability of the simplified optoelectronic 3-
R regenerator comes in three ways: (1) The finite bandwidth of the receiver and
the electrical amplifier combined filtered out most of the noise outside the
electrical bandwidth; (2) within the electrical bandwidth, the beating noise is
then suppressed by the nonlinear E-O transfer function of the TW-EAM; (3) the
pulse train gated by the TW-EAM comes from a clean CW source inside the
regenerator so that the OSNR of the output signal is high.
On the other hand, all-optical approaches do not have the advantage of
electrical filtering and the high frequency noise can be transferred to the output
signal. In addition, output OSNR can be limited if the input and output signals
are mixed optically, especially when the wavelengths are very close.
The input OSNR is further degraded to 16 dB with a noisier eye diagram
and a 2.8-dB power penalty, as shown in Figure 6.30. An error floor is observed
below 10-10 BER. The amplitude of the electrical driving signal is reduced by
20% due to the increased optical noise. Nevertheless, the regenerated signal has
a negative power penalty of 1.6 dB at 10-9 and the eye is clean. The OSNR is
also as high as 56 dB. However, the error floor still exists after regeneration.
CHAPTER 6: 3-R Regeneration
270
This is due to the fact that the regenerator itself does not correct errors in the
signal, as discussed in Section 6.1. It only refines the shape and timing of the
optical signal so that the accumulation of error is slower in the subsequent
transmission. In addition, the reduced driving voltage may also have reduced the
regeneration strength. These results may suggest that regeneration should be
implemented before the OSNR drops below a certain level.
Figure 6.30 Results with 16-dB OSNR input signal (a) eye diagrams; (b) BER curves
The regeneration of dispersion-degraded signal is evaluated by
propagating the input signal through 25-km of SMF-28 fiber without any
dispersion compensation. As shown in Figure 6.31(a), the eye of the input signal
is broadened and distorted, which leads to 0.5 dB of power penalty. The peak to
CHAPTER 6: 3-R Regeneration
271
peak amplitude of the electrical driving signal is also 20% smaller as in the 16-
dB OSNR case. Nevertheless, the regenerated signal is well shaped without
significant power penalty. A slight change in BER slope is observed, which may
be caused by the mixed effect of decreased electrical amplitude and increased
intersymbol interference due to dispersion. Even though no negative power
penalty is obtained in this case, the dispersion experienced by the input signal is
gone after regeneration and the signal is re-timed and re-shaped.
Figure 6.31 Results with dispersed input signal (a) eye diagrams; (b) BER curves
The ASE noise suppression capability can be further improved by
inverting the polarity of the electrical driving signal. This is due to the fact that
the ASE beating noise is stronger on the mark and the TW-EAM has a much
wider flat transmission region at high reverse bias, as shown in Figure 6.23. The
CHAPTER 6: 3-R Regeneration
272
bias point can be adjusted towards higher reverse bias to suppress more ASE
noise on the mark while keeping the space transparent. This would require a
large ∆V in the driving signal (analogous to the PAW-Regeneration case), which
is met with the use of an electrical amplifier. In addition, the transition voltage
of the TW-EAM should be as low as possible so that the spare part of driving
signal, ∆V minus the transition voltage, can be reserved to remove the amplitude
noise in the input signal.
The concern of degraded signal to noise ratio caused by the electrical
amplifier [22] can be counteracted by nonlinear E-O transformation with large
driving amplitude. In this particular experiment, up to 56 dB of electrical gain
was employed but very effective regeneration can still be obtained. The use of a
photodiode to drive the EAM directly is the most compact realization of this
concept but the regeneration performance would strongly rely on the voltage
swing produced by the photodiode relative to the transition voltage.
The proposed approach can also function as a 3R regenerative
wavelength converter if the internal CW wavelength can be tuned within the
operating wavelength range of the EAM. Given the recent developments in
integration technology [23], high level integration of this concept for an even
lower cost and compact simplified optoelectronic 3R regenerator is promising,
as will be demonstrated in Section 6.4.
CHAPTER 6: 3-R Regeneration
273
6.3.3 Modeling
In this section, the theoretical model described in Section 5.3.1 is
modified to simulation the simplified optoelectronic 3-R regenerator
demonstrated in the previous section. The sequence of simulation is briefly
described in Figure 6.32.
Figure 6.32 Sequence of simulation for simplified optoelectronic 3-R regenerator
The major architectural difference from PAW-Regeneration (Figure
6.15) is that the TW-EAM gates a regenerated pulse train instead of a CW. This
mitigates the pulse deformation problem in PAW-Regeneration with highly
degraded input signal.
The impulse response of the receiver and the electrical amplifier used in
the simulation is shown in Figure 6.33(a). There are small bumps at 250 ps and
400 ps because of the reflections inside the electronic circuits, which will result
in imtersymbol interference. The bandwidth of the receiver and the amplifier
CHAPTER 6: 3-R Regeneration
274
combined is only 4.51 GHz, derived by Fourier transforming the impulse
response. The low bandwidth obtained here is due to the fact that two 10-Gb/s
components (the receiver and the amplifier) are cascaded.
Figure 6.33 (a) impulse response of the receiver and the electrical amplifier combined; (b) the
corresponding power spectrum of the impulse response
Figure 6.34 Electrical driving signal (a) measured; (b) simulated.
Figure 6.34 shows the measured and the simulated eye diagrams of the
electrical driving signal. The impulse response used in the simulation is properly
tuned so that the calculated eyes match the measured ones. Even though the
electrical bandwidth is below 5 GHz, the eye of the driving signal is still open
CHAPTER 6: 3-R Regeneration
275
and clear. The limited bandwidth gives a smooth pulse shape, which is desired
for improved timing tolerance as discussed in the previous section. What really
matters is that the digital information is transferred with a large ∆V.
First, the gating window of the TW-EAM is simulated, which can be
obtained by replacing the pulse train with a CW. The three cases shown in
Figure 6.25 are simulated and the results are plotted in Figure 6.35. Close match
in eye shape is obtained in general. The amplitude of the electrical driving signal
is 4 V peak to peak, which gives the best match. The higher noise on the space
level of the measured eye in the first case is caused by the facts that (1) the
fluctuations on the space level of the measured driving signal is a little bit higher
than that of the simulated signal shown in Figure 6.34. This means that the
actual impulse response of the experimental setup has more bumps than the one
used in the simulation, even though close matching is obtained in general. (2)
small fluctuations on the space are enlarged by the nonlinear E-O transfer
function of the TW-EAM and result in an even more noisy eye. Fortunately, the
cases of interest are those having higher reverse biases where this discrepancy is
not a problem (the fluctuations are well suppressed at high reverse biases).
Even though the gating window shapes are well matched, the bias
voltages are all shifted by – 1.8 V in the simulation compared to the
experimental value. This problem was not encountered in the simulations
presented in Section 5.3, where the material of the TW-EAM is different from
CHAPTER 6: 3-R Regeneration
276
the one used in this case. The E-O transfer function shown in Figure 6.23 is used
in the current simulation, which is measure with CW input with no electrical
driving signal and low optical input power (2 dBm at the fiber input). However,
in the actual regeneration experiment, the electrical driving power is strong and
the dissipated power inside the TW-EAM can heat up the device. The raised
temperature can cause a red-shift of the absorption edge, which results in
smaller bias voltages in the experiment. Nevertheless, it seems that the E-O
transfer function is shifted without much change in shape since the bias steps are
not changed among the three cases shown in Figure 6.35. Therefore, the design
of low power dissipating CPW line in the TW-EAM is important in this regard.
Figure 6.35 Gating window shapes at different bias voltages (a) measured; (b) simulated. The amplitude of the electrical driving signal is 4 V peal to peak for call cases.
CHAPTER 6: 3-R Regeneration
277
The regenerated signal with high-quality input is shown in Figure 6.36.
The experimental bias voltage is – 2 V and the bias in the simulation is – 3.8 V.
Other than that, close agreement is obtained.
Figure 6.36 Regenerated optical eye diagrams with high-quality input signal (a) experimentally
measured (10 ps/ div); (b) numerically simulated at Vb= - 3.8 V.
6.3.4 Simulation of cascaded performance
Since the modeling presented in the previous section works well to
model the regenerator, it is applied in this section to study the cascaded
performance of the simplified optoelectronic 3-R regenerator. This is a
particularly important issue when the regenerator is used as a regenerative
wavelength converter. Cascaded performance of optical NRZ wavelength
conversion was studied in detail in Section 5.3.2 and one of the conclusions is
that the cascadibility is ultimately limited by the finite rise and fall time of the
wavelength converter. However, for RZ signal regeneration, it is possible to
avoid this problem when the pulses are regenerated in each stage of conversion.
This will be verified numerically in this section.
CHAPTER 6: 3-R Regeneration
278
The cascadibility strongly depends on the bias voltage of the TW-EAM,
which determines the shape of the gating window together with the amplitude of
the driving signal and the shape of the transfer function. Three voltages, – 3.5 V,
– 3.8 V and – 4.6 V, are chosen as shown in Figure 6.37 to represent three types
of situation. At – 3.5 V, the mark is well clamped but the extinction ratio is a
little bit compromised. On the contrary, at – 4.6 V, the extinction is excellent but
clamping on the mark is reduced. At – 3.8 V, the gating window is something in
between. This is the bias value obtained by fitting with the measured eye
diagram in Figure 6.36.
Figure 6.37 Simulated gating window shapes at three different biases (a) - 3.5 V; (b) - 3.8 V; (c) - 4.6 V
Figure 6.38 shows the output eye diagrams after several cascades for the
three bias situations. It is assumed that the peak input power at each stage of
regenerator is the same, which gives 4 Vpp of electrical driving signal. With only
one stage of regeneration (or wavelength conversion if the wavelength is
changed), the differences between the three biases are not significant. However,
after several cascades, the imperfections in clamping or in extinction
CHAPTER 6: 3-R Regeneration
279
accumulate, which result in eye closures. However, in the Vb = – 3.8 V case, the
eye diagram seems to remain the same even up to 5 cascades.
Figure 6.38 Simulated eye diagrams of the regenerated signals with different numbers of
cascade at (a) – 3.5 V; (b) – 3.8 V; (c) – 4.6 V of bias.
In order to quantify the change in signal quality, the eye height penalty
defined by Equation 5.2 in Section 5.3.2 is used. The width of the mask is still
0.2 unit internal (UI). Figure 6.39 shows the dependence of eye height penalty
on bias voltage for several different numbers of cascade. It is clear that
cascadibility strongly depends on bias voltage. It is interesting that in a small
range of bias around – 4.0 V, the eye height penalty does not seem to increase
with up to 5 cascades. To probe the cascadibility further, the regenerator is
CHAPTER 6: 3-R Regeneration
280
numerically cascaded up to 50 times around this bias point and the calculated
eye height penalties are shown in Figure 6.40. It is amazing from this figure that
in the bias range of – 3.7 V to – 4.1 V, the eye height penalty converges to stable
and small values (less than 0.4 dB). The simulated eye diagram after 50
cascades looks almost the same as that with only one stage of regeneration
(Figure 6.41). This implies that the regenerator may be cascaded for unlimited
times. Surely this is not true since statistical noise, which is not considered in
this model, will accumulate and eventually degrade the signal. However, these
results do show that it is possible for this regenerator to be free from
deterministic degradations. This is in contrast to optical NRZ wavelength
conversions, where the pulse shape degrades unavoidably because of finite rise
and fall times, even in the absence of statistical noise.
Figure 6.39 Eye height penalty as function of reverse bias for one, three, and five
times of regeneration
CHAPTER 6: 3-R Regeneration
281
Figure 6.40 Eye height penalty as function of cascades at different bias points. The gray arrows
indicate the direction of increased reverse bias.
Figure 6.41 Simulated eye diagrams with – 3.8 V of bias voltage (a) input signal; (b) after one
stage of regeneration; (c) after 50 stages of regeneration
6.3.5 Simulation of regeneration capabilities
The regeneration of degraded optical signal is simulation in this section.
First, the input signal is degraded with amplitude jitter (noise). This is done by
scaling the amplitude of each bit by a factor (1+ x), where x is a random variable
averaged at zero and following the Gaussian distribution characterized by the
standard deviation, σ. Three regenerator configurations are considered: the
CHAPTER 6: 3-R Regeneration
282
amplitude of the electrical driving signal takes values of 3 Vpp, 4 Vpp, and 5 Vpp.
Note that 4 Vpp is the case that matches the experiment presented in the previous
section where the corresponding bias voltage is – 3.8 V. The bias voltages for
the cases with 3 Vpp and 5 Vpp are chosen so that the extinction is the same as in
the 4 Vpp case. This means that for larger driving amplitudes, more ∆V is
clamped on the mark level.
Figure 6.42 Eye diagrams of the amplitude jitter degraded input signal (upper) and the regenerated signal (lower) for the regenerator with (a) Vpp = 3, Vb = - 4.2 V; (b) Vpp = 4,
Vb = - 3.8 V; (c) Vpp = 5, Vb = - 3.4 V.
The upper row of Figure 6.42 shows the input signals with 10% standard
deviation. The eyes may look different from each other but the standard
deviations are all 10%. The regenerated eyes are shown on the lower row of
Figure 6.42. As the amplitude of the electrical driving signal increases, the
amplitude jitter in the regenerated signal decreases. For the 5 Vpp case, almost
CHAPTER 6: 3-R Regeneration
283
70% reduction of amplitude jitter is obtained. The pulse shape deformation
problem encountered in PAW-Regeneration as shown in Figure 6.21 does not
happen in this approach.
Figure 6.43 Relationship of input and output standard deviations of the mark level
for the three regenerator configurations
The standard deviation of the amplitude jitter is varied up to 20 % and
the relationship of the input and the output are plotted in Figure 6.43 for the
three regenerator configurations. It is clear that with larger driving amplitudes,
the reduction of amplitude jitter is more significant. On the other hand, if the
driving amplitude is low like the 3 Vpp case, the amplitude jitter of the output
can easily be larger than that of the input (above the 1:1 line). In the “direct-
drive” approach, where the photodiode directly drives the EAM, the highest
reported driving amplitude is only 2 Vpp [22] (measured with 50-Ohm
termination). Although this case is not explicitly shown in Figure 6.43, it can be
CHAPTER 6: 3-R Regeneration
284
anticipated that the curve with 2 Vpp should be well above the 1:1 line (if the
same EAM is used), which means that there is no regeneration capability at all.
After the conversion, the amplitude jitter most likely gets worse. In reality, the
impedance of TW-EAMs is lower than 25 Ohm and the actual voltage swing on
the EAM can be even lower. This shows again the importance of strong driving
amplitude for effective regeneration.
Figure 6.44 Eye diagrams of the timing jitter degraded input signal (upper), the corresponding
gating window, and the regenerated signal (lower) for the regenerator with (a) Vpp = 3, Vb = - 4.2 V; (b) Vpp = 4, Vb = - 3.8 V; (c) Vpp = 5, Vb = - 3.4 V.
Next, the regeneration of timing jitter degraded signal is simulated. The
timing jitter is imposed on the input signal bit by bit. The shift of the pulse in
CHAPTER 6: 3-R Regeneration
285
each bit is determined by a random variable, whose distribution is even in a
specified range (instead of a Gaussian distribution). The upper row of Figure
6.44 shows the input signals with 20 ps peak to peak timing jitter. The gating
windows shown in the middle row are clamped and broadened on the mark.
Therefore, even in the presence of timing jitter, there is still a large and clear eye
opening to gate the regenerated pulse train.
The output signals are shown in the lower row in Figure 6.44. The timing
jitter is significantly reduced to 27%, 17%, and 11% of the original value,
respectively, in the three cases. These are all better than the 50% timing jitter
reduction achieved with PAW-Regeneration as presented in Section 5.4.3 and
Section 6.2.4. Pulse shape deformation is not a problem even with 20 ps (0.2 UI)
of timing jitter. It is verified again that stronger driving amplitude leads to better
regeneration capability. The reason is that there is more clamped and flattened
region in the gating window to allow the timing to be shifted between the pulse
train and the gate. In Figure 6.27, it is measured experimentally that the timing
tolerance of the implemented regenerator is 36 ps within 1 dB of power penalty.
Figure 6.45 shows the calculated input and output timing jitter relationship for
the three cases. Compared to the amplitude reduction capability in Figure 6.43,
the capability of timing jitter reduction is even more effective.
CHAPTER 6: 3-R Regeneration
286
Figure 6.45 Relationship of input and output timing jitters for the three configurations
In a real regenerator, the timing jitter reduction capability may also
depend on the frequency of the jitter. The simulated jitter here is imposed in a
bit by bit fashion. Therefore, the frequencies are close to the bit-rate. As shown
in Chapter 4, timing jitters with frequencies over the lock-in range of the clock
recovery unit (usually several MHz or lower) are suppressed in the recovered
clock. The jitter only transfers to the timing of the gating window. However,
because of the flatted shape of the window, the regenerated pulse train does not
sense the effect of the jitter and the regenerated output is free of high frequency
jitter.
In contrast, if the frequency of the input jitter is below the lock-in range
of the clock recovery unit, both the gate and the regenerated pulse train will
have the same kind of jitter. As a result, these low frequency jitters will be
CHAPTER 6: 3-R Regeneration
287
transferred to the regenerated signal. Transfer of low frequency jitter is actually
required for a practical clock recovery unit since the receiver or the regenerator
has to follow the low frequency timing drift in the incoming signal caused by
fiber transmission.
Figure 6.46 Regenerated optical signals with varied pulsewidth (a) 20 ps; (b) 30 ps; (c) 40 ps.
The input signal is 40 ps wide with 20 ps peak to peak timing jitter.
Figure 6.47 Input and output timing jitters with different regenerator pulsewidths
The jitter reduction capability can be improve further if the width of the
regenerated pulse train inside the regenerator is shortened. Figure 6.46 shows
CHAPTER 6: 3-R Regeneration
288
the regenerated signals with 20 ps, 30 ps, and 40 ps pulsewidths. It is clear that
jitter is further reduced with shorter pulses. This is due to the fact that a shorter
pulse samples a smaller portion of the gating window and senses less
transmission change related to timing jitter. With the 20-ps pulse train, the
timing jitter is reduced from 20 ps to 0.8 ps, a 96 % reduction. The input and
output jitter relationships for several different regenerator pulsewidths are
plotted in Figure 6.47, showing a very strong re-timing capability.
In summary, the simplified optoelectronic 3-R regenerator is
demonstrated experimentally and studied numerically to verify its feasibility and
effectiveness for amplitude and timing jitter reduction. As a regenerative
wavelength converter, the simulation also shows superb cascadability.
6.4 Integrated optoelectronic 3-R regeneration
The concept of simplified optoelectronic 3-R regenerator has been
demonstrated and proven to be very effective in Section 6.3. The merit of this
approach would be further enhanced if it can be even more compact while
keeping all the good features. In this section, monolithic integration technology
is brought in to this purpose, where the CW laser source and the modulators are
all integrated on a single chip in order to reduce the size and the coupling loss of
the 3-R regenerator compared to the discrete version presented in Section 6.3. In
CHAPTER 6: 3-R Regeneration
289
addition, the commercial clock recovery unit used in Section 6.3 is replaced by
the injection locking ring oscillator introduced in Section 4.4 so that the entire
regenerator can be constructed with very few components and hybrid integration
can be applied to minimize the 3-R regenerator / wavelength converter into very
small dimensions.
6.4.1 Experimental Setup
The schematic setup of the integrated optoelectronic 3-R regenerator is
shown in Figure 6.48. The monolithically integrated device is fabricated on an
InP substrate with the integration technologies documented in Ref. [23] and the
references therein. The shallow QW TW-EAMs used in Section 5.2.4 to
improve the speed of PAW-Conversion is also based on this integration
platform, which promises further integration with other components such as
semiconductor optical amplifiers (SOAs), power combiners or even optical
filters to minimize the form factor. The CW source is a tunable sampled-grating
distributed Bragg reflector (SGDBR) laser, capable of continuous wavelength
tuning in the entire C-band. This makes the proposed concept very agile as a
regenerator or a wavelength converter. This is very critical for the
reconfigurable WDM networks. In the implemented device, the SGDBR laser is
followed by an SOA to boost the power. The OSNR degradation imposed by the
SOA is very little because of a high input power.
CHAPTER 6: 3-R Regeneration
290
Figure 6.48 Configuration of the integrated optoelectronic 3-R regenerator
Two EAMs are integrated in tandem following the SOA. EAM1 is 200
µm long and EAM2 is 600 µm. Since they are both terminated with an open, the
modulation bandwidth is limited by their lengths and the devices are behaving
like lumped devices at 10-Gb/s. For the best result, EAM1 is chosen as the
nonlinear optical gate because of its higher bandwidth and EAM2 is driven by
the recovered electrical clock to carve pulses. The sequence is inverted
compared to the discrete version presented in the previous section. However, the
superposition principle should apply and the final results are the same.
The E-O transfer function of the 200 µm EAM (EAM1) is shown in
Figure 6.49. It is very unfortunate that the transition edge is red-shifted
compared to the discrete version and the flat-top of the transfer function is
CHAPTER 6: 3-R Regeneration
291
basically gone. This would significantly degrade the regeneration capability of
the 3-R regenerator. A new generation of device with material engineering is
required. However, since there is still a flat transmission region at high reverse
bias for this particular device, one-end re-shaping should still be feasible. Due to
the fact that the ASE beating noise is stronger on the mark of the input signal,
the polarity of the electrical driving signal is inverted by reversing the
microwave connection to EAM1, as indicated in Figure 6.48. This should
enhance the capability of the regenerator to remove the ASE noise on the mark
of the input signal.
Figure 6.49 E-O transfer functions of the EAMs used in the discrete version and
the integrated version of simplified optoelectronic 3-R regenerator
The clock recovery part of the regenerator subsystem is an injection
locking ring oscillator, which is exactly the one characterized in detail in Section
4.4. It was demonstrated there that this injection locking clock recovery is
capable of recovering a low jitter electrical clock ( ~ 200 fs RMS timing jitter)
CHAPTER 6: 3-R Regeneration
292
and the dependence on the OSNR of the input signal is extremely low as shown
in Figure 4.26. In the experiment, the phase of the recovered electrical clock
relative to the gating window is adjusted manually through an adjustable
microwave delay line. The input power to the photodetector is fixed at -1.65
dBm in all experiments.
6.4.2 10-Gb/s RZ signal regeneration
Following the experiments on the simplified optoelectronic 3-R
regenerator in Section 6.3.2, the integrated 3-R regenerator is first tested with a
high-quality input signal with an OSNR of 40 dB. The operating wavelength is
1555.34 nm for both the input and output signals. The electrical amplifier after
the receiver can be operated in two modes: unsaturated and saturated.
In the unsaturated mode the amplification is basically linear and the
output signal is only modified by the bandwidth of the amplifier, which is
specified at 10 GHz. In the saturated mode, the input signal to the amplifier is
increased by changing the input power to the receiver. When the amplifier is
saturated, its voltage transfer function becomes nonlinear and the output signal
is clamped at certain output amplitude. This would make the boost amplifier
working like a limiting amplifier and the clamping effect is helpful to remove
the amplitude fluctuations.
CHAPTER 6: 3-R Regeneration
293
Figure 6.50 shows the measured eye diagrams with high-quality input
signal. For the saturated case, the input optical power is increased by 6.8 dB in
order to saturate the amplifier, compared to the unsaturated case. It is clear that
the output signal of the saturated amplifier is well clamped and the mark and the
space levels are very clean and flat compared to that of the unsaturated output.
With the clock signal to EAM2 turned off, the gating window shape can be
obtained, as shown in the middle row in Figure 6.50. The gating window with a
saturated amplifier looks pretty much like an NRZ signal due to the double
nonlinearity of the saturated amplifier and the EAM. The noise is increased a
little bit especially on the mark level of the optical gate. This could be caused by
the impedance mismatch of the EAM which causes multiple reflections and
intersymbol interference. Note that the polarity is inverted.
For the case with unsaturated amplifier, the mark of the driving signal
(now the space of the gating window) is clamped due to the flat region of the
transfer function at high reverse bias. However, the fluctuation on the space of
the driving signal is not clamped due to the lack of a flat transmission region at
low reverse bias. Instead, the fluctuation gets worse on the mark of the gating
window. Multiple reflections may have also increased the noise on the mark.
The regenerated signals are shown in the bottom row of Figure 6.50. As
can be expected, the output eye with unsaturated amplifier is noisier. The BER
measurements are shown in Figure 6.51. Compared to the high-quality input, the
CHAPTER 6: 3-R Regeneration
294
power penalty with saturated amplifier is + 0.6 dB and + 1.6 dB with
unsaturated amplifier. Note that the receiver used in this experiment is more
sensitive than the one used in the discrete version. Here, the receiver sensitivity
for the high-quality signal is improved by 3.5 dB.
Figure 6.50 Eye diagrams of the integrated optoelectronic 3-R regenerator with high-quality input signal
Since the electrical driving signal of the saturated amplifier looks clean,
the 0.6 dB power penalty should come from the impedance mismatch or signal
integrity problems of the EAM. The microwave probe used here has a 50 Ohm
parallel resistor for improved impedance matching as discussed in Section 5.2.4.
Nevertheless, the matching is not perfect and multiple reflections may still exist.
If a normal probe is used instead (without the parallel resistor), the eye diagram
CHAPTER 6: 3-R Regeneration
295
is even worse. On the other hand, for the unsaturated amplifier, the noise in its
output signal (most likely caused by pattern dependence due to a rippled impulse
response) is transferred to the gating window and finally to the regenerated
signal. The lack of both flat transmission regions in the current EAM makes this
problem worse. The same amplifier was used in the discrete version and superb
performance was obtained because the discrete EAM has a more ideal transfer
function and also a better microwave termination. The signal degradation due to
electronics is one of the main issues of conventional optoelectronic regenerators.
Comparing the input and the output powers, the case with unsaturated
amplifier has a power gain of 7.8 dB while it is only 0.2 dB for the saturated
case. Note that no optical amplifier was used for amplifying the input or the
output signal, indicating the re-amplification potential of this approach.
Figure 6.51 BER curves with a high-quality (40-dB OSNR) input signal
CHAPTER 6: 3-R Regeneration
296
Next, the OSNR of the input signal is degraded to 20 dB. The
configurations of the regenerator remain the same as in the high-quality input
case. The eye of the electrical driving signal is perceivably noisier with
unsaturated amplifier while the change is not much for the saturated amplifier.
However, it can be observed that the noise on the mark of the gating window
with saturated amplifier does increase compared to the high-quality input case
because the noise is amplified by the non-flat E-O transfer function. The
regenerated eyes both look clean and pretty much the same as in the high-quality
input case. The subtle differences can only be observed through the BER
measurements.
Figure 6.52 Eye diagrams of the integrated optoelectronic 3-R regenerator with a 20-dB OSNR input signal
CHAPTER 6: 3-R Regeneration
297
Figure 6.53 BER curves with 20-dB OSNR input signal
Figure 6.53 shows the measured BER curves. The ASE degraded 20-dB
OSNR signal has 1.2 dB of power penalty compared to the original high-quality
signal. A gradual change in slope is observed indicating the emergence of an
error floor. The regenerated signal with unsaturated amplifier has a + 0.9 dB
power penalty compared to the degraded signal at 10-9 BER while the power
penalty is – 0.4 dB with saturated amplifier. The negative power penalty mainly
comes from the suppression of amplitude noise by the saturated amplifier. The
BER curve with saturated amplifier shows a steeper and constant slope similar
to that of the high-quality signal. On the other hand, the BER curve with
unsaturated amplifier follows closely with that of the degraded signal, implying
no reduction of noise. The + 0.9 dB power penalty should be caused by the
signal degradation caused by the unsaturated electrical amplifier as already
discussed in the case with high-quality input signal.
CHAPTER 6: 3-R Regeneration
298
The OSNR of the input signal is further decreased to 16 dB and + 4.4 dB
of power penalty occurs compared to the high-quality signal. After regeneration,
both configurations give a regenerated eye that is cleaner than the input eye, as
shown in Figure 6.54. The BER results in Figure 6.55 show that – 2.3 dB of
power penalty can be obtained with saturated amplifier while it is – 1.0 dB with
unsaturated amplified. Error floors are observed in all three curves. As discussed
earlier in this chapter, the regenerator does not correct errors that go beyond the
decision threshold. Therefore, it is not possible for the regenerator to completely
remove the error floor. The same phenomenon was also observed in the discrete
version. However, the measured negative power penalties indicate that the signal
quality does get improved by the regenerator so that after the EDFA in the pre-
amplified receiver, the signal does not degrade as much as the original signal.
The negative power penalty obtained with unsaturated amplifier mainly comes
from the electrical filtering effect. On the other hand, in the case of saturated
amplifier, the clamping effects due to saturation further reduce the in-band noise
and more negative power penalty can be obtained.
Similar to the discrete version, the OSNR of the regenerated signal is
determined solely by the laser source used in the regenerator, unless optical
amplifiers are used at the output stage. Figure 6.56 shows the spectrums for the
cases with high-quality and 16-dB OSNR input signals. The regenerated signals
all have an OSNR of 45 dB, no matter what the input OSNR is. The side-bands
CHAPTER 6: 3-R Regeneration
299
of the regenerated signal come from the finite suppression ratio of the SGDBR
laser.
Figure 6.54 Eye diagrams of the integrated optoelectronic 3-R regenerator
with a 16-dB OSNR input signal
Figure 6.55 BER curves with 16-dB OSNR input signal
CHAPTER 6: 3-R Regeneration
300
Figure 6.56 Optical spectrums of the input signal and the regenerated signal (a) with high-
quality (40-dB OSNR) input; (b) with 16-dB OSNR input
6.5 Summary
Three TW-EAM-based 3-R regenerators are proposed and studied in
detail in this chapter. The compact 3-R PAW-Regenerator is a very unique 3-R
regenerator that is based on a single TW-EAM with clock recovery, re-shaping,
and re-timing realized in one setup simultaneously. Several unique properties of
the TW-EAM were applied to obtain such an ultra-compact 3-R regenerator.
The limitations of its performance were also studied in detail. The technical
limitations mainly come from the limited detection bandwidth of the TW-EAM
while the intrinsic limitations are inherent to the regeneration mechanisms. The
inherent limitations can be regarded as the price for merging three functions
simultaneously into a single TW-EAM. The regeneration strength is thus only
CHAPTER 6: 3-R Regeneration
301
moderate. However, it can be a good choice for a compact 3-R regenerative
wavelength converter when the input signal is not seriously degraded.
For stronger regeneration capabilities, the simplified optoelectronic 3-R
regenerator was proposed, where the nonlinear E-O transfer function of the
TW-EAM was utilized to simplify the configuration of traditional O-E-O
regenerators. Only electrical amplifiers were required to boost the amplitude of
the driving signal but not a complicated signal processing circuit. This would
put the least burden on electronics while keeping all the benefits it provides.
Since the standard 3-R architecture is followed, very strong and effective 3-R
regenerations were demonstrated. Numerical simulations showed excellent
cascadibility and strong regeneration of timing jitter and amplitude jitter
degraded signals.
Finally, an integrated version of the simplified optoelectronic 3-R
regenerator is presented, where the transmitter part of the regenerator is
monolithically integrated on a single chip. Even though the transfer function of
the realized device was not perfect, effective ASE noise suppression capability
was demonstrated. If hybrid integration of electronic and optoelectronic
components are implemented, a very high-performance, low cost, and small-size
3-R regenerator or regenerative wavelength converter is promising for the
demand of future optical networks.
CHAPTER 6: 3-R Regeneration
302
References
[1] J.-X. Cai, D. G. Foursa, L. Liu, C. R. Davidson, Y. Cai, W. W. Patterson, A. J. Lucero, B. Bakhshi, G. Mohs, P. C. Corbett, V. Gupta, W. Anderson, M. Vaa, G. Domagala, M. Mazurczyk, H. Li, M. Missov, A. N. Pilipetskii, and Neal S. Bergano, “RZ-DPSK Field Trial over 13,100 km of Installed Non Slope-Matched Submarine Fibers”, Optical Fiber Communication Conference (OFC), paper PDP34, 2004
[2] D. J. Blumenthal, B.-E. Olsson, G. Rossi, T. E. Dimmick, L. Rau, Milan Masanvic, O. Lavrova, R. Doshi, O. Jerphagnon, J. E. Bowers, V. Kaman, L. A. Coldren, and J. Barton, “All-Optical Label Swapping Networks and Technologies”, J. Lightwave Technol., vol. 18, no. 12, pp. 2058-2075, 2000
[3] C. Bornholdt, J. Slovak, C. Schmidt, and B. Sartorius, “Mitigation of PMD @ 40 Gb/s Using an Optically Clocked 3R Regenerator”, Optical Fiber Communication Conference (OFC), paper OTuO6, 2005
[4] V. Kaman, X. Zheng, C. Pusarla, A. J. Keating, R. J. Helkey, and J. E. Bowers, “Mitigation of optical crosstalk penalty in photonic cross-connects using forward error correction”, Electron. Lett, vol. 39, no. 8, pp. 678-679, 2003
[5] J. Livas, “Optical Transmission Evolution: From Digital to Analog to ? Network Tradeoffs Between Optical Transparency and Reduced Regeneration Cost”, J. Lightwave Technol., vol. 23, no. 2, pp. 219-224, 2005
[6] M. Rochette, J. N. Kutz, J. L. Blows, D. Moss, J. T. Mok, and B. J. Eggleton, “Bit-Error-Ratio Imporvement With 2-R Optical Regenerators”, IEEE Photon. Technol. Lett., vol. 17, no. 4, pp. 908-910, 2005
[7] F. G. Sun, Z. G. Lu, G. Z. Xiao, and C. P. Crover, “Optimal Position of Nonlinear Regenerators in Optical Communication Links”, IEEE Photon. Technol. Lett., vol. 16, no. 5, pp. 1406-1408, 2004
[8] P. Ohlen and E. Berglind, “Noise Accumulation and BER Estimates in Concatenated Nonlinear Optoelectronic Repeaters”, IEEE Photon. Technol. Lett., vol. 9, no. 7, pp. 1011-1013, 1997
[9] O. Leclerc, B. Lavigne, E. Balmefrezol, P. Brindel, L. Pierre, D. Rouvillain, and F. Seguineau, “Optical Regeneration at 40 Gb/s and Beyond,” J. Lightwave Technol., vol. 21, no. 11, pp. 2779-2790, 2003
[10] D. Wolfson, A. Kloch, T. Fjelde, C. Janz, B. Dagens, M. Renaud, “40-Gb/s all-optical wavelength conversion, regeneration, and demultiplexing in an SOA-based all-active Mach-Zehnder interferometer”, IEEE Photon. Technol. Lett., vol. 12, no. 3, pp. 332-334, 2000
[11] T. Otani, T. Miyazaki, and S. Yamamoto, “40-Gb/s Optical 3R Regenerator Using Electroabsorption Modulator for Optical Networks,” J. Lightwave Technol., vol. 20, pp. 195-200, 2002
[12] J. Leuthold, G. Raybon, Y. Su, R. Essiambre, S. Cabot, J. Jaques, and M. Kauer, “40 Gbit/s transmission and cascaded all-optical wavelength conversion over 1000000 km”, Electron. Lett, vol. 38, no. 16, pp. 890-892, 2002
CHAPTER 6: 3-R Regeneration
303
[13] H. J. Thiele, A. D. Ellis and I. D. Phillips, “Recirculating loop demonstration of 40Gbit/s all-optical 3R data regeneration using a semiconductor nonlinear interferometer”, Electron. Lett, vol. 38, no. 16, pp. 890-892, 2002
[14] G. Raybon, Y. Su, J. Leuthold, R.-J. Essiambre, T. Her, C. Joergensen, P. Steinvurzel, K. Dreyer, K. Feder, “40 Gbit/s Pseudo-linear Transmission Over One Million Kilometers”, Optical Fiber Communication Conference (OFC), paper FD10, 2002
[15] S. Watanabe, F. Futami, R. Okabe, Y. Takita, S. Ferber, R. Ludwig, C. Schubert, C. Schmidt, and H. G. Weber, “160 Git/s Optical 3R-Regenerator in a Fiber Transmission Experiment”, Optical Fiber Communication Conference (OFC), paper PD16, 2003
[16] E. S. Awad, P. S. Cho, C. Richardson, N. Moulton, and J. Goldhar, “Optical 3R Regeneration Using a Single EAM for All-Optical Timing Extraction With Simultaneous Reshaping and Wavelength Conversion”, IEEE Photon. Technol. Lett., vol. 14, pp. 1378-1380, Sep. 2002
[17] C. Bornholdt, J. Slovak and B. Sartorius, “Semiconductor-based all-optical 3R regenerator demonstrated at 40 Gbit/s”, Electron. Lett., vol. 40, pp. 192-193, Feb. 2004
[18] H.-F. Chou, Z. Hu, J. E. Bowers, and D. J. Blumenthal, “Compact Optical 3R Regeneration Using a Traveling-Wave Electroabsorption Modulator”, IEEE Photonics Technology Letters, vol. 17, no. 2, pp. 486-488, 2005
[19] H.-F. Chou and J. E. Bowers, “Simplified optoelectronic 3R regenerator using nonlinear electro-optical transformation in an electroabsorption modulator”, Optics Express, vol. 13, no. 7, pp. 2742-2746, April 2005
[20] H.-F. Chou, J. E. Bowers, and D. J. Blumenthal ,"Novel Photocurrent-assisted wavelength (PAW) converter using a traveling-wave electroabsorption modulator with signal monitoring and regeneration capabilities", Optical Fiber Communication Conference (OFC´04), paper FD4, Feb. 2004, Los Angeles, CA
[21] W. Kuebart, B. Lavigne, M. Witte, G. Veith, and O. Leclerc, “40 Gb/s transmission over 80000 km dispersion shifted fibre using compact opto-electronic-3R regeneration,” in Proc. European Conference on Optical Communication 2003, Rimini, Italy, September 21-25, 2003, paper Mo4.3.1.
[22] Y. Kisaka, A. Hirano, M. Yoneyama, and N. Shimizu, “Simple 2R repeater based on EA modulator directly driven by uni-traveling-carrier photodiode,” Electron. Lett., vol. 35, pp. 1016-1017, 1999
[23] M. N. Sysak, J. S. Barton, L. A. Johansson, J. W. Raring, E. J. Skogen, M. L. Masanovic, D. J. Blumenthal, and L. A. Coldren, “Single-Chip Wavelength Conversion Using a Photocurrent-Driven EAM Integrated With a Widely Tunable Sampled-Grating DBR Laser,” IEEE Photonics Technology Letters, vol. 16, pp. 2093-2095, 2004
[24] T. Yoshimatsu, S. Kodama, K. Yoshino, and H. Ito, “100 Gbit/s error-free retiming operation of monolithic optical gate integrating with photodiode and electroabsorption modulator”, Electron. Lett., vol. 40, pp. 626-628, May 2004
CHAPTER 6: 3-R Regeneration
304
This page intentionally left blank
305
Chapter 7 Summary and Future Work 7.1 Summary
This dissertation is about developing compact sub-systems to meet the
demands of future optical networks by utilizing various aspects of the traveling-
wave electroabsorption modulator (TW-EAM). It can be briefly summarized
with a flowchart as shown in Figure 7.1. On the device side, the TW-EAM has
many unique properties that can be utilized to provide novel solutions to system-
level demands. In the scope of this dissertation, the properties explored include:
(A) the traveling-wave electrode and its associated distributed effect,
(B) the photocurrent signal generated inside the device,
(C) the nonlinear E-O transformation of the TW-EAM,
(D) the integrability with other optoelectronic devices.
CHAPTER 7: Summary and Future Work
306
There are two general philosophies that guided the direction of research
in this dissertation. The first philosophy is that the proposed sub-systems must
reach higher performance and more functionality with less complexity. This
would ensure the scalability of the proposed approaches. The other philosophy is
that mature electronic technologies should be considered if better performance
can be obtained without violating the first philosophy. It is believed that the
right blend of electronics and optics can provide the best solution.
Traveling-Wave Electroabsorption Modulator
Traveling-waveelectrodes
Nonlinear E-OtransformationPhotocurrent Integration
technology
Traveling-wavephenomena
PAW-Conversion160G scaled OTDM clock recovery with
simultaneous demux
Standing-wave enhanced mode
OTDM Gating Operations
40G pulse generation160G add-drop multiplexing
with 40G base-rate160G demultiplexing
Injection-lockingclock recovery
RF-Driven PAW-Conversion
Compact 3-RPAW-Regeneration
Simplified optoelectronic
3R regeneration
Integratedoptoelectronic
3R regeneration
Novel Solutions
Sys
tem
& S
ub-S
yste
mD
evic
e
OTDM WDMClock Recovery
Figure 7.1 Flowchart of the dissertation
On the system and sub-system side, the research of this dissertation can
be classified into three categories: OTDM, Clock Recovery, and WDM.
CHAPTER 7: Summary and Future Work
307
For OTDM applications, the focus was on building essential signal
processing sub-systems with electrically-driven TW-EAMs, which can lower the
complexity and the cost associated with OTDM whenever the speed can be
handled by the device. The line-rate of particular interest was 160 Gb/s, which
can be built based on 10-Gb/s ETDM or even 40-Gb/s ETDM that is becoming
available in recent years. Properties (A) and (C) are the enabling factors in this
part of research.
First, the traveling-wave phenomena in 40-GHz optical pulse generation
were studied. The experimental results were supported by numerical simulations
based on a transmission line model. This confirmed the traveling-wave
operation of the device and indicated the importance of distributed effect when
the device dimension is comparable to the microwave wavelength.
To generate shorter pulses, a standing-wave enhanced mode of the TW-
EAM was proposed and verified experimentally and theoretically. By properly
adjusting the length of the termination CPW line, the microwave distribution
along the active waveguide can be optimized to generate optical pulses as short
as 2.4 ps at 40 GHz.
As optical gates, two standing-wave enhanced TW-EAMs were utilized
to demonstrate the first 160-Gb/s add-drop multiplexing with a 40-Gb/s base-
rate. An average power penalty of 1 dB was demonstrated for all operations (
including the dropped, the through, and the added channels). Up to date, this
CHAPTER 7: Summary and Future Work
308
remains the only semiconductor-based solution that can operate with a 40-Gb/s
base-rate. For demultiplexing, the standing-wave enhanced mode can lower the
driving voltage required for 80- to 10-Gb/s demultiplexing using a single EAM.
An average power penalty as low as 0.55 dB was obtained. To increase the line-
rate speed up to 160 Gb/s, a TW-EAM was driven by a shaped electrical signal
composed of harmonics at 10 GHz and 20 GHz. The first single-stage 160- to
10-Gb/s EAM-based demultiplexer was demonstrated with power penalties
lower than that of the previously reported two-stage configuration (1 dB vs. 2.8
dB). These works showed that electrically-driven TW-EAMs can be compact
and efficient optical demultiplexers up to 160 Gb/s with both 40-Gb/s and 10-
Gb/s base-rates.
For clock recovery applications, property (B) is intensively utilized
together with (A) and (C). When combined with an optoelectronic phase-locked
loop, the TW-EAM enabled a clock recovery sub-system at 40-Gb/s with
simultaneous demultiplexing capability. Three functions (detection,
demultiplexing, and pulse generation) were performed simultaneously within a
single TW-EAM by using microwave harmonic frequencies and independent
optical wavelengths. The line-rate speed was successfully extended to 160 Gb/s
using the scaled clock recovery technique. Detailed modeling was presented for
this scaled technique, with emphasis on how the locking dynamics scales with
the number of multiplexed channels. The results showed that a proper
CHAPTER 7: Summary and Future Work
309
demultiplexing window width is critical to ensure the optimal lock-in range and
detuning should be kept low for a better lock-in time and stability.
To achieve an even more compact clock recovery sub-system, the
injection locking clock recovery technique was also studied, where the TW-
EAM is used as part of the microwave ring oscillator (property (A)) and the
photocurrent signal generated by the TW-EAM (property (B)) is utilized to
injection-lock the oscillator. Only an electrical amplifier and a microwave band-
pass filter were need in addition to the TW-EAM, which resulted in a very
compact configuration. Detailed characterization was carried out at 10 Gb/s,
which showed very effective jitter suppression capability. This compact clock
recovery approach was utilized in subsequent 3-R regenerators.
For WDM applications, compact sub-systems for wavelength conversion
and 3-R regeneration were realized by using all four device level properties.
These sub-systems are intended to work at the ETDM base-rate with very
compact configurations in order to provide cost-effective solutions in high-
channel-count WDM systems. By combining properties (A) and (B), a novel
photocurrent-assisted wavelength converter (PAW-Converter) was proposed and
demonstrated, which also provides additional signal monitoring capability. The
existence of the photocurrent-assisted mechanism inside the TW-EAM was
carefully verified with several experiments.
CHAPTER 7: Summary and Future Work
310
The conversion speed of PAW-Converter for NRZ signal was increased
from 2.5 Gb/s to 10 Gb/s by detailed optimizations in active material and
microwave termination. Power penalties as low as 0.2 dB was obtained at 10
Gb/s and operation in the entire C-band was demonstrated. A first-order model
was developed and the cascadibility was studied with numerical simulations. It
was found that cascadibility of NRZ optical wavelength conversion is ultimately
limited by the finite rise and fall times.
To improve the performance of RZ signal wavelength conversion, RF-
driven PAW-Conversion was proposed, which successfully reduced the power
penalty from 7.5 dB down to 1.0 dB even with the original active material.
Regenerative capabilities such as re-shaping and re-timing were also found in
RF-driven PAW-Conversion so that it is also denoted as PAW-Regeneration. To
make the regenerator even smaller, RF-driven PAW-Conversion was merged
with the injection locking clock recovery. The result was an extremely compact
3-R PAW-Regenerator, where the three required functionalities for RZ signal
regeneration (clock recovery, pulse generation, and non-linear gating) were
implemented simultaneously with a single TW-EAM. A 1-dB negative power
penalty compared with the degraded input signal was obtained. The performance
limitations were studied in detail by numerical simulations. Some of the
limitations can be mitigated by increasing the detection speed of the TW-EAM
but others are more intrinsic, which leads to a reduced regeneration capability.
CHAPTER 7: Summary and Future Work
311
To obtain stronger regeneration, a simplified optoelectronic 3-R
regenerator was proposed and demonstrated, where the nonlinear E-O
transformation of the TW-EAM (property (C)) was used to replace the
electronic signal processing circuits of conventional O-E-O regenerators. All the
benefits of the optoelectronic approach were preserved but with a reduced
complexity and better efficiency. Regeneration of ASE degraded signals was
demonstrated with up to 1.6 dB of negative power penalty. This simplified
optoelectronic 3-R regenerator not only can regenerate the signal on the same
wavelength but also can operate as a regenerative wavelength converter.
Extensive numerical simulations were carried out which showed that the
cascadibility as a regenerative wavelength converter is excellent and not limited
by the waveform degradation as in the NRZ case. Strong regeneration
capabilities on amplitude noise and timing jitter degraded signals were also
demonstrated by numerical simulations.
The merit of the proposed optoelectronic regenerator was increased
further by using the integration technology developed in recent years. An
integrated version of the simplified optoelectronic 3-R regenerator was
demonstrated by using a monolithically integrated transmitter consisting of a
tunable SGDBR laser, a semiconductor optical amplifier, and two TW-EAMs.
The size of the regenerator was significantly reduced. In addition, the injection
locking clock recovery was also used in the demonstration so that only electrical
CHAPTER 7: Summary and Future Work
312
amplifiers and a passive microwave filter were required from electronics in
order to build the regenerator. Effective regeneration of ASE degraded signal
was demonstrated with up to 2.3 dB negative power penalty.
7.2 Suggestions for future research
7.2.1 Techniques for higher OTDM line-rate speeds
The standing-wave enhanced mode of the TW-EAM has been
instrumental to enhanced 40-GHz gating operations that include optical pulse
generation, 160- to 40-Gb/s optical demultiplexing, and 160-Gb/s add-drop
multiplexing with a 40-Gb/s base-rate. As shown in Chapter 2, even though the
microwave distribution can be optimized with a proper length of termination
CPW line, the impedance seen at the input port of the TW-EAM may not be
optimal for efficient microwave coupling. Therefore, an impedance matching
network at the input port should be implemented to improve the microwave
coupling efficiency. Due to the narrow-band nature of the gating operation,
impedance matching can be achieved by using single or double stubs. Similar
research was reported in Ref. [1] for lumped EAMs. This would be crucial for
TW-EAMs to generate shorter pulses and to reduce the power fluctuations
among the through channels in add-drop multiplexing.
CHAPTER 7: Summary and Future Work
313
One feature of electrically-driven TW-EAM optical gate is that the
OTDM line-rate speed scales with the advance in electronics. Currently, a 4:1
multiplexing factor can be achieved with a 40-Gb/s base-rate for gating
operations such as demultiplexing and add-drop multiplexing. This means a
line-rate speed of 160-Gb/s. If impedance matching can be implemented for the
standing-wave enhanced mode, the simulation results in Figure 2.13 suggested
that the gating window width and output power can both be improved and 320-
to 40-Gb/s demultiplexing should be feasible. However, for add-drop
multiplexing, impedance matching may only improve the cascadibility of 160-
Gb/s operation with less power variations among the through channels. For 320-
Gb/s operation, the gating windows may be marginal unless the transfer function
of the TW-EAM can also be improved through material and device engineering
so that an ideal step-like transfer function with higher modulation efficiency can
be obtained.
Another solution for a higher line-rate speed is to add the 80-GHz
harmonic frequency to the 40-GHz driving signal, analogous to the single-stage
160- to 10-Gb/s demultiplexer in Chapter 3, where 10- and 20-GHz harmonics
were added to shape the electrical driving signal. Over 160-Gb/s add-drop
multiplexing and over 320-Gb/s demultiplexing should be feasible this way.
Surely this is strongly dependent on the available electronics and the response of
the TW-EAM at 80 GHz.
CHAPTER 7: Summary and Future Work
314
7.2.2 Improved QW design for better performances
For the wavelength converters and 3-R regenerators proposed in this
dissertation, the step-like nonlinear E-O transfer function is critical to provide
vertical re-shaping for better performances. In addition, for the PAW-family
sub-systems, the detection speed is also essential for improved speed and
reduced pattern dependence. These all require further works on the design of
new materials. In general, the optimization of QWs is complicated and most of
the properties are correlated. Trade-offs may occur as in the case for saturation-
based EAM wavelength converters. Fortunately, the material requirements for a
faster PAW-Converter are in the same direction as for high saturation-power
EAMs and no obvious trade-off was observed. Techniques like shallow QWs
and reduced bandgap offset can be used. The step-like shape may involve red-
shift of the absorption edge. A recent development in intra-step-barrier QWs
may suggest a direction to meet both saturation and shape requirements on the
E-O transfer function [2].
7.2.3 Monolithic and hybrid integrations for higher speed
For the optoelectronic 3-R regenerators proposed in this dissertation, one
major advantage is that their operation speed scales with the advances in
electronics. The proposed approaches using the nonlinear E-O transformation is
particularly valuable to relax the burden on electronics when the operating speed
CHAPTER 7: Summary and Future Work
315
is close to the limits. Even though the bandwidth demonstrated in this
dissertation is only 10 Gb/s due to the bandwidth of the components available,
scaling the operation up to 40 Gb/s and beyond should be feasible given the
recent developments in TW-EAMs and the bandwidth of microwave amplifiers.
The integrated 3-R regenerator is demonstrated for a single wavelength channel
in this dissertation. The concept would be even more competitive economically
if it can be integrated monolithically to a large scale (higher capacity per chip).
In addition, hybrid integration with associated microwave components would be
necessary for high-speed operations beyond 10 Gb/s.
References
[1] T. Kuri, K. Kitayama, and Y. Takahashi, “60-GHz-Band Full-Deplex Radio-On-Fiber System Using Two-RF-Port Electroabsorption Transceiver”, IEEE Photon. Technol. Lett., vol. 12, no. 4, pp. 419-421, 2000
[2] D. S. Shin, W. X. Chen, Y. Zhuang, Y. Wu, S. A. Pappert, D. Chow, D. Yap, P. Deelman, P. K. L. Yu, “Suppressing electroabsorption with intra-step-barrier quantum wells for high-power electroabsoption modulators”, Electron. Lett., vol. 38, no. 19, pp. 1140-1142, 2002