A thesis submitted in fulfilment of the requirement for the degree of Doctor of Philosophy

School of Electrical Engineering and Telecommunications The University of New South Wales

November 2010 THE UNIVERSITY OF NEW SOUTH WALES Thesis/Dissertation Sheet

Surname or Family name: JI

First name: PHILIP Other name/s: NAN

Abbreviation for degree as given in the University calendar: PhD

School: ELECTRICAL ENGINEERING AND TELECOMMUNICATIONS Faculty: ENGINEERING

Title: DEVELOPMENT OF NOVEL FIBRE OPTIC DEVICES AND SUBSYSTEMS FOR NEXT GENERATION DWDM SYSTEMS

Abstract

As the backbone for the global communication network, optical dense wavelength division multiplexed (DWDM) systems are facing challenges in capacity, flexibility, reliability and cost effectiveness.

In my thesis research I developed five novel optical devices or subsystems to combat these technical challenges. Each of these devices/subsystems is described in an individual chapter including background survey, proposal of new features, theoretical analysis, hardware design, prototype fabrication and characterization, and experimental verification in DWDM systems.

The first is a novel tunable asymmetric interleaver that allows the interleaving ratio to be adjusted dynamically. Two design methods were proposed and implemented. Spectral usage optimisation and overall system performance improvement in 10G/40G and 40G/100G systems were successfully demonstrated through simulations and experiments.

The second is a colourless intra-channel optical equalizer. It is a passive periodic filter that restores the overall filter passband to a raised cosine profile to suppress the filter narrowing effect and mitigate the inter-symbol interference. 20% passband widening and 40% eye opening were experimentally achieved.

The third is a flexible band tunable filter that allows simultaneous tuning of centre frequency and passband width. Based mainly on this filter, a low cost expendable reconfigurable optical add/drop multiplexer (ROADM) node was developed. Its flexible switching features were experimentally demonstrated in a two-ring four-node network testbed.

The fourth is a transponder aggregator subsystem for colourless and directionless multi-degree ROADM node. Using the unique characteristics of the coherent receiver, this technology eliminates the requirement of wavelength selector, thus reduces power consumption, size and cost. I experimentally demonstrated that it can achieve < 0.5 dB penalty between receiving single channel and 96 channels.

The last is a real-time feedforward all-order polarization mode dispersion (PMD) compensator. It first analyses spectral interference pattern to retrieve phase information and calculate PMD, then it uses a pulse shaper to restore the pulse shape and thus compensates the PMD. These functions were demonstrated through experiments and simulations.

All of these novel devices and subsystems deliver new functional features and are suitable to be applied in the next generation DWDM systems to improve capacity, flexibility, and reliability and to reduce cost.

Declaration relating to disposition of project thesis/dissertation

I hereby grant to the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or in part in the University libraries in all forms of media, now or here after known, subject to the provisions of the Copyright Act 1968. I retain all property rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation.

I also authorise University Microfilms to use the 350 word abstract of my thesis in Dissertation Abstracts International (this is applicable to doctoral theses only).

…………………………………………………………… ……………………………………..……………… ……….……………………...…….… Signature Witness Date

The University recognises that there may be exceptional circumstances requiring restrictions on copying or conditions on use. Requests for restriction for a period of up to 2 years must be made in writing. Requests for a longer period of restriction may be considered in exceptional circumstances and require the approval of the Dean of Graduate Research.

FOR OFFICE USE ONLY Date of completion of requirements for Award: ORIGINALITY STATEMENT

‘I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgement is made in the thesis. Any contribution made to the research by others, with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project's design and conception or in style, presentation and linguistic expression is acknowledged.’

Signed ……………………………………………......

Date ……………………………………………......   

This thesis would not have been possible without the guidance and the help of several individuals who in one way or another contributed and extended their valuable assistance in the preparation and completion of this study:

First and foremost, the late Professor Pak-Lim Chu, for leading me into this wonderful world of fibre optics, and for being my mentor and being the godly role model that I always look up to.

My supervisor, Professor Gang-Ding Peng, for all the precious guidance, advice and instruction throughout my entire post-graduate study.

My co-supervisor, Dr. Ting Wang, for giving me the great opportunities to explore many new and exciting research fields, and for the invaluable guidance and support during my research.

All my fellow students and staffs at the Photonics and Optical Communications Group, University of New South Wales, for the assistance and encouragement throughout my long PhD pursuit.

All my colleagues at NEC Laboratories America, for the fruitful collaborations and all the important technical supports in this research.

My late parents, for their constant prayers and encouragement, and for their life-long sacrifice. They have always been my source of motivation and strength in the pursuit of knowledge and truth.

My dearest wife, Jennifer, for her constant love and support which accompanied and sustained me in this long and challenging journey.

My two beautiful children, Jonathan and Zachary, for all the joy I received.

iv !"#$%&'%(%) !"#$%&'%(%)!"#$%&'%(%)

All my brothers and sisters in Christ from Overseas Christian Fellowship Australia, West Sydney Chinese Christian Church, and Princeton Christian Church, for their prayers, fellowship and encouragement that upheld me during my study.

Most importantly, my Heavenly Father, my Lord and Saviour, the Wonderful Counsellor, the Creator of the heavens and the earth, and the Creator of the ultimate broadband communication network, for the blessings, grace, mercy, guidance, and provision, throughout this research and throughout my entire life. May all glory and honour be unto Him.

The heavens declare the glory of God, and the sky above proclaims his handiwork. Day to day pours out speech, and night to night reveals knowledge. There is no speech, nor are there words, whose voice is not heard. Their voice goes out through all the earth, and their words to the end of the world. … Let the words of my mouth and the meditation of my heart be acceptable in your sight, O Lord, my rock and my redeemer. Psalm 19:1-4, 14 (ESV)

v  

As the backbone for the global communication network, optical dense wavelength division multiplexed (DWDM) systems are facing challenges in capacity, flexibility, reliability and cost effectiveness. In my thesis research I developed five novel optical devices or subsystems to combat these technical challenges. Each of these devices/subsystems is described in an individual chapter including background survey, proposal of new features, theoretical analysis, hardware design, prototype fabrication and characterization, and experimental verification in DWDM systems. The first is a novel tunable asymmetric interleaver that allows the interleaving ratio to be adjusted dynamically. Two design methods were proposed and implemented. Spectral usage optimisation and overall system performance improvement in 10G/40G and 40G/100G systems were successfully demonstrated through simulations and experiments. The second is a colourless intra-channel optical equalizer. It is a passive periodic filter that restores the overall filter passband to a raised cosine profile to suppress the filter narrowing effect and mitigate the inter-symbol interference. 20% passband widening and 40% eye opening were experimentally achieved. The third is a flexible band tunable filter that allows simultaneous tuning of centre frequency and passband width. Based mainly on this filter, a low cost expendable reconfigurable optical add/drop multiplexer (ROADM) node was developed. Its flexible switching features were experimentally demonstrated in a two- ring four-node network testbed. The fourth is a transponder aggregator subsystem for colourless and directionless multi-degree ROADM node. Using the unique characteristics of the coherent receiver, this technology eliminates the requirement of wavelength selector, thus reduces power consumption, size and cost. I experimentally demonstrated that it can achieve < 0.5 dB penalty between receiving single channel and 96 channels. The last is a real-time feedforward all-order polarization mode dispersion (PMD) compensator. It first analyses spectral interference pattern to retrieve phase information and calculate PMD, then it uses a pulse shaper to restore the pulse shape and thus

vi *+),!)*+),!)*+),!)

compensates the PMD. These functions were demonstrated through experiments and simulations. All of these novel devices and subsystems deliver new functional features and are suitable to be applied in the next generation DWDM systems to improve capacity, flexibility, and reliability and to reduce cost.

vii   

ADC Analogue-to-digital converter AOM Acousto-optical modulator ASE Amplified spontaneous emission AWG Arrayed waveguide grating AWGN Additive white Gaussian noise B&S Broadcast and select CAPEX Capital expense CD Chromatic dispersion CL&DL Colourless and directionless CMRR Common mode rejection ratio CS-RZ Carrier-suppressed return-to-zero CW Continuous wave CWDM Coarse wavelength division multiplexing DB Duobinary DCF Dispersion compensation fibre DCM Dispersion compensation module DFB Distributed feedback DFE Decision feedback equalizer DGD Differential group delay DGE Dynamic gain equalizer DM Deformable mirror DMD Differential mode dispersion DOP Degree of polarization DPSK Differential phase shift keying DQPSK Differential quadrature phase shift keying DSM Demultiplexer-switch-multiplexer DSP Digital signal processing DWDM Dense wavelength division multiplexing ECL External cavity laser EDC Electronic dispersion compensator EDFA Erbium-doped fibre amplifier EMD Empirical mode decomposition ENOB Effective number of bits FBG Fibre Bragg grating

viii #++, #- %,(+ #++, #- %,(+#++, #- %,(+

FBTF Flexible band tunable filter FFE Feed-forward equalizer F-P Fabry-Perot FSR Free spectral range GMPLS Generalized multi-protocol label switching G-T Gires-Tournois GUI Graphical user interface HT Hilbert transform IMF Intrinsic mode function IP Internet protocol ISI Inter-symbol interference ITO Indium tin oxide ITU International Telecommunications Union ITU-T ITU’s Telecommunication Standardization Sector LC Liquid crystal LCoS Liquid crystal on silicon LED Light emitting diode LO Local oscillator M-ASK Multilevel amplitude shift keying MGT Michelson Gires-Tournois MI Michelson interferometer MZI Mach-Zehnder interferometer NRZ Non-return-to-zero OADM Optical add/drop multiplexer OEO Optical-electrical-optical OEQ Optical equalizer OFDM Orthogonal frequency division multiplexing OOK On-off keying OPEX Operation expense OSA Optical spectrum analyser OSNR Optical signal-to-noise ratio OTDM Optical time domain multiplexing OXC Optical cross-connect PCB Printed circuit board PD Photodetector PDL Polarization dependent loss PDM Polarization division multiplexing PLC Planar lightwave circuit

ix #++, #- %,(+ #++, #- %,(+#++, #- %,(+

PM Phase modulator, or polarization maintaining PMD Polarization mode dispersion PolSK Polarization shift keying PPG Pulse pattern generator PRBS Pseudo-random bit stream/sequence p.s.d. Power spectral density PSK Phase shift keying PSP Principal states of polarization QAM Quadrature amplitude modulation QPSK Quadrature phase shift keying RMS Root mean square ROADM Reconfigurable optical add/drop multiplexer RZ Return-to-zero SINR Signal-to-interference-and-noise ratio SLM Spatial light modulator SMF Single mode fibre SOP State of polarization SSB Single-side-band SSE Spontaneous source emission TA Transponder aggregator TAI Tunable asymmetric interleaver TCP Transmission Control Protocol TDM Time division multiplexing TEC Thermal-electrical cooler TEF Tunable edge filter TFF Thin film filter TIA Trans-impedance amplifier TOAD Tunable optical add/drop TODC Tunable optical dispersion compensator VOA Variable optical attenuator VSB Vestigial-side-band WB Wavelength blocker WDM Wavelength division multiplexing WSS Wavelength-selective switch WXC Wavelength cross-connect

x    

The work presented in this thesis is published in the following papers and patents:

Referred international conference and journal papers:

• P. N. Ji, T. Wang and L. Zong, “ROADM for Metro DWDM Network Using Wavelength Selective Devices”, Proc. ATFO, pp. 139-150, 2004

• L. Zong, P. N. Ji, et al., “Testbed for ROADM and WXC Based Metro WDM Networks”, in Network Architectures, Management, and Applications III, K.-W. Cheung, G.-K. Chang, G. Li and K. Sato (eds), Proc. SPIE, Vol. 6022, pp. 379-385, 2005

• P. N. Ji, A. Dogariu, et al., “Optical Tunable Asymmetric Interleaver”, Technical Digest of OFC-NFOEC 2006, OTuM7, 2006

• P. N. Ji, L. Zong, L. Xu and T Wang, “Passband Width Tunable Filter and Its Application in Low Cost ROADM Node”, Proc. 5th ICOCN/ 2nd ATFO, pp. 44-48, 2006

• P. N. Ji, L. Xu, et al., “Inter-Symbol Interference Comparison for Wavelength and Waveband Switching in All-Optical Optical Cross-Connect Nodes”, Proc. OSA Photonics in Switching 2006, O.14.4, pp. 187-189, 2006

• P. N. Ji, J. Yu, et al., “Waveband Deaggregator Cascading Effect in Multi- Granularity Optical Cross-Connect Nodes”, Proc. 19th LEOS, WW4, 2006

• A. Dogariu, P. N. Ji and T Wang, “All-Order PMD Compensation Using Spectral Interference and Pulse Shaping”, Proc. IEEE Sarnoff Symposium 2007, pp. 1-4, 2007

• A. Dogariu, P. N. Ji, and T. Wang, “Real-Time All-order PMD Compensation Using Spectral Interference”, Proc. IEEE Sarnoff Symposium 2008, PP. 1-4, 2008

• P. N. Ji, J. Yu, D. Qian and T. Wang, “Spectrally Efficient 3.4b/s/Hz Waveband Switching over Wavelength-Selective Switch-Based ROADM on 114 Gb/s PolMux-

xi .*!)#+ .*!)#+.*!)#+

RZ-8PSK DWDM System”, in Optical Transmission, Switching and Subsystems VI, K. Kitayama, P. C. Ghiggino, et al. (eds), Proc. SPIE, Vol. 7136, 71361X, 2008

• P. N. Ji, M.-F. Huang, J. Yu and T. Wang, “Reconfigurable Waveband Cross- Connect and Its Application in 112 Gb/s WDM System”, Technical Digest of OFC- NFOEC 2009, OTuF2, 2009

• A. Dogariu, P. N. Ji, L. Cimponeriu and T. Wang, “PMD Compensation via Real- Time Phase Retrieval from Spectral Interference”, Optics Communications, Vol. 282, No. 18, pp. 3706-3711, 2009

• P. N. Ji, and Y. Aono, “Colorless and Directionless Multi-Degree Reconfigurable Optical Add/Drop Multiplexers,” Proc.19th WOCC, OA1, 2010

• P. N. Ji, Y. Aono and T. Wang, “Reconfigurable Optical Add/Drop Multiplexer Based on Bidirectional Wavelength Selective Switches”, Proc. Photonics in Switching 2010, PWB1, 2010

• P. N. Ji, J. Yu, G.-D. Peng and T. Wang, “Passband Optimisation for Hybrid 40G/100G System Using Tunable Asymmetric Interleaver”, Technical Digest of ECOC 2010, Mo.1.A.3, 2010

Issued patents and pending patent applications:

• T. Wang, P. N. Ji, et al., “Flexible Band Tunable Add/Drop Multiplexer and Modular Optical Node Architecture”, US patent application US2005/0281558, published December 2005

• T. Wang, P. N. Ji, et al., “Flexible Band Tunable Filter”, US patent US7099529, issued August 2006

• P. N. Ji, T. Wang, et al., “Optical Tunable Asymmetric Interleaver and Upgrade for Dense Wavelength Division Multiplexed Networks”, US patent US2007/0116468, published May 2007

• T. Wang, P. N. Ji, et al., “Flexible Waveband Aggregator and Deaggregator and Hierarchical Hybrid Optical Cross-Connect System”, US patent US7239772, issued July 2007

xii .*!)#+ .*!)#+.*!)#+

• T. Wang, L. Xu, P. N. Ji, et al., “Optical Equalization Filtering of DWDM Channels”, US patent application US2007/0206898, published September 2007

• A. Dogariu, P. N. Ji, et al., “All Order Polarization Mode Dispersion Compensation with Spectral Interference Based Pulse Shaping”, US patent application US2008/0002972, published January 2008

• P. N. Ji, L. Xu, et al., “Inter-Symbol Interference-Suppressed Colorless DPSK Demodulator”, US patent application US2008/0240736, published October 2008

• P. N. Ji, L. Xu, et al., “Intra-Channel Equalizing Optical Interleaver for DWDM Communication Systems”, US patent application US2009/0162066, published June 2009

• P. N. Ji, Y. Aono, et al., “Transponder Aggregator Without Wavelength Selector for Colorless and Directionless Multi-Degree ROADM Node”, US patent application 12-900220, published October 2009

• P. N. Ji, Y. Aono, et al., “Real-Time Receiver Optimization in Filter-Less Coherent Receiving System”, US provisional patent application 61/332390, filed April 2010

xiii     

Acknowledgements ...... iv Abstract ...... vi Glossary of Terms ...... viii Publications ...... xi Table of Contents ...... xiv List of Figures ...... xx List of Tables ...... xxvi

Chapter 1. Introduction ...... 1 1.1 Background of the Research ...... 1 1.1.1 Brief History of Fibre Optic Communications...... 1 1.1.2 Demands of Fibre Optic Communication Systems ...... 3 1.1.3 DWDM Fibre Optic Communication System Technologies ...... 4 a. System Configuration and Architecture ...... 4 b. Subsystems / Devices ...... 5 1.1.4 Key Limiting Factors and Mitigation Techniques ...... 7 1.2 Objectives of the Research ...... 9 a. Targeted Problems ...... 9 b. Research Topics ...... 10 1.3 Thesis Outline ...... 11

Chapter 2. Development and Experimental Demonstration of Tunable Asymmetric Interleaver ...... 13 2.1 Background of the Study ...... 13 2.1.1 Capacity Upgrade in DWDM Systems ...... 13 2.1.2 Optical Interleaver ...... 15 a. Optical Interleaver and Its Functions ...... 15 b. Technologies and Implementations of Conventional Optical Interleaver. 18 2.1.3 DWDM Systems Upgrade Using Optical Interleaver ...... 18

xiv *% #- #)%)+ *% #- #)%)+ *% #- #)%)+

2.1.4 Issues of DWDM System Upgrade Using Conventional Symmetric Optical Interleaver ...... 20 2.2 Development and Implementation of Tunable Asymmetric Interleaver ...... 23 2.2.1 Target Functions of Tunable Asymmetric Interleaver ...... 24 2.2.2 TAI Design 1: Based on Cascaded Symmetric Interleavers ...... 25 a. Design Principle ...... 25 b. Implementation ...... 27 c. Characterization of Optical Properties ...... 33 d. Characterization of Asymmetry Tuning ...... 36 2.2.3 TAI Design 2: Based on Programmable Optical Processor ...... 41 a. Design Principle ...... 41 b. Implementation and Asymmetric Passband Profiles ...... 42 2.2.4 Comparison between the Two TAI Designs ...... 45 a. Insertion Loss ...... 45 b. Polarization Dependent Loss ...... 45 c. Chromatic Dispersion ...... 46 d. Differential Group Delay ...... 47 e. Tuning Speed ...... 47 f. Passband Profiles ...... 47 g. Control Complexity ...... 48 h. Hardware size ...... 48 i. Cost ...... 48 2.3 Experimental Demonstration of TAI ...... 48 2.3.1 Simulation of TAI for 10G/40G Hybrid System Upgrade ...... 48 a. Simulation Tool ...... 48 b. Simulation Model and Parameters ...... 49 c. Simulation Results and Analysis ...... 52 2.3.2 Experimental Demonstration of 40G/100G Hybrid System Upgrade ...... 55 a. Experimental Setup ...... 55 b. Experiment Results and Analysis ...... 58 2.4 Conclusions ...... 59

xv *% #- #)%)+ *% #- #)%)+ *% #- #)%)+

Chapter 3. Development and Experimental Demonstration of Colourless Optical Intra-Channel Equalizer ...... 61 3.1 Background of the Study ...... 61 3.1.1 ISI in DWDM Transmission ...... 61 3.1.2 ISI Mitigation Methods ...... 66 a. Increase Signal’s Spectral Efficiency ...... 66 b. Digital Coding Techniques ...... 67 c. Pre-filtering ...... 67 d. Signal Equalization ...... 67 3.2 Development and Implementation of Colourless Optical Intra-Channel Equalizer ...... 68 3.2.1 Operation Principle ...... 68 a. Signal Equalization Using Linear Filters ...... 68 b. Colourless OEQ for ISI Suppression in Multiple DWDM Channels ...... 70 3.2.2 Device Design ...... 72 a. Basic Design: Fabry-Perot Interferometer ...... 72 b. Parameter Design and Verification ...... 74 c. Design Optimisations ...... 78 I. Dip Depth Variation ...... 78 II. Passband Profile Symmetry ...... 79 III. Effect of Frequency Offset ...... 80 IV. Location of Equalization ...... 81 V. Optimum Power ...... 82 3.2.3 Colourless OEQ Prototype and Characterization ...... 83 3.3 Experimental Demonstration ...... 86 3.3.1 Passband Widening Measurement ...... 86 3.3.2 Back-to-Back Experiment ...... 88 3.3.3 Transmission Experiment ...... 91 3.4 Conclusions ...... 93

Chapter 4. Development and Experimental Demonstration of Flexible Band Tunable Filter and Its Application in Low Cost and Expendable ROADM Node ...... 94

xvi *% #- #)%)+ *% #- #)%)+ *% #- #)%)+

4.1 Background of the Study ...... 94 4.2 Development and Implementation of Flexible Band Tunable Filter ...... 98 4.2.1 Flexible Band Tunable Filter and Its Functions ...... 98 4.2.2 Design of FBTF ...... 99 4.2.3 FBTF Prototype ...... 103 a. Dielectric Thin Film F-P Interferometric Filter ...... 103 b. Tuning Mechanism ...... 104 c. Prototype Construction ...... 105 4.2.4 FBTF Prototype Characterization ...... 107 4.3 Experimental Demonstration of Low Cost and Expendable ROADM Node Based on FBTF ...... 110 4.3.1 ROADM Node Architecture ...... 110 a. Motivations ...... 110 b. Node Architecture ...... 112 c. Modules for the ROADM Node ...... 112 d. Node Features and Benefits ...... 115 4.3.2 FBTF-Based ROADM Node Prototype and Network Testbed ...... 117 a. Network Testbed ...... 117 b. Design and Construction of Modules ...... 118 I. Express Module ...... 119 II. Tunable Optical Add/Drop Module ...... 121 III. OXC Module ...... 123 IV. External Hardware ...... 124 c. ROADM Control Configuration and Software ...... 127 I. Communication between Controller Computer and the ROADM Modules ...... 127 II. Management Software ...... 128 4.3.3 FBTF-Based ROADM Experiment and Results ...... 131 a. Experiment Setup ...... 131 b. Network Operation Experiments ...... 134 c. Signal Quality Experiments ...... 138 4.4 Conclusions ...... 141

xvii *% #- #)%)+ *% #- #)%)+ *% #- #)%)+

Chapter 5. Development of Transponder Aggregator without Wavelength Selector ...... 143 5.1 Background of the Study ...... 143 5.1.1 Colourless and Directionless ROADM ...... 143 5.1.2 Wavelength Selection Methods in Transponder Aggregator ...... 145 5.2 Development of Transponder Aggregator without Wavelength Selector ...... 146 5.2.1 Operation Principle ...... 146 5.2.2 Theoretical Analysis ...... 148 a. Signal ...... 149 b. Interference ...... 150 c. Noise ...... 151 d. Signal-to-Interference-and-Noise Ratio ...... 152 5.2.3 Simulation ...... 153 5.2.4 Design Software ...... 155 5.3 Experimental Demonstration of Transponder Aggregator without Wavelength Selector ...... 157 5.3.1 Experiment Setup ...... 157 5.3.2 CMRR Adjustment and Measurement Technique ...... 159 5.3.3 Experiment Results and Analysis ...... 163 5.4 Conclusions ...... 168

Chapter 6. Development of a Feedforward Real-Time Technique for All-Order PMD Measurement and Compensation ...... 170 6.1 Background of the Study ...... 170 6.1.1 PMD Issue in DWDM Transmission ...... 170 6.1.2 Current PMD Measurement and Compensation Methods ...... 171 6.2 Feedforward Real-Time All-Order PMD Measurement and Compensation Technique ...... 172 6.2.1 PMD Model and Theory ...... 172 6.2.2 PMD Measurement and Compensation System ...... 176 6.2.3 Step 1: PMD Measurement by All-Order Phase Retrieval ...... 177 6.2.4 Step 2: PMD Compensation by Pulse Shaping ...... 180

xviii *% #- #)%)+ *% #- #)%)+ *% #- #)%)+

6.3 Experimental Demonstration of Feedforward Real-Time All-Order PMD Measurement and Compensation Technique ...... 182 6.3.1 First Order PMD Measurement Experiment ...... 182 6.3.2 Higher Order PMD Measurement Demonstration ...... 185 6.3.3 Pulse Shaping Experiment ...... 187 6.4 Conclusions ...... 191

Chapter 7. Conclusions ...... 193 7.1 Thesis Summary ...... 193 7.2 Future Research ...... 195

References ...... 196

xix    

Figure 1.1 Schematic of a DWDM transmission system ...... 5 Figure 1.2 Application locations of the five subsystems/devices studied in this thesis ...... 10 Figure 2.1 Functions of optical interleaver ...... 16 Figure 2.2 Interleavers in cascade ...... 16 Figure 2.3 Operation of a 4×4 cyclic interleaver ...... 16 Figure 2.4 100 GHz-spaced DWDM transmission link ...... 19 Figure 2.5 100 GHz to 50 GHz DWDM network upgrade using interleaver ...... 19 Figure 2.6 Interleaver profiles ...... 21 Figure 2.7 Spectra of 40 Gbit/s signal under several types of modulation schemes ...... 22 Figure 2.8 Compatibility of various modulation formats with 50GHz interleaver passband ...... 22 Figure 2.9 Fixed asymmetric interleaver constructed by cyclic interleavers ...... 24 Figure 2.10 Function of tunable asymmetric interleaver ...... 25 Figure 2.11 Design of an asymmetric interleaver ...... 26 Figure 2.12 Tunable asymmetric interleaver using cascaded shift symmetric interleavers ...... 27 Figure 2.13 Schematic of a Michelson interferometer ...... 28 Figure 2.14 Periodic outputs of Michelson interferometer ...... 29 Figure 2.15 Schematic of a Gires-Tournois etalon ...... 30 Figure 2.16 Schematic of a Michelson-Gires-Tournois interferometer-based interleavers ...... 32 Figure 2.17 Spectra of the individual interleaver components and combined tunable asymmetric interleaver at room temperature...... 34 Figure 2.18 Passband edge comparison ...... 35 Figure 2.19 Temperature control feedback circuit for the TAI ...... 37 Figure 2.20 Passband profiles of TAI ...... 38 Figure 2.21 Asymmetry tuning at different temperatures ...... 39 Figure 2.22 Temperature response of the TAI unit ...... 40

xx +) #- +) #- +) #- '.,%+ '.,%+ '.,%+

Figure 2.23 A passband from the TAI’s Odd Output at interleaving ratios from 10% to 90% in 5% increments ...... 42 Figure 2.24 Passbands of the Odd Output and Even Output at interleaving ratios from 10%:90% to 90%:10% in 10% increments ...... 43 Figure 2.25 PDL of the LCoS-based TAI at different interleaved ratios with respect to different clear channel passbands ...... 44 Figure 2.26 PDL profiles of TAIs constructed ...... 46 Figure 2.27 CD profiles of TAIs constructed ...... 46 Figure 2.28 DGD profiles of TAIs constructed ...... 47 Figure 2.29 Simulation model of the 10G/40G DWDM transmission system ...... 50 Figure 2.30 Simulated optical spectra with various types of modulation schemes ... 51 Figure 2.31 Eye closure values of signals with different modulation schemes as a function of interleaving ratio for the 40 Gbit/s channel ...... 52 Figure 2.32 Q-factor values of signals with different modulation schemes as a function of interleaving ratio for the 40 Gbit/s channel ...... 52 Figure 2.33 Experimental setup of TAI in 40G/100G hybrid system ...... 56 Figure 2.34 OSNR requirement as a function of interleaver passband width ...... 58 Figure 3.1 Measured transmission characteristics of an optical interleaver ...... 63 Figure 3.2 Simulated optical spectra and eye diagrams of 43 Gbit/s 33% RZ-DPSK signals under different filtering conditions ...... 64 Figure 3.3 Passband narrowing by the cascading ROADM filters ...... 65 Figure 3.4 Eye closure of 40G DPSK signal caused by ISI at different number of cascading 50 GHz WSS filters in the transmission link ...... 66 Figure 3.5 Equalization filter for ISI suppression ...... 69 Figure 3.6 Optical equalization for optical DWDM systems ...... 69 Figure 3.7 Optical spectrum and eye diagram of 43 Gbit/s 33% RZ-DPSK signals passing through optical multiplexer, demultiplexer and OEQ filter in a 50 GHz-spaced DWDM system ...... 70 Figure 3.8 A periodic comb filter which can be used as an optical equalization filter for simultaneous ISI suppression in multiple DWDM channels ...... 71 Figure 3.9 Reducing OEQ quantity by using colourless OEQ for all channels ...... 71 Figure 3.10 Model of a Fabry-Perot interferometer ...... 73

xxi +) #- +) #- +) #- '.,%+ '.,%+ '.,%+

Figure 3.11 Light intensity transmission curve and phase change under different mirror transmission coefficients ...... 74 Figure 3.12 Optical characteristics of optical devices used in the simulation ...... 75 Figure 3.13 VPI Simulation model for DWDM transmission link with colourless intra-channel OEQ ...... 76 Figure 3.14 Simulated eye diagrams of the received 42.8 Gbit/s RZ DPSK signals 77 Figure 3.15 Effect of OEQ dip depth ...... 79 Figure 3.16 Passband profiles for the symmetric OEQ and half OEQ ...... 80 Figure 3.17 Effect of OEQ passband profile symmetry ...... 80 Figure 3.18 Effect of OEQ frequency offset ...... 81 Figure 3.19 Effect of OEQ location in the transmission link ...... 82 Figure 3.20 Designed profile of a colourless OEQ with 50 GHz FSR ...... 83 Figure 3.21 Mechanical drawing and photo of the constructed colourless OEQ prototype ...... 84 Figure 3.22 Measured insertion loss profile of the constructed colourless OEQ prototype ...... 85 Figure 3.23 Measured optical characteristics of the colourless OEQ prototype ...... 85 Figure 3.24 Experiment setup for the passband widening experiment ...... 86 Figure 3.25 Optical spectra of the filtering elements with and without OEQ ...... 87 Figure 3.26 42.8 Gbit/s RZ-DPSK back-to-back OEQ experiment setup ...... 88 Figure 3.27 Signal degradation due to filtering effect ...... 90 Figure 3.28 Eye opening improvement by OEQ ...... 90 Figure 3.29 Measured BER without and with the colourless OEQ under different EDC reference voltage settings after 5 span transmission over SPWV160 system ...... 92 Figure 4.1 DSM-based ROADM architectures ...... 95 Figure 4.2 B&S-based ROADM architectures ...... 96 Figure 4.3 WSS-based ROADM architecture ...... 97 Figure 4.4 Examples of the possible passband set by the FBTF ...... 99 Figure 4.5 The operation principles of flexible band tunable filter ...... 100 Figure 4.6 Example of WDM channel spectra at various points within a FBTF .. 101 Figure 4.7 FBTF using additional low λ TEF ...... 102 Figure 4.8 Schematic of a thin film FP interferometric filter ...... 103

xxii +) #- +) #- +) #- '.,%+ '.,%+ '.,%+

Figure 4.9 Mechanical angle tuning for optical FP interferometric filter ...... 104 Figure 4.10 Internal structure of a TEF in the FBTF prototype ...... 106 Figure 4.11 GUI of the FBTF control software in LabVIEW ...... 106 Figure 4.12 Example of measured insertion loss profile ...... 107 Figure 4.13 Measured insertion loss profile at various points of FBTF ...... 108 Figure 4.14 Measured PDL, DGD and CD at various ports ...... 109 Figure 4.15 More examples of centre wavelength and passband width tuning by FBTF ...... 111 Figure 4.16 Schematics of some key functional modules for the stackable ROADM node ...... 113 Figure 4.17 Experimental WDM network testbed with stackable ROADM nodes 117 Figure 4.18 The Express Module prototype ...... 119 Figure 4.19 The TOAD Module prototype ...... 122 Figure 4.20 The OXC Module prototype ...... 123 Figure 4.21 Front panel of the TOAD Module ...... 125 Figure 4.22 Example of the operation of Channel LEDs ...... 125 Figure 4.23 A complete prototype module with front panel and cover ...... 126 Figure 4.24 Two ROADM nodes with five modules in the NEC SPWV-40 shelf . 126 Figure 4.25 Network level management interface ...... 129 Figure 4.26 Node level management interfaces for Node 2 and Node 3 ...... 129 Figure 4.27 Module level management interface for Express Module, Tunable Optical Add/Drop Module, and OXC Module ...... 130 Figure 4.28 ROADM network testbed and experiment system ...... 131 Figure 4.29 Photo of the ROADM network testbed and experiment system ...... 133 Figure 4.30 Measured optical signal spectra at various points in the ROADM experiment testbed ...... 135-136 Figure 4.31 BER performance for the OC-48 SONET signal in the ROADM testbed ...... 139 Figure 4.32 Eye diagrams of the OC-192 SONET signal at BTB position and ROADM node Express Module output ...... 140 Figure 4.33 BER performance for the OC-192 SONET signal in the ROADM testbed ...... 140 Figure 5.1 A 3-degree colourless and directionless ROADM node ...... 145

xxiii +) #- +) #- +) #- '.,%+ '.,%+ '.,%+

Figure 5.2 Various channel selection methods in Transponder Aggregator ...... 146 Figure 5.3 Proposed channel separation method in the TA without wavelength selector ...... 147 Figure 5.4 Canonical quadrature of a polarization- and phase-diversity coherent receiver ...... 149 Figure 5.5 Spectrum of theoretical and simulated signal-signal interference for three example systems ...... 153 Figure 5.6 OSNR penalty for 50 GHz-spaced PDM-QPSK signals at different WDM channel numbers and different channel power levels ...... 154 Figure 5.7 OSNR penalty for 50 GHz-spaced PDM-QPSK signals at different WDM channel numbers and different CMRR settings ...... 155 Figure 5.8 Examples of the filterless ROADM receiver design software GUI ..... 156 Figure 5.9 Experimental setup of the 96-channel filterless coherent receiver experiment ...... 157 Figure 5.10 Example of CMRR frequency response under different adjustments .. 162 Figure 5.11 Spectra of received WDM channels at different numbers and the LO 164 Figure 5.12 Measured BER vs. OSNR for different CMRR conditions ...... 165 Figure 5.13 Measured and theoretical OSNR penalties for different CMRR conditions ...... 166 Figure 5.14 Measured BER vs. OSNR for different channel settings ...... 167 Figure 5.15 OSNR penalties for single-ended detector and balanced detectors with different CMRR conditions ...... 168 Figure 6.1 The random birefringent fibre as a stack of randomly oriented waveplates ...... 173 Figure 6.2 Mach-Zehnder interferometer where the phases along x and y lead to spectral modulation due to interference ...... 174 Figure 6.3 Spectral interference obtained in Ref. [149] from a Mach-Zehnder interferometer ...... 175 Figure 6.4 Schematic of all-order PMD compensation scheme ...... 176 Figure 6.5 Empirical mode decomposition of a simulated interference pattern. From top to bottom: original signal, intrinsic mode functions (IMF1-IMF6), and the signal and EMD-processed signal obtained from superposition of the modes IMF2-IMF4 ...... 180

xxiv +) #- +) #- +) #- '.,%+ '.,%+ '.,%+

Figure 6.6 Pulse shaping operation ...... 181 Figure 6.7 First order PMD measurement experiment setup ...... 183 Figure 6.8 Spectral interference patterns with 0 to 40 ps PMD delays ...... 184 Figure 6.9 Measured DGD from spectral interference as function of the group delay imposed by the PMD emulator ...... 184 Figure 6.10 Second- and third- order PMD simulated noisy interference signals and EMD-based filtered signals, the true phase used to generate the interference patterns, and the unwrapped phase estimated from the EMD- processed signals ...... 186 Figure 6.11 Demonstration of the pulse distortion due to PMD and PMD compensation using the proposed method ...... 187 Figure 6.12 Schematic of an AOM-based pulse shaper ...... 188 Figure 6.13 Mechanical drawing of the AOM device ...... 188 Figure 6.14 Experimental setup of the AOM-based pulse shaper. Yellow lines and arrows indicate the travelling path of light ...... 189 Figure 6.15 Examples of spectrum manipulation by pulse shaping ...... 190

xxv    

Table 2.1 Eye patterns, ECVs, and Q-factors for different modulation schemes at 50%:50% and 30%:70% interleaving ratios ...... 54 Table 4.1 Comparison of different technologies for tunable filter ...... 105 Table 4.2 Test channels and operations in the ROADM network operation experiment ...... 135 Table 5.1 CMRR calculation for four u2t balanced receivers ...... 161 Table 5.2 CMRR calculation for four u2t balanced receivers after adjustment ... 162

xxvi )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

)%, /)%, /  

As the backbone for the global communication network, optical fibre dense wavelength division multiplexing system is facing increasing demands for capacity, distance, flexibility, reliability and cost efficiency. The main objective of this thesis is to develop novel subsystems and devices to deal with some of the practical technical challenges. In this thesis, five subsystems or devices, namely the tunable asymmetric interleaver, colourless intra-channel equalizer, flexible band tunable filter, transponder aggregator without wavelength selector, and all-order polarization mode dispersion compensator, will be proposed and studied.

1.1 Background 1.1.1 Brief History of Fibre Optic Communications Even though the history of human communications using optical signal can be traced back to ancient time of fire beacon and smoke signals, the modern optical communication had its origin in 1960, when the laser was demonstrated for the first time. To realise feasible optical communication, two key elements are required. The first is a broad bandwidth source such as a laser source. The second element is a suitable medium which has low signal loss, allowing for information-carrying light being transmitted over sufficiently long distance. The first proposed medium – light pipe, where light rays are refocused by lenses in vacuum – is not practical. It is not economical either. The next candidate, the thin film waveguide, also has limits for long distance transmission. Optical fibre has also been considered, however the huge attenuation rate of 1000 dB/km makes it impossible to transmit beyond several meters. It was until 1966 when Charles K. Kao claimed that the loss of optical fibre could be reduced below 20 dB/km to be a telecommunications medium for inter-office communications. Together with and George Hockham, Kao demonstrated that the high-loss of existing optical fibre arose from impurities in the glass, rather than from an underlying problem with the technology itself [1].

1 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

Since then, fibre optic communication systems have revolutionised the telecommunications industry and become the technology for the global backbone network. The first generation of fibre optic communication system operates around 850 nm wavelength and uses GaAs semiconductor lasers, with bit rate around 45 Mbit/s with repeater spacing of up to 10 km. With the use of InGaAsP laser, the second generation fibre optic system operates at 1.3 μm wavelength and delivers bit rate up to 1.7 Gbit/s with repeater spacing up to 50 km. The third generation system utilizes new low loss fibre with loss of about 0.2 dB/km at 1.55 μm wavelength region and dispersion shifted fibre, and realises 2.5 Gbit/s transmission with repeater spacing exceeding 100 km. In the 1990s, wavelength division multiplexing (WDM) technology was proposed, where multiple optical channels with different wavelengths are transmitted within the same fibre simultaneously to offer many times increase in the amount of bandwidth per fibre. The development of another major technology, namely the optical amplifier, reduces the need for repeaters. In particular, the invention of erbium-doped fibre amplifier (EDFA) enables the simultaneous amplification of multiple channels within the transmission window, making the WDM system more practical and low cost. With these technologies, the system capacity of fibre optic system doubles every 6 months from 1992 until a bit rate of 10 Tbit/s was reached by 2001 [2]. In the last few years, the capacity of fibre optic communication systems continues to grow and the transmission distance continues to expand, with the technologies such as ultra-low loss fibre, advanced modulation formats, coding schemes, and impairment mitigation techniques. Recently, record capacity of 32 Tbit/s per fibre with 580 km transmission distance has been reported using polarization multiplexed 8-level quadrature amplitude modulation (8QAM) modulation format and digital coherent detection [3]. For single channel, the transmission bit rate has reached 1.07 Tbit/s through a combination of differential quadrature phase shift keying (DQPSK), optical time domain multiplexing (OTDM) and polarization multiplexing technologies [4]. Considering the capacity-distance product, 84.3 Petabit/s.km was achieved, where 135 channels of 111 Gbit/s per channel coherent orthogonal frequency division multiplexing (OFDM) signal were transmitted over 6,248 km of fibre [5].

2 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

1.1.2 Demands of Fibre Optic Communication Systems The next generation fibre optic communication systems are facing several key demands. Firstly, the key driving force for fibre optic communication research and development is the growing bandwidth capacity demand, particularly the demand from the global Internet traffic, which has been growing exponentially. The Internet protocol (IP) data traffic not only exceeded the traditional voice traffic to be the dominant source of content, IP is gradually taking over time division multiplexing (TDM) as the voice transmission technology. It is predicted that the U.S. residential VoIP market will grow at a compound annual rate of 20 percent over the next four years [6]. However, the biggest IP traffic generator is video. In the latest network index white paper by Cisco in the June 2010, it is forecasted that Internet video will approach 40% of consumer Internet traffic by the end of 2010, and the sum of all forms of video (TV, video-on- demand, Internet, and peer-to-peer) will continue to exceed 91% of global consumer traffic by 2014. Among them, advanced Internet video (3D or HD) will increase 23- fold between 2009 and 2014, and video-on-demand traffic will double every 2.5 years through 2014. The overall global IP traffic will quadruple from 2009 to 2014 and the annual traffic will exceed three quarters of a zettabyte (a zettabyte is a trillion gigabytes) in 2014 [7]. As the backbone to provide the transportation pipelines for such traffic volumes, the optical network has received demands for larger bandwidth capacity. 40 Gbit/s network is being deployed by network companies worldwide, and 100 Gbit/s transmission has attracted significant amount of research interest and 100 Gbit/s transmission field trials have been reported [8-9]. Secondly, the transmission distance needs to be extended. Currently optical regenerators are placed after certain distance of transmission. These regenerators convert the optical signal to electrical signal and perform re-amplification, re-shaping, and re-timing because the optical signal has been distorted after the transmission. After regeneration, the electrical signal is converted back to optical signal for further transmission. Since such optical-electrical-optical (OEO) regenerators are costly and consume large amount of electrical power, it is desirable for the transmission distance to be extended before regeneration is required. In order to achieve that, various system and transmission impairments need to be mitigated. Thirdly, since there are large numbers of data channels in the high-capacity transmission system, and the IP traffic tends to be dynamic in nature, the WDM system 3 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

also needs to be offer the flexibility in provisioning and routing. This is reflected in the wide deployment of reconfigurable optical add/drop multiplexer (ROADM). It allows the operator to remotely reconfigure the network or perform automatic provisioning. Multi-degree ROADM or wavelength cross-connect (WXC) will lead to greater degree of flexibility. Another aspect of flexibility is the mobility. Even though the fibre optic network does not offer much mobility, it is a requirement of many end user devices. As fibre reaches the last mile and enters the home, the last meters of the network are often handled by wireless communication technology. Therefore the seamless integration of optical and wireless technology is a goal for next generation network. Reliability is another important issue in optical network. Effective and efficient protection and restoration mechanism is becoming increasingly critical in the optical network with high data rate, large channel count, long reach and dynamic configuration. Eventually, the deciding factor to the adaptation and deployment of new technology is the cost. The owner of the network needs to justify the economy benefit of using them. The cost is reflected in various aspects, including the hardware cost, footprint, power consumption, upgrade ability, interoperability with the other systems, and the operation cost.

1.1.3 DWDM Fibre Optic Communication System Technologies a. System Configuration and Architecture Dense WDM (DWDM) transmission system refers to the type of WDM system in which the neighbouring optical channels have spacing of 100 GHz (around 0.8 nm in terms of wavelength) or less. It usually operates only on the conventional transmission window (or C-band) ranging from 1530 nm to 1565 nm. This is also the effective operation region of EDFA. In contrast to coarse WDM (or CWDM) system where the channel spacing is typically 20 nm and the channels spread across the 1310 nm and 1550 nm transmission windows, the DWDM system can accommodate more channels within the operation band (typically 40 channels at 100 GHz spacing or 80 channels at 50 GHz spacing within the C-band) and avoid the water peak wavelength of the fibre where scattering caused by oxygen-hydrogen bond vibration occurs in the standard silica optical fibre. Thus DWDM system offers higher bandwidth and longer transmission distance, and becomes the backbone of global telecommunication network. In a DWDM fibre optic communication system, multiple optical channels are combined together for transmission. These channels can also be processed individually 4 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

in optical switching nodes. Partial or all channels can be dropped/added or crossed to other DWDM links in a multi-node DWDM network. Figure 1.1 shows the basic schematic of a DWDM transmission system.

Data λ Transmitter Receiver 1 in 1 D Odd M channels E

… … U M X U Data λ X Receiver N-1 in N-1 Transmitter ROADM … Data λ Transmitter 2 in 2 Repeater Receiver Interleaver Interleaver D M Drop Add E U M

… Or cross … X Even U Data channels λ X N in N Transmitter Receiver

Figure 1.1 Schematic diagram of a DWDM transmission system. b. Subsystems / Devices The DWDM transmission system contains several key subsystems and devices, as illustrated in Figure 1.1. Below are some brief descriptions of each subsystem or device and its functions. The transmitter contains optical source, modulator and modulation circuitry to impose the high-speed data onto the laser beam. Typically the optical source is a distributed feedback (DFB) laser with continuous wave (CW) output, and LiNbO3 modulator is used to modulate the intensity or phase of the laser output with the data. For lower transmission rate of 10 Gbit/s or below per channel, standard intensity modulation with non-return-to-zero on-off-keying (NRZ OOK) is used. Different modulation schemes (such as QPSK, OFDM) and multiplexing techniques (such as polarization multiplexing) are used for high-speed transmission. Multiple channels with different wavelengths are combined into a single fibre using an optical multiplexer. This multiplexer is typically an arrayed waveguide grating (AWG)-based passive optical device. It multiplexes these DWDM channels on standard ITU (International Telecom Union) defined grids with fixed channel spacing such as 100 GHz. To increase the capacity of the fibre, an optical interleaver, which is a comb- shape passive filter, is placed after the multiplexer to combine the signals from two multiplexers so that the channel density can be doubled. The DWDM signals from

5 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

these two multiplexers are on alternate channel grids, and they are called odd and even channels. The combined DWDM signal travels down the fibre optic link. Repeaters, typically comprised of EDFA and dispersion compensation module (DCM), are placed after every span. Recently, some transmission systems eliminate the in-line dispersion compensator and use digital signal processing at the receiver to perform the CD compensation, while some systems use hybrid structure of both in-line DCM and electronic CD compensation. In some ultra-long haul transmission links and submarine links, Raman amplifier is added to deliver even higher optical power amplification level. In the middle of the transmission link, there might be need for part of the DWDM channels to be dropped locally, while others added to the fibre. Some channels are crossed with other DWDM transmission paths. These are performed on optical switching nodes. In the past, this was performed manually or use large scale switching array. In recent years, switching nodes with dynamic switching capability, such as ROADM nodes, have been designed and deployed to perform flexible add, drop and cross-connect operation. At the receiver end, another interleaver is used to separate the odd channels and even channels. This interleaver has the same design and characteristics as the interleaver at the transmission end, but it functions in the opposite direction. Sometimes it is called the de-interleaver. The odd and even groups are demultiplexed into individual channels using optical demultiplexers, typically AWG-based device too. Each demultiplexed DWDM optical signal is converted to electrical signal using an optical receiver that correspond to the modulation and multiplexing technique used in the transmitter. Before the signal reaches the photodetector, fixed or tuneable optical dispersion compensation is usually performed to remove the dispersion accumulated throughout the transmission link. Since the photodetector can only convert optical intensity to electrical signal, additional devices such as optical delay interferometer or 90 degree optical hybrid are needed to demodulate signals with advanced modulation and multiplexing (such as PSK signal or polarization multiplexed signal) into intensity signals to be detected. The signal can then be further processed electronically to compensate for transmission impairments and recover the original data.

6 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

1.1.4 Key Limiting Factors and Mitigation Technologies The DWDM optical networks are developed to meet the data, distance, flexibility, reliability and low cost requirements described earlier. However, they are facing different technological limiting factors. In this session some key limiting factors are discussed and some related mitigation technologies are reviewed. The first factor is the limited transmission windows available in the optical fibre. As described above, the C-band contains only 35 nm spectrum. Even after adding the long wavelength transmission band (L-band), the available spectrum is only 95 nm (1530 nm to 1625 nm). Beyond this spectrum the EDFAs cannot operate and the DWDM communication becomes unfeasible. Therefore the spectral efficiency of the DWDM signals needs to be improved to increase the per-channel data rate and allow more data to be transmitted within the limited spectrum. The main solution to achieve this is to use advance modulation formats and multiplexing techniques. Quadrature modulation formats that use phase, intensity, multi-level amplitude or combinations have been used to increase the data rate per symbol. The examples are ((D)QPSK), duobinary (DB), multilevel amplitude shift keying (M-ASK) and QAM [10-12]. Polarization shift keying (PolSK) has also been proposed [13]. In terms of multiplexing techniques, time, frequency, code and polarization have been used to combine multiple channels together to increase the transmission data rate [14]. (OFDM WDM transmission has also been attracting strong research interests lately [15]. Recently a record spectral efficiency of 7.0 bit/s/Hz were reported using coherent polarization multiplexing with OFDM technology [16]. Secondly, the DWDM signals suffer from various transmission impairments, which limit the transmission distance and capacity. A transmission impairment is the fibre attenuation and insertion loss from optical components. Fibre dispersions, including chromatic dispersion (CD) and polarization mode dispersion (PMD), also cause pulse spreading along the transmission link and limit the transmission distance. Another source of signal degradation is fibre nonlinearity, which arises from various optical effects such as Kerr effect, stimulated Brillouin scattering, and stimulated Raman effect. Other factors that affect the transmission performance include the crosstalk between WDM channels and filter narrowing caused by passive optical components, which will cause inter-symbol interference (ISI). Different technologies were proposed to mitigate various transmission impairments and to increase transmission distance, such as transmission fibre design (a recent example is the ultra- 7 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

low loss fibre developed by Corning [3, 17]), error correction coding (such as low density parity check codes [18-19]), receiving techniques with better OSNR tolerance (such as homodyne and heterodyne coherent detection [20]), optical dispersion compensator (such as fibre Bragg grating-based or Gires-Tournois etalon-based CD compensators [21-22] and feedback PMD compensator [23-25]), optical filter narrowing effect analysis and mitigation through intra-channel equalization [26-28], and special digital signal processing (DSP) algorithms for dispersion mitigation [29-30] and nonlinear effect compensation [31]. Another limiting factor is that the existing DWDM networks are still not very flexible. For example, the optical switching is either performed at the fibre level or at the wavelength level. No intermediate level switching is allowed. Also, the transponders in the current multi-degree ROADM nodes are limited to a particular degree, and each add/drop port can only process transponder with a pre-determined wavelength. These limit the efficient sharing of hardware resources. Manual intervention is required if channel or signal direction is changed at the access or egress point of the network, or if a signal needs to be rerouted around network congestion points. To tackle these issues and to increase the flexibility of the network, different network topologies (such as point-to-point, ring and mesh) [32-34] with various hierarchies and switching granularities [35-37] are studies, switching node architectures are designed to provide advanced features such as reconfigurable optical add/drop multiplexing [38-40] and colourless and directionless wavelength cross-connect [41- 42], network routing control and traffic/resource management methods are proposed (such as new dynamic bandwidth/channel allocation and load balancing algorithms [43- 45] and adaptive lightpath configuration via LightLabel [46]), and there has been extensive research on higher layer traffic aggregation and scheduling protocols such as generalized multi-protocol label switching (GMPLS) [47-50]. Sometimes, the DWDM network suffers from accidents such as fibre cut, nature disaster or hardware failure. Various failure finding methods are investigated (such as adopting failure location algorithm and reusing downstream light source) [51-52], different network protection and restoration techniques are proposed and investigated (such as path protection/restoration or link protection/restoration, dedicated or shared back up, pre-configured or dynamic discovery) [53-56].

8 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

1.2 Objectives of the Research a. Targeted Problems This research focuses on the physical layer of the optical network, in particular the fibre optic device and subsystem aspects of the DWDM system. The goal is to design and develop new devices and subsystems to improve the transmission performance of DWDM systems and increase flexibility while reducing the hardware and operation expenses. In particular, my thesis research addresses five technical issues in this regard. Firstly, the hybrid configuration issue will be addressed. The upgrade of the transmission system is usually performed gradually, which means that there will be transponders with different data rates or different modulation technologies used concurrently within the same DWDM system. The current DWDM system does not take into account such hybrid configuration, and thus the optical spectrum is not utilized to the optimum level and the overall transmission capacity and performance are limited. Secondly, the filtering effect due to various passive devices on a DWDM system will be addressed. As the data rate for each channel increases to meet the capacity demand, the signal spectrum is also getting wider. At the same time, the DWDM channels are becoming denser, and thus the passive devices in the transmission system, such as the interleaver, multiplexer, and optical switches in the ROADM node, will cause more filtering effect, which in turn leads to inter-symbol interference in the optical signal. Thirdly, the issue related to flexible and reconfigurable switching nodes will be addressed. Currently optical components such as wavelength blocker (WB) or wavelength-selective switch (WSS) are required in the DWDM switching node to perform reconfigurable channel add/drop and cross-connect function. Besides high cost, the switching node with these components cannot provide flexible upgrade capability when the demand of add/drop or switching function increases in the node. Fourthly, the issue related to replace optical multiplexer with transponder aggregator will be addressed. As the ROADM node becomes colourless and directionless, transponder aggregator will be required to replace the conventional optical demultiplexer. This will lead to high cost, high power consumption, and high optical loss at the switching node.

9 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

Finally, how to compensate polarization mode dispersion will be addressed. As the data rate increases, polarization mode dispersion becomes a significant issue because it limits the transmission distance and leads to higher error rate. The current optical dispersion compensators in the market are for chromatic dispersion only, and cannot compensate for PMD since it is a stochastic process and cannot be predicted. The PMD compensation technologies proposed by researchers are either slow or require complex operation. b. Research Topics The aim of my research work is to combat these technical challenges by developing novel optical devices and subsystems in a DWDM system. Most of these devices and subsystems are passive. The innovations rise from the new structure design and implementation technique, and the results are new functions and capabilities with improved performance. Through them, the DWDM systems can be more efficient, the operation can be easier, some system impairments can be mitigated, and the capital expense (CAPEX) and the operation expense (OPEX) will be lower. Five research topics will be studied in this thesis: • Improving spectrum utilization and providing seamless in-service future upgrade capability by tunable asymmetric interleaver • Improving inter-symbol interference tolerance of the optical signal during transmission by intra-channel equalizer and related integrated passive devices • Enabling low cost yet flexible and expendable ROADM function by applying flexible band tunable filter • Simplifying switching node and reduce hardware cost by developing transponder aggregator without wavelength selector • Mitigating PMD impairment by all-order PMD compensator that is based on spectral interference technique and pulse shaping The application locations of these five subsystems/devices in the DWDM transmission system are shown on Figure 1.2.

10 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

Data Research topic 1: Improve Research topic 3: Enable λ Transmitter Receiver 1 in 1 spectrum utilization and low cost yet flexible and D Odd provide seamless in-service expendable ROADM M channels E U upgrade by tunable function by applying M X asymmetric interleaver flexible band tunable filter U Data λ X N-1 in N-1 Transmitter Receiver ROADM … DC Data λ 2 Transmitter 2 in Repeater D Interleaver Interleaver Receiver M Drop Add E U Research topic 5: M X Even U Data channels Research topic 4: Simplify ROADM Mitigate PMD λ X N in N Transmitter node and reduce hardware cost by impairment by all-order Receiver developing transponder aggregator PMD compensator Research topic 2: Improve without wavelength selector inter-symbol interference tolerance by colourless intra-channel equalizer Figure 1.2 Application locations of the five subsystems/devices studied in this thesis.

1.3 Thesis Outline The first chapter of this thesis (this chapter) presents some backgrounds of the research, which is the DWDM optical communications system. It includes some brief history, application demands, and various research fields. Then it lists five technical challenges faced by the current DWDM system, and the proposed device and subsystem research topics to solve these challenges. The following chapters are divided according to the individual research topics. Research Topic 1 (Chapter 2) is on developing a novel tunable asymmetric interleaver device. It is the first interleaver that produces asymmetric interleaving passband ratios and allows continuous in-service tuning of the interleaving ratio. I demonstrate that this device enables spectrum optimisation in hybrid DWDM systems during system upgrade through both simulation and experiment. Research Topic 2 (Chapter 3) is on developing a novel colourless intra-channel equalizer. This low cost passive device is designed to restore a raised cosine passband profile in all DWDM channels. It has been experimentally demonstrated to be very effective in simultaneously mitigating the inter-symbol interference in all channels of the DWDM transmission link. Some integration device based on this colourless intra- channel equalizer are also introduced and demonstrated. Research Topic 3 (Chapter 4) is on developing a flexible band tunable filter. Unlike conventional tunable filter which only offers tunability for the centre frequency, this novel device provides two degrees of tuning simultaneously, namely the centre

11 )%, /)%, / ),#&.!)# ),#&.!)# ),#&.!)#

frequency and passband width. This device becomes a key element that enables the design and construction of a low cost yet flexible and expendable ROADM node. Research Topic 4 (Chapter 5) is on developing a transponder aggregator subsystem for a colourless, directionless and contentionless ROADM node. It does not require any wavelength selector as in the existing transponder aggregators, instead it uses the local oscillator in the coherent receiver to select the target channel. Through theoretical analysis and experimental verification, it is demonstrated that this scheme produces little OSNR degradation even in a 96-channel DWDM system. This novel technology simplifies the ROADM node, reduces hardware cost, size and power consumption, and improves the system reliability. Research Topic 5 (Chapter 6) is on developing a novel all-order PMD compensation method. Unlike the existing methods that require feedback loops with large number of iterations or complicated calculation for the state of polarization, this method uses spectral interference to extract the PMD information at all orders in real time. The accuracy of PMD measurement at the first order and higher orders are demonstrated experimentally and through simulation. Together with a pulse shaper, an simple and effective all-order PMD compensator subsystem is proposed to mitigate this important impairment in the DWDM transmission. Each chapter in Chapters 2 to 6 begins with a background briefing, a research work overview, and a problem description. Then it provides theoretical analysis and proposes the device or subsystem. Based on the design, hardware prototypes is implemented and characterized. Experiments are then conducted to verify the performance of the invented device/subsystem, followed by experimental data analysis and error analysis before a chapter conclusion. With this structure, the literature research will not be placed in an individual chapter, but instead spread into each topic and each chapter. Finally, the seventh chapter is the summary of this research. It also suggests some related research topics for future study.

12 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

)%, )%, 000      

       

   

Optical interleaver is usually needed in DWDM system upgrade. However all the conventional optical interleavers have symmetric passbands, which cannot achieve optimum spectrum allocation for 10G/40G or 40G/100G hybrid DWDM systems. In this chapter I propose a novel optical interleaver whose odd/even passband ratio can be continuously tuned to any symmetric or asymmetric setting. Two design methods are proposed and implemented. The prototypes demonstrated the target passband ratio tuning feature. Simulation and experimental verifications in 10G/40G system and 40G/100G system have been carried out. The results show that this tunable asymmetric interleaver can optimise the spectrum utilization in the DWDM system and lead to better overall performance.

2.1 Background 2.1.1 Capacity Upgrade in DWDM Systems In the recent years, there have been rapidly increasing demands of network traffic volume due to the spreading of Internet technology and the growth of business and personal applications. As a result, the bandwidth of existing DWDM networks with 10 Gbit/s data rate over 100 GHz channel spacing is becoming inadequate. To upgrade the traffic capacity, higher channel density and/or high data rate per channel are required. There are three typical ways for the upgrade. Firstly, make the channel spacing denser such as reducing the channel separation between adjacent WDM channels from 100 GHz to 50 GHz, since the standard DWDM wavelength assignment by International Telecommunications Union’s Telecommunication Standardization Sector (ITU-T) is based on grids with doubling density (such as 200 GHz, 100 GHz, 50

13 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

GHz and even 25 GHz). Each step will double the channel number within the certain operation band, and in turn double the traffic capacity per fibre. Secondly, the data bit rate per channel can be increased. 40 Gbit/s channels are gradually replacing 10 Gbit/s channels in the global DWDM networks. Some network operators even decide to skip 40 Gbit/s step and upgrade to 100 Gbit/s directly. Thirdly, the operating spectrum of the fibres can be expanded. Currently most of the DWDM systems in the world operate at C-band with wavelength ranging from 1530 nm to 1565 nm. If L-band (long band) and S-band (short band) are also utilized, the operation wavelength range can be expanded to 1460 nm to 1625 nm. There are also technologies that use other bands such as O-band (original band, 1260 nm to 1360 nm) and E-band (extended band, 1360 nm to 1460 nm). These three approaches are not exclusive to one another. Two or all three of them can be implemented at the same system to deliver greater level of bandwidth upgrade. Among these three approaches, the operation band expansion is limited by the characteristic of fibre. Some of the legacy fibres can only deliver satisfactory transmission performance at the certain band. For example, most of the older SMF-28 fibres in the US long-haul optical network are suitable for C-band operation but not optimal for L-band, while most of the fibres laid in Japan are dispersion shifted fibres that are more suitable for L-band operation due to their large nonlinear impairments (such as four wave mixing) in the C-band. The network equipments are also typically optical spectrum dependent. For example, EDFA, the most common type of amplifier in DWDM network, cannot amplify S-band signal. Also, the multiplexers and demultiplexers are wavelength dependent. Since it is costly to replace the entire fibre cable routes and network equipments, it is not a feasible approach to upgrade the network bandwidth by expanding the operation spectrum if the existing network infrastructure (fibre and equipment) does not support the added band. Most of the existing optical networks fall into this category, therefore the most suitable upgrade solution for them is to narrow down the adjacent channel spacing to 50 GHz and increase the data rate per channel to 40 Gbit/s or higher. However, since it is costly to forego all the legacy systems that run at 10 Gbit/s with 100 GHz channel spacing and replace the entire network to 40 Gbit/s channels, the practical initial step taken by most carriers is to insert 40 Gbit/s channel between two adjacent 100 GHz- spaced 10 Gbit/s channels. With this upgrade, the transmission data rate can be increased 5 times. For example, within a typical transmission band of 4 THz, the total 14 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

transmission data rate from the DWDM channels can be increased from 400 Gbit/s (40 channels × 10 Gbit/s/channel) to 2 Tbit/s (40 channels × 10 Gbit/s/channel + 40 channels × 40 Gbit/s/channel) with the upgraded hybrid 10G/40G system. This may be sufficient to for their bandwidth requirement for several years. In following steps during the subsequent years as the 10G/40G channels are gradually filled, the 10 Gbit/s transponders can be gradually replaced with 40 Gbit/s transponders or even straight to 100 Gbit/s transponders to be ready for further increases in bandwidth demand. The hybrid interleaved upgrade can also minimize the affect on the existing traffics, making the upgrade operation less complicated. To achieve the 10G/40G or future 40G/100G upgrade, new techniques and equipment will be required. This chapter introduces a novel method and corresponding optical device to enable low cost upgrade while maintaining good network performance. The performance improvement is verified through numerical simulation and transmission experiments.

2.1.2 Optical interleaver a. Optical Interleaver and Its Functions Optical interleaver has been shown to be an effective and economical way to double the channel capacity in a DWDM system [57]. An optical interleaver is a passive optical device that can segregate a group of channels into odd and even sets with doubled channel spacing. For example, a 50GHz/100GHz (50G/100G in short) interleaver can segregate a group 50 GHz spacing channels into two groups of channels, each with 100 GHz spacing (Figure 2.1(a)). The spectrum at the outputs of the interleaver has a periodic profile (in this example, the free spectral range (FSR) is 100 GHz), therefore it is also called a “comb filter”. Typically optical interleaver allows lights to travel in both directions, allowing it to perform both interleaving and de- interleaving functions (Figure 2.1(b)). In common usage, the term “interleaver” refers to both interleaver and de-interleaver since they are essentially the same device but with different optical transmission direction. Therefore in this thesis, the term “interleaver” is used in both interleaving and de-interleaving applications. Interleavers can also be placed in cascade to further de-interleave or interleave optical channels into finer or coarser groups (Figure 2.2). In some interleaver designs, these interleavers are integrated into the same device to form a 1×n interleaver with n greater than 2. Some interleavers have cyclic structures. An n×n cyclic interleaver functions as the 15 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

integration of n units of 1×n or n×1 interleaver with cyclic shift of n output ports (Figure 2.3).

100 GHz 100 GHz

50 GHz In 1 Out 1 50 GHz Odd Odd In Out Even Even Out 2 In 2 Interleaver Interleaver

100 GHz 100 GHz (a) Interleaving function (b)De-interleaving function Figure 2.1 Functions of optical interleaver.





 

    



    

Figure 2.2 Interleavers in cascade.

 

 

 

 

   

Figure 2.3 Operation of a 4×4 cyclic interleaver.

16 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

All conventional optical interleavers are passive fixed wavelength devices. Firstly, the FSR of each interleaver output is fixed. In DWDM applications, the FSR value is typically 25GHz, 50 GHz, 100 GHz or 200 GHz. Secondly, the centre frequencies (or wavelengths) of the interleaver output’s passbands are pre-determined. During the manufacturing process, the optics of the interleaver is fine tuned so that the passband centres will fall on the standard ITU-T grids. Thirdly, the passband splitting ratio of the interleaver outputs is always 50:50. In other words, the passband of an Odd channel has the same passband width as the passband of an Even channel. For an optical interleaver to deliver good performance in DWDM network node and link there are several requirements. These requirements are mostly based on its optical characteristics. • The optical interleaver should produce a square-like spectral profile. A square- like profile includes a flat passband top, a wide passband width, and steep passband edges at both sides. Interleaver with squire-like spectral profile will be less sensitive to wavelength shift of the optical source that arises from poor laser quality or temperature variation. It will also allow more interleavers to be placed in cascade without significantly reducing the filter passband width. • As for other passive components in optical communication system, the insertion loss at the passed bands should be low. • High isolation at the rejected bands is desirable. It reduces crosstalk between the Odd and the Even channels. • Dispersion figures should be low, especially when high bit rate traffic is transmitted in each DWDM channel. These figures include chromatic dispersion and polarization mode dispersion. Low polarization dependent loss is also required. • The interleaver needs to be reliable. Some measures to achieve high reliability are: easy alignment, easy integration within optical link, insensitive to environmental change (such as temperature variation), insensitive to vibration, small number of elements, and compact footprint. • Finally, while maintaining good optical characteristics, the cost of the device should be low.

17 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

b. Technologies and Implementations of Conventional Optical Interleaver Generally speaking, all optical interleavers are based on optical interferometer principle. However there are many different technologies to realize an optical interleaver. In the optical structure aspect, there are linear interferometer type utilizing Mach-Zehnder interferometer (MZI) or Michelson interferometer (MI) structure [58- 67], arrayed waveguide grating (AWG) type [68], coupled resonator type (such as fibre grating or etalon) [69-71], resonance-based interferometer type [72-74], ring resonator type [75], and so on. In the physical material aspect, some interleavers have all fibre structure (including fibre Bragg grating (FBG) and fibre ring resonator) [58-60, 69, 75], some are planar lightwave circuit (PLC)-based [66-68], some have thin film components [70-71], some use free space optics for interferometer [62, 72-74], some use birefringent crystal for interferometer [63-65], and some have combination of more than one materials. Each of the different technologies and implementations of interleaver has its technical strengths and weaknesses in its optical performance. For example, simple single stage MZI or MI type interleaver cannot deliver flat-top spectral response, a desirable feature for application in DWDM system. This can be improved by adding more interferometers in cascade. However the insertion loss of the device will also increase accordingly. Resonance-based interleaver (such as FBG-based, thin film- based, MI with Gires-Tournois resonator, and MZI with ring resonator) can achieve good spectral response with small loss. However the resonance effect leads to higher chromatic dispersion and might thus require additional dispersion compensation mechanism [76]. For different application, different compromises need to be made among these parameters to deliver the optimal performance.

2.1.3 DWDM System Upgrade Using Interleaver This section illustrates how the optical interleaver is used in DWDM system upgrade application. Figure 2.4 shows the current 10 Gbit/s DWDM transmission link with 100 GHz channel spacing. At the source node the 10 Gbit/s data from the DWDM transponders (also called “line cards”) are combined by a 100 GHz DWDM multiplexer and transmitted in a single fibre. For ITU-T C-band transmission, the channel centre frequencies are 191 + 0.1×k THz where k is an integer typically between 1 and 50. The multiplexed DWDM signals are amplified by repeaters at the end of each span. When the signals reach the destination node, they are separated by a 100 GHz DWDM 18 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

demultiplexer into individual channels at respective fibres to be detected by the receivers at the destination transponders. In actual DWDM system, the optical paths are duplex with a reverse direction link set up. When the network is upgraded, additional transponders with 40 Gbit/s data rate are added. These transponders have 50 GHz centre frequency offset compared with the existing 10 Gbit/s transponders. The channel centre frequencies can be represented as 191.05 + 0.1×k THz with the same definition of k as above. Corresponding DWDM multiplexer and demultiplexer with 50 GHz frequency shift are placed at the source node and the destination node respectively to combine and separate these additional channels. The output of the multiplexer for the existing 10 Gbit/s signals and the output



        G G G

 !              

Figure 2.4 100 GHz-spaced DWDM transmission link.



         G G G

 

      G G G " #$ " #$  



Figure 2.5 100 GHz to 50 GHz DWDM network upgrade using interleaver. 19 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

of the multiplexer for the added 40 Gbit/s signals are combined using a 50G/100G optical interleaver at the source node (Figure 2.5). A 50G/100G optical de-interleaver is placed at the destination node to separate these two groups of signals into two demultiplexers respectively. For the reverse path (not represented separately this figure), another pair of 50G/100G interleaver and de-interleaver is used. The optical spectra at various points of the DWDM link are shown as the insets on Figure 2.5. At the inputs of the interleaver at the source node, both DWDM signals from the two multiplexers have 100 GHz spacing between adjacent channels. At the output of the interleaver, the channel spacing becomes 50 GHz. With this network upgrade method, the existing 10 Gbit/s system does not need to be replaced. This will keep the hardware cost to minimum and also reduce the amount of labour required. A more straightforward way to combine the two 100 GHz multiplexed channel groups is to use a 50:50 coupler, since the centre wavelengths/frequencies of the DWDM channels at these groups have 50 GHz offset and do not overlap each other. Even though this method is simpler and costs less, it is not commonly used because of the deterioration of optical performance. Comparing to the interleaver solution, using coupler will introduce larger insertion loss at the channel passing band. More importantly, the isolation of the unwanted signal at the inter-channel gap is significantly lower and therefore large inter-channel crosstalk will be introduced at the combined output. Furthermore, optical splitter cannot replace the interleaver at the de-interleaving side.

2.1.4 Issue of DWDM System Upgrade Using Conventional Symmetric Optical Interleaver The presence of 50G/100G optical interleaver in the optical path will not introduce significant variation of optical performance in the 10 Gbit/s channels. The only effect is the slight extra insertion loss at the source and destination nodes. This can be compensated by adjusting the gain of the booster amplifiers and pre-amplifiers at these nodes. However at 40 Gbit/s transmission, the insertion of 50G/100G optical interleaver will cause significant optical performance deterioration. This is the filtering

20 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

effect caused by the spectral width difference between the interleaver passband and the 40 Gbit/s signal spectrum. With different design and implementation technologies, the passband profile of optical interleaver can vary greatly. Different profiles are suitable for different applications. For example, some interleavers have Gaussian-shape passbands such as Figure 2.6(a). They usually have lower insertion loss and therefore are suitable for optical signals with narrower laser line width and better wavelength stability. Other interleavers have flat-top profiles such as shown on Figure 2.6(b) (usually higher order super-Gaussian shapes such as in second or third order super-Gaussian functions) and are suitable for filtering signals with wider spectra. The difference in passband profiles can also be reflected in the slopes of the passband edges. Steeper slopes are usually associated with wider passbands. For DWDM transmission, particularly for higher bit rate systems such as 40 Gbit/s per channel systems, interleavers designed with flat-top profiles and steeper passband edge are used. However, despite all the efforts to make the interleaver output passband more square-like, the actual passband profile can never be an exact square shape (Figure 2.6(c)) due to limits of fundamental optics. Therefore, the clear channel passband width of the 50G/100G interleaver output is smaller than 50 GHz. Typically the optical interleaver vendors only guarantee minimum passband width of 20 GHz to 30 GHz at 0.5 dB level from the peak even with the flat-top design.

(a) (b) (c) Figure 2.6 Interleaver profiles (a) Gaussian; (b) Flat-top; (c) Ideal square.

Spectral analysis clearly shows that the available passband width of the optical interleaver cannot accommodate the bandwidth required by most 40 Gbit/s signals. Figure 2.7 illustrates the simulated spectra of 40 Gbit/s signal under several types of modulation schemes individually. With conventional schemes such as NRZ-OOK and RZ-OOK, and even with improved schemes such as Carrier Suppressed Return-to-Zero (CS-RZ) and Differential Phase Shift Keying (DPSK), the signal spectra are quite wide (Figure 2.7(a, b, c and e)). When comparing them with the passband spectrum of a

21 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

typical 50G/100G interleaver output (Figure 2.8), it is clear that the spectra under these modulation formats cannot fit into the interleaver passband and therefore the signal will suffer distortion after the interleaver [77].

Figure 2.7 Simulated spectra of 40 Gbit/s signal under several types of modulation schemes.

Figure 2.8 Compatibility of various modulation formats with 50GHz interleaver passband [77].

22 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

Improving bandwidth efficiency of the DWDM transmission system has become a popular research field by researchers worldwide to tackle this issue and to enable satisfactory performance when transmitting 40 Gbit/s signal over 50 GHz spacing. These researches mainly focus on developing new modulation schemes. The proposed modulation schemes, such as DQPSK, Differential PolSK and optical duobinary (DB), aim to achieve 80% bit/s/Hz bandwidth efficiency by encoding in multiple polarizations and/or by encoding multiple bits per symbol. Figure 2.7(d) and Figure 2.8 show that duobinary scheme has the narrowest spectrum and therefore can fit into the passband of a typical 50G/100G interleaver output. Even though some of these new modulation schemes offer higher spectral efficiency and allow 50 GHz interleaver passband to accommodate 40 Gbit/s signal, the problem of passband width remains as the DWDM network is facing the growing demands to increase the per channel data rate to 100 Gbit/s and beyond and to reduce the DWDM channel spacing further to 25 GHz. The latest advanced modulation schemes include polarization multiplexed RZ-PQSK or RZ-8PSK [3, 8-12], DPSK- 3ASK [78], and optical OFDM [5, 16]. Since these transmission systems are complicated and costly to implement, the upgrade is usually conducted in steps. Therefore hybrid system with different data rate or modulation format is common, and the system will change gradually. The interleavers used in the current DWDM systems do not provide any flexibility to accommodate different types of optical signals in such hybrid systems since they are passive device with fixed passband profiles. Therefore, it is desirable to have an optical interleaver that allows flexible adjustment of interleaving ratios between the Odd channels and the Even channels. In the following section, a novel interleaver with such capability is proposed. This tunable asymmetric interleaver provides the solution to reach the optimum transmission performance in the overall hybrid DWDM system.

2.2 Development and Implementation of Tunable Asymmetric Interleaver Optical interleaver with asymmetric interleaving ratio has been proposed before by Cao and Mao [79]. By setting appropriate thickness of optical retardances and reflectivity of the reflective coating in the interferometer components in the interleaver, different Odd/Even splitting ratio can be obtained. However this is a fixed structure. In

23 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

other words, the centre frequency of the passbands and the asymmetry ratio cannot be adjusted once the interleaver is manufactured. Another method to implement asymmetric interleaver is to connect a cyclic interleaver and a cyclic de-interleaver with the same FSR while leaving some ports disconnected as illustrated in Figure 2.9. This is also a fixed structure, and the asymmetry ratio is limited by the number of cyclic interleaver ports.

1 5 9 1 5 9

2 6 10

1 2 3 4 5 6 7 8 9 10 11 12

AWG 2 3 4 6 7 8 10 11 12 3 7 11 AWG

4 8 12

Figure 2.9 Fixed asymmetric interleaver constructed by cyclic interleavers.

In this work, a novel interleaver that offers both continuous passband frequency shifting and continuous asymmetric ratio adjusting capabilities is designed and demonstrated. It is called the tunable asymmetric interleaver (TAI).

2.2.1 Target Function of Tunable Asymmetric Interleaver The TAI has capabilities beyond the existing interleavers. Figure 2.10 shows the example of various output capabilities of the proposed TAI. It can function as a conventional interleaver with symmetric 50:50 output ratio and wavelengths centred at ITU-T grid (Figure 2.10(a)); or it can adjust the passband width ratio between 2 outputs (the interleaving ratio) while maintaining ITU-T grid-centred passbands (Figure 2.10(b) and (c)), the ratio can be set continuously from 50:50 to 0:100 (Figure 2.10(d)); it also allows non-ITU-T grid centred passbands (Figure 2.10(e)), similar to tunable shift interleaver, while allowing asymmetric output ratio simultaneously; in the case that the wavelength shift reaches 50% of the FSR and the output passband ratio is 50:50, the outputs of the interleaver will be exactly opposite to the conventional interleaver (Figure 2.10(f), c.f. Figure 2.10(a)). Two methods to design such TAI device are proposed here, the first one uses cascaded symmetric interleavers, the second one uses programmable optical processor.

24 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

%$ %$

)*    λ %$ %$

)+*    λ %$ %$

),*    λ %$ %$

) * 

 λ %$ %$

)*    λ %$ %$

)-*   GG λ

&'(" 

Figure 2.10 Function of tunable asymmetric interleaver.

2.2.2 TAI Design 1: Based on Cascaded Symmetric Interleavers a. Design Principle In the first design, the TAI is constructed using two symmetric interleavers in cascade: one is a 1×2 tunable shift interleaver and a 2×2 cyclic tunable shift interleaver [80, 81]. These 2 interleavers have the same FSR settings, which are twice the FSR of the resultant TAI. For example, to construct a 1×2 50G/100G TAI, the two tunable shift interleavers should be 100G/200G interleavers. Figure 2.11 illustrates the design of a TAI using the example of a fixed asymmetric interleaver based on the same structure. Figure 2.11(a) shows the functions of the two interleaver components with same FSR but certain level of centre wavelength offset which can be observed as the shift between Edge 1 and Edge 2 in the diagram. The regions with dotted border lines are the passbands. Each passband contains 4 neighbouring channels. The FSR values for the interleaver output ports are all equal to eight times of the input channel spacing. Within the first eight channels, Output 1 of the first interleaver has a passband containing Channels 1-4 of the input signal, while the

25 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

passband of Output 2 contains Channel 5-8. The output passbands of the second interleaver has a wavelength shift equal to the input channel separation. So the Output 1 contains Channel 2-5 of Input 1, while Output 2 contains Channel 1 and Channels 6-8. Since this is a 2×2 cyclic interleaver, Output 1 also contains the complementary channels from Input 2, namely Channel 1 and Channels 6-8. Similarly Output 2 contains Channels 2-5 of Input 2.

% (

   1       . / 0 1  .    . / 0 . 

    . / 0 1  .     . / 0 1  .  

    . / 0 1  .     . / 0 1  .  

)* % (

   1    . / 0  .     . / 0 1  . 2    3 

 . / 0 .   1 

)+* % ( % (

Figure 2.11 Design of an asymmetric interleaver (a) Two interleaver components; (b) Combined asymmetric interleaver.

When these two symmetric interleavers are placed in cascade (Figure 2.11(b)), the two outputs of the first interleaver become the inputs of the second interleaver (points A and B on the diagram). Due to the passband centre wavelength offset between these two interleavers, Channels 2-4 from the input reach Output 1 of the combined interleaver via point A. Even though Channel 5 from point A can also reach Output 1, the signal at point A does not contain Channel 5, therefore only 3 channels reach Output 1 via point A. Similarly, Channels 6-8 of the input signal reach Output 1 via point B. The complementary channels (Channel 1 and 5) end up at Output 2 via point A and B respectively. As a result, Output 1 contains Channels 2-4 and Channel 6- 8, representing 75% of the first 8 input channels, while Output contains 2 channels or 25%. Due to the periodic behaviour of the interleavers, same ratio is obtained at any adjacent 8 channels and therefore at the overall spectrum. Asymmetry at the interleaver 26 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

output is thus achieved. The separation between Edge 1 and Edge 2 in the diagram determines the asymmetry ratio. It should be noted that the two outputs has FSR value equal to four times the input channel separation, comparing to the eight times channel spacing FSR at the two interleaver components. The same principle can be applied in the tunable case. If the first 1×2 symmetric interleaver becomes a tunable shift interleaver, the spectral position of Edge 1 can be tuned. Similarly, Edge 2 in the second interleaver (2×2 symmetric interleaver) can be varied if it becomes a tunable shift interleaver. With different Edge 1 and Edge 2 values, different passband widths at the output ports can be obtained. Thus the asymmetry figure (that is, the passband ratio) can be adjusted and the interleaver becomes a TAI (Figure 2.12). If only one of the 2 interleaver components is tunable, only one edge can be tuned. This is the single stage TAI and does not provide full passband centre wavelength tuning capability. With dual stage TAI, the centre of the passbands can be fixed on the ITU-T grid by selecting equal distance from both edges to the ITU-T grid wavelength/frequency.

  . / 0  .     . / 0 1  .    

  1 

% ( % (

Figure 2.12 Tunable asymmetric interleaver using cascaded shift symmetric interleavers. b. Implementation The implementation of TAI using this design is not limited to any particular type of technology. The only two requirements to construct a TAI comparing to conventional interleaver technology are: tunability for wavelength shift and cyclic capability. Both of these two requirements can be achieved by most of the interleaver technologies. Tunable shift interleaver can be constructed using birefringent elements with mechanical rotation control or by adjusting the resonator in the resonator-based interferometer structure such as varying the Gires-Tournois (G-T) cavity length in a

27 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

Michelson G-T (MGT) interferometer-based interleaver [65]. As for the cyclic aspect, most of the 1×2 interleaver design structure can be extended to a 2×2 structure due to the symmetrical property of the interleaver. Of course, the requirement for the alignment and calibration is stricter and therefore cyclic interleavers are slightly more difficult to manufacture in some technologies. For technologies such as PLC-based AWG, there is little extra difficulty to make cyclic interleaver. In this work, the MGT interferometer technology is used to construct the TAI prototype because it has good optical performance and simple configuration [76]. Here is the brief description of theoretical principles of MI and G-T resonator. The Michelson interferometer produces interference fringes by splitting a beam of monochromatic light into two beams propagating in perpendicular directions, each strikes a mirror and reflects back. When the reflected beams are brought back together by the beam splitter/combiner, an interference pattern results. The schematic of a Michelson interferometer is shown on Figure 2.13.

8 -#, "$ 7" 

L2

I2B E2B L1  3 I1B E1B 

Iin Ein

4)56* +7#"  E2A E1A

I2A I1A  2

Figure 2.13 Schematic of a Michelson interferometer.

For a semi-reflective 50:50 beam splitter placed at 45° angle of the incident beam, the reflectivity value R=0.5, so the reflected components are: ~ = − ~ = ~ = ~ E1A R 1 R E in 0.5 0.5 Ein 0.5 Ein (2.1)

~ = − − ~ = ~ = ~ E1B 1 R 1 R E in 0.5 0.5 Ein 0.5 Ein (2.2)

~ = − ~ = ~ = ~ E2 A 1 R R E in 0.5 0.5 Ein 0.5 Ein (2.3)

28 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

~ = ~ = ~ = ~ E2B R R E in 0.5 0.5 Ein 0.5 Ein (2.4)

Therefore the intensity at Output A is:

= ~ + ~ = ()()~ + ~ ~* + ~* I A E1A E2 A E1A E2 A E1A E2 A

~ 2 ~ 2 ~ ~ ()()φ −φ φ −φ = + + ()i 1 A 2 A + i 2 A 1A 0.25 Ein 0.25 Ein 0.5 Ein Ein e e

= ~ 2 ()+ Δφ 0.5 Ein 1 cos A (2.5)

= ~ 2 ()+ Δφ Similarly: I B 0.5 Ein 1 cos B (2.6) = + Because the conservation of energy, we have Iin I A I B . And since

~ 2 4π I = E , we get cos Δφ + cos Δφ = 0 , where Δφ = f (L − L ) +φ in in A B A c 1 2 A Δφ = π + Δφ Δφ Therefore A k B where k is an odd integer. This means that A and Δφ π B are (odd numbers of) out of phase. In other words, outputs A and B are complementary. Due to constructive and destructive interference that govern the transmission function, their outputs are periodic in frequency (Figure 2.14). 1×2 (or 2×2) interleaver is thus constructed. T

IA

IB

Figure 2.14 Periodic outputs of Michelson interferometer.

Gires-Tournois resonator or etalon is a special type of Fabry-Perot (F-P) interferometer/etalon, because its two surfaces have different reflectivity. One of them is highly reflective with reflectivity close to ideal value of 1, while the other surface has lower reflectivity. Therefore a G-T resonator is sometimes also called an asymmetric F- P interferometer/etalon. (In strict technically definition, etalon is different from interferometer. Etalon structure refers to a transparent plate with two reflective

29 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

surfaces, while interferometer refers to two parallel highly-reflective mirrors. However the terminology is often used inconsistently.)

n

θi θt

100% r

d

Figure 2.15 Schematic of a Gires-Tournois etalon.

Figure 2.15 shows the schematic of a G-T etalon. One of the surface has ideal 100% reflectivity, the other surface has complex amplitude reflectivity of r (r < 100%). Due to multiple-beam interference, light incident on the lower reflectivity surface of a G-T etalon is (almost) completely reflected, but has a phase shift that depends strongly on the wavelength of the light. The complex amplitude reflectivity of a G-T etalon is given by: r − e−iδ R = (2.7) 1 − re−iδ where 2π sinθ δ = nd cosθ and n is the refractive index the media, n = i λ t θ sin t d is the length of the resonating cavity (thickness of the plate)

%t is the angle of refraction the light makes within the plate, and ( is the wavelength of the light. For the special case of r = 100%,

2π C dn S Δφ = D2 − 2d tanθ sinθ T λ D θ t i T E cos t U

4πd C n sinθ S = D − t nsinθ T λ D θ θ t T E cos t cos t U

30 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

4πdn = cosθ (2.8) λ t 4πdn For normal incidence, θ = 0 , therefore Δφ = . t λ For constructive interference, we have Δφ = m ⋅ 2π where m is an integer. Therefore: 4πdn f m ⋅ 2π = = 4πdn m (2.9) λ m c c And so f = m (2.10) m 2dn c According to the definition of FSR f = , we get f = m ⋅ f . And 0 2dn m 0 therefore: 4πdn f f f Δφ = = 2π = 2π = 2π (2.11) λ m c f0 fm m 2dn The schematic of the MGT-based interleaver is shown on Figure 2.16(a). At output A, two beams travelling along these two paths interfere to form the interleaver output. The phase variation between these 2 paths φA comes from both Michelson interferometer and the G-T resonator. It is determined by the path length difference:

Δφ = Δφ + Δφ = π f ()− + π f A A−Michelson A−GT 2 2 L1 L2 2 (2.12) c f0 where L1 is the separation between the beam splitter and the reflective mirror, while L2 is the separation between the beam splitter and the G-T resonator. f0 is the FSR of the

G-T resonator (50 GHz, 100 GHz, etc.). Calibrate the lengths L1 and L2 to match f0:

− = − c L1 L2 ,then we have 4 f0 C S Δφ = π f D− c T + π f = π f A 2 2D T 2 (2.13) c E 4 f0 U f0 f0

So when f = 2m⋅f0 where m is an integer, we have φA = 0, and φB = 4; and when f = (2m+1)⋅f0, we have φA = 4, and φB = 0; since φA and φB are odd number of 4 out of phase in a Michelson interferometer. 2×2 cyclic interleaver can also be

31 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

constructed with the similar structure (Figure 2.16(b)). The second input is placed perpendicular to the first input.

!-#, "$7" 

L 2 d

IA L1  2 n  Iin "'  " 4+7    #" , 7+"

IB  3

(a)

!-#, "$7" 

L 2 d

IA L1  2 n   Iin "'  " 4+7    #" , 7+"

IB    3

(b) Figure 2.16 Schematic of a Michelson-Gires-Tournois interferometer-based interleavers: (a) 1×2 interleaver, (b) 2×2 cyclic interleaver.

With some tuning mechanism, the lengths L1 and/or L2 can be adjusted, so the

0’s and 4’s can be obtained at different frequencies. So we will have φA = 0 and φB

= 4 when f = 2m⋅f0 + f, and φA = 4 and φB = 0 when f = (2m+1)⋅f0+ f. If the reflection mirror is replaced by another G-T resonator, the output passband profile can be further improves with better square-like behaviour and superior channel isolation [72]. This G-T resonator does not need to be tunable. To further improve the passband profile, one of the single cavity G-T resonators is replaced by a

32 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

multi-cavity G-T resonator, however the chromatic dispersion of the interleaver worsen significantly. Another alterative design is to replace both single cavity G-T resonators with multi-cavity ones. This configuration produces very good CD result with the trade-off of narrower passbands. It should be noted that the polarized version of MGT interferometer is not suitable for constructing tunable shift interferometer. Because with polarized MGT structure, the two polarization states function as the two beams in the interferometer. A result is that these two interfering beams propagate on the same paths and do not have path length difference. This is a desirable property for conventional interleaver due to its high tolerance and ruggedness, however since the path length difference is necessary for to achieve wavelength shift, the polarized MGT structure is not suitable for TAI implementation. Among various tuning technologies, thermal tuning is used in this prototype. This is because thermal tuning does not require physical moving part such as stepper motor used in mechanical tuning, and thermal controller with fine adjustment capability is widely available. One disadvantage of thermal tuning is the tuning speed. However since the thermal mass of the GT cavity is low due to the small physical size, fast (sub second level) tuning can be achieved using Peltier type temperature controller (also called thermal-electrical cooler or TEC, although it can also be used for heating). This tuning speed is sufficient for our proposed applications in DWDM network, since interleaver tuning and reconfiguration is only required occasionally, unlike optical circuit provisioning or optical burst switching. Due to the device fabrication limitation, only the first symmetric interleaver element has tunability in the constructed prototype. The cyclic interleaver has fixed passband centre wavelengths. Therefore the prototype is essentially a single edge TAI. However this does not affect the proof-of-concept purpose of the prototype. c. Characterization of Optical Properties The passband profile and tuning capability of the constructed cascaded symmetric interleaver-based TAI is measured using Agilent 81910A photonic all- parameter analyser, which consists of a low spontaneous source emission (SSE) sweep laser source, high sensitivity power sensors, polarization controller and optical test head and is able to measure insertion loss, polarization dependent loss (PDL), CD, PMD and return loss of passive optical device simultaneously. 33 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

0

Even Even Even 5 Passband 1 Odd Passband 2 Odd Passband 3 Odd Passband 1 Passband 2 Passband 3 10

15 1x2 shift interleaver

20 2x2 cyclic interleaver

25 1x2 tunable asymmetric

Insertion loss (dB) interleaver 30

35

40 193.55 193.6 193.65 193.7 193.75 193.8 193.85 Frequency (THz)

Figure 2.17 Spectra of the individual interleaver components and combined tunable asymmetric interleaver at room temperature.

Figure 2.17 shows a section of the output spectra of the individual and combined interleavers at room temperature. The shaded curve is the spectrum of one of the outputs of the 1×2 shift interleaver when the temperature controller is set to room temperature. The dotted curve is the spectrum at one of the outputs of the 2×2 symmetric cyclic interleaver measured individually. The solid curve is the resultant output spectrum when these 2 individual symmetric interleavers are connected together in series. It can be observed that the resultant TAI has a FSR equal to half of the FSR of the 2 interleaver components. Half of the TAI passbands (such as Odd Passband 2 on Figure 2.17) each contains one edge of the 1×2 shift interleaver (the rising edge under the frequency spectrum) and one edge of the 2×2 cyclic interleaver (the falling edge under the frequency spectrum). The other half (such as Odd Passbands 1 and 3 on Figure 12) each has the rising edge from the other output of the 1×2 shift interleaver and the falling edge from the other output of the 2×2 cyclic interleaver (not shown here). It is vice versa for the Even Passbands contained at the other output of the TAI interleaver. The spectral locations of these passbands are marked by Even Passbands 1- 3 on Figure 2.17. Since the passbands of the combined TAI is constructed by the intersected spectral area of two interleaver components, the slopes of the passband edges are the same as the slopes of the individual interleavers. As described above, the individual

34 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

interleavers are originally designed to interleave signals with periods equal to twice the SFR of the TAI. So the passband edge slope is only optimized for larger FSR and larger spectral separation between the WDM signals. With half the FSR, the requirement for steepness of the passband edge and width of passband is higher. Therefore TAI constructed by interleavers with standard square-like profiles will have worse square-like profile compared to the single stage interleaver. Figure 2.18 is an example. Figure 2.18(a) is the output of one of the interleaver components that make up the TAI. This interleaver output has a FSR of 200 GHz. Its rising edge slope is suitable for 200 GHz FSR applications that separates the odd and even channels of DWDM signals with 100 GHz spacing. This rising edge forms the rising edge of the TAI’s output passband (Figure 2.18(b)). Despite the fact that the resultant TAI has a FSR of 100 GHz and can separate DWDM signals with 50 GHz spacing, the steepness of its rising and falling edge does not reach the steepness level of a single stage interleaver with the same 100 GHz FSR (Figure 2.18(c)). This will lead to more crosstalk from the neighbouring channels and narrower useable passband.

!""( (

)*

!""( ( )+*

!""( (

),*

,;4 

9:!4

Figure 2.18 Passband edge comparison (a) interleaver component; (b) combined tunable asymmetric interleaver; (c) standard symmetric interleaver.

35 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

To solve this problem, stricter passband profiles will be required for the two interleaver components. For a 50G/100G interleaver (interleaver with 100 GHz FSR at the output to separate DWDM signals with 50 GHz channel spacing), the two 100G/200G interleaver components should be designed to meet the slope requirement of standard 50G/100G interleaver. As described above, there are various techniques to improve the passband profile to be more square-like (higher super-Gaussian order), including replacing a single cavity G-T resonator with a multi-cavity resonator. The peak insertion loss figures of the passbands are about 1.25 to 1.5 dB for the 2 interleaver components respectively. The combined TAI has peak insertion loss of about 3 dB, including the connectors at both ends and between 2 interleaver stages. This figure is sufficiently low for most of the applications proposed in this report. For actual production, these 2 interleaver components will be integrated inside one package and no fibre connectors and adapters are required in between. This can reduce the insertion loss further. 1.5 dB maximum insertion loss for the TAI without connectors can be easily achieved. d. Characterization of Asymmetry Tuning The tunable shift interleaver in the TAI prototype unit uses thermal tuning with Peltier type heater/cooler, and a thermal coupler is used to provide temperature reading. Because there is no temperature controlling equipment available that can take the thermal coupler readings, a temperature controller feedback circuit is set up to perform thermal tuning for the TAI (Figure 2.19). The feedback circuit consists of the thermal coupler probe from the tunable shift interleaver, a Micromega Autotune PID temperature controller to read thermal coupler output, a Crydom PCV series power relay and a regular DC power supply. Based on the reading from the thermal coupler probe, the temperature controller decides whether to close the circuit and supply power to the device. If the temperature controller circuit is closed, the power relay will close the power supply circuit which then provides up to 5 V voltage to either heat up or cool down the tunable shift interleaver through the Peltier thermoelectric device. Because the temperature is displayed as current in a thermal coupler, only the absolute number is shown. Therefore two sets of power relay and power supply are required, one for cooling and the other for heating. For the future TAI units, thermistor can be used for temperature reading. Unlike the thermal coupler, thermistors are thermally sensitive resistors with positive or 36 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

negative resistance/temperature coefficient. Using thermistor, a single circuit can perform both heating and cooling functions. Thermistor-based temperature controllers are also more widely available.

Tunable asymmetric interleaver

1x2 Tunable 2x2 cyclic Power shift symmetric supply interleaver interleaver

Power Thermal relay coupler

Temperature controller

(a)

Broadband Power optical source supplies

Temperature Asymmetric controller passbands Thermal Power coupler relays

2x2 cyclic 1x2 Tunable symmetric shift interleaver interleaver

(b) Figure 2.19 Temperature control feedback circuit for the TAI. (a) Schematic; (b) Photo of the experimental setup.

In the tuning experiment, the temperature of the tunable shift interleaver is varied from 7°C to 75°C. As expected, the Odd:Even passband splitter ratio at the TAI output varies accordingly. Figure 2.20 (a) and (b) show the passband profiles of one 37 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

output that varies from 20% to 100% of the FSR in steps of 10% and the other output from 80% to 0% respectively. These passbands have the same falling edge because the falling edge is determined by the second interleaver component, namely the 2×2 cyclic interleaver. As described earlier, the cyclic interleaver in this prototype is not shiftable, therefore the passband cannot be centred at a same frequency such as the ITU-T grid.

8

18

08

/8

.8

8

8

8

8

(a)

08 /8

.8

8

8

8

8

8

8

(b) Figure 2.20 Passband profiles of TAI: (a) Output port 1 with interleaving ratio varies from 20% to 100%, (b) Output port 2 with interleaving ratio varies from 80% to 0%. 38 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

193.55 193.6 193.65 193.7 193.75 193.8 193.85 0

5

10 /6. = 15 20

25 04  (dB) loss Insertion 30

35 Frequency (THz) 193.55 193.6 193.65 193.7 193.75 193.8 193.85 0

5 6 = 10 15 /4 20 25 Insertion loss (dB) 30

35 Frequency (THz)

193.55 193.6 193.65 193.7 193.75 193.8 193.85 0

5

10

60 = 15 .4 20 25 Insertion loss (dB) 30

35 Frequency (THz)

193.55 193.6 193.65 193.7 193.75 193.8 193.85 0

5 /6 = 10 15 4 20 25 Insertion loss (dB) 30

35 Frequency (THz)

193.55 193.6 193.65 193.7 193.75 193.8 193.85 0

5 6 = 10 15

4. 20

25 Insertion loss (dB) loss Insertion

30

35 Frequency (THz)

193.55 193.6 193.65 193.7 193.75 193.8 193.85 0

5

 6/ = 10 4/ 15 20

25 Insertion loss (dB) 30

35 Frequency (THz)

193.55 193.6 193.65 193.7 193.75 193.8 193.85 0

5 6 = 10 15 40 20 25 Insertion loss (dB) 30

35 Frequency (THz)

193.55 193.6 193.65 193.7 193.75 193.8 193.85 0

5 06 = 10 15 41 20 25 Insertion loss (dB) 30

35 Frequency (THz)

193.55 193.6 193.65 193.7 193.75 193.8 193.85 0

5 .6/ = 10 15 4 20 25 Insertion loss (dB) loss Insertion

30

35 Frequency (THz)

Figure 2.21 Asymmetry tuning at different temperatures.

39 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

The Odd and Even passbands at each temperature are plotted together in Figure 2.21. Flat-top profiles are observed at wider passbands. There is no significant insertion loss variation relating to the asymmetry ratio as long as the passband is 30% or wider. The temperature response curve for the TAI is plotted on Figure 2.22. The centre frequency of one passband is used for Y-axis. A good linear response is observed. The slope shows that each degree of tuning corresponds to about 688 MHz tuning. Since the thermal controllers available at present can easily provide 0.1°C temperature resolution and accuracy, frequency tuning resolution of less than 0.1 GHz and centre frequency accuracy of within ±0.05 GHz can be achieved. This level is satisfactory for DWDM applications with 50 GHz or higher channel spacing (less than 0.1% frequency jitter).

192.945

192.94

192.935

192.93

192.925

192.92

192.915

192.91 Center frequency (THz frequency Center 192.905

192.9

192.895 0 10203040506070 Temperature (degree C)

Figure 2.22 Temperature response of the TAI unit.

The tuning time for the experimental prototype is several seconds. It is due to the less efficient temperature controller circuit used here. The tuning time can be reduced to less than a second with higher power source and faster feedback. The experimental results above show that the constructed TAI prototype can deliver the novel functions and satisfactory optical performance for network applications. Some of the parameters, such as the passband profile and the CD, can be improved by using other types of interleaver technology without changing the basic TAI structure.

40 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

2.2.3 TAI Design 2: Based on Programmable Optical Processor a. Design Principle The second method to implement the TAI is to use a liquid crystal on silicon (LCoS)-base programmable optical processor to achieve tunable asymmetric interleaving of the DWDM signals. LCoS is originally a display technology which combines liquid crystal and semiconductor technologies, to create a solid-state display engine with up to WUXGA resolution. It has been proposed that LCoS can be employed to control the phase of light at each pixel to produce beam-steering since almost two decades ago [82], but it is only in the past few years that this technology was applied for optical communications. The key element in an LCoS-based programmable optical processor is a two- dimensional array of LCoS pixels, each pixel being addressable by an analogue AC voltage to control the phase of light reflected from that pixel. This control process is to create a linear optical phase retardation in the direction of the intended deflection. Through this phase pattern control of the LCoS pixels in the two-dimensional array on which the dispersed DWDM signal (an expanded optical beam) is projected, the optical signal can be steered or attenuated to obtain the target passband profiles. By using a large number of phase steps, a highly efficient, low insertion loss switch can be created [83, 84]. The LCoS technology has been used to construct wavelength-selective switch (WSS). It has also been used for dispersion trimming [85] and pulse burst generation [86]. In this work, a new application is proposed for the LCoS-based optical processor, that is, to construct a TAI and obtain asymmetrically interleaved passbands. The LCoS-based TAI can be considered as a special LCoS-based WSS where the port count is 1×2 and the channels widths are not identical. Similar to the operation principle of an LCoS-based WSS, the input DWDM signal from the LCoS-based TAI is dispersed through a diffractive grating and projected onto the centre region of a XGA resolution 2-D LCoS array with the dimension of 1024×768 pixels. This centre effective region covers about 750×500 pixels. The physical dimension of each LCoS pixel is 13×13 microns. The longer axis of the 2-D array is used as the wavelength axis, while the shorter axis performs switching. Along the wavelength axis, each column of pixel corresponds to about 7 GHz frequency range. However this is not the resolution limit of this device since sub-column phase control is achievable. Therefore the achievable

41 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

actual resolution is about 100 MHz. Based on the target asymmetric passband profiles for each output port, the corresponding phase pattern (also called a phase map) is calculated for the entire effective LCoS region. The phase level is represented by an 8- bit digital signal for each pixel, giving a total of 256 possible levels. The corresponding analogue AC voltages are applied to the effective pixels to realize the phase map and thus realize the target passbands at both output ports. b. Implementation and Asymmetry Passband Profiles An LCoS-based programmable optical processor from Finisar (WaveShaper 4000E) is used in this experiment to generate the desired asymmetric interleaver output profiles. Corresponding software developed by Finisar is used to calculate the corresponding phase map for each interleaver configuration and to control the voltages of the pixels on the LCoS engine. Figure 2.23 shows one passband at the Odd Output with the interleaving ratio tuned from 10% to 90% in 5% increments. Figure 2.24 shows the passbands of the two interleaver outputs at different asymmetric interleaving ratios. Both figures are for a 50G/100G TAI. They are measured using Ando AQ6317B optical spectrum analyser (OSA) with 0.1 nm resolution bandwidth.

0 90%

-20 10% Normalized intensity (dB) intensity Normalized -40

1547.0 1547.5 1548.0 Wavelength (nm) Figure 2.23 A passband from the TAI’s Odd Output at interleaving ratios from 10% to 90% in 5% increments.

42 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

Odd Odd Odd 10%:90% Even 20%:80% Even 30%:70% Even 0 0 0

-20 -20 -20

-40 -40 -40

1546 1547 1548 1549 1546 1547 1548 1549 1546 1547 1548 1549

Normalized intensity (dB) intensity Normalized Wavelength (nm) (dB) intensity Normalized Wavelength (nm) (dB) intensity Normalized Wavelength (nm)

Odd Odd Odd 40%:60% 50%:50% 60%:40% 0 Even 0 Even 0 Even

-20 -20 -20

-40 -40 -40

1546 1547 1548 1549 1546 1547 1548 1549 1546 1547 1548 1549

Normalized intensity (dB) intensity Normalized Wavelength (nm) (dB) intensity Normalized Wavelength (nm) (dB) intensity Normalized Wavelength (nm)

Odd Odd Odd 70%:30% Even 80%:20% Even 90%:10% Even 0 0 0

-20 -20 -20

-40 -40 -40

1546 1547 1548 1549 1546 1547 1548 1549 1546 1547 1548 1549

Normalized intensity (dB) Wavelength (nm) Normalized(dB) intensity Wavelength (nm) (dB) intensity Normalized Wavelength (nm)

Figure 2.24 Passbands of the Odd Output and Even Output at interleaving ratios from 10%:90% to 90%:10% in 10% increments.

In this system with 50 GHz channel spacing, each passband at the interleaver output has a Gaussian profile at low interleaving ratios (< 20%) and super-Gaussian behaviour at higher interleaving ratios. The super-Gaussian profile order increases from 2 to 4 as the interleaving ratio increases from 20% to 90%. The excess insertion loss from the TAI is about 3.8 dB. For the special case of 50%:50% interleaving ratio (that is, symmetric interleaving), the passband profiles of this LCoS-based TAI are similar to the passband profiles of a standard symmetric 50G/100G interleaver, and therefore there is no passband width penalty due to the additional asymmetric interleaving function. Comparing with the passbands of the TAI prototype constructed using the cascaded symmetric interleavers method, this LCoS-based TAI has steeper passband edges similar to the conventional symmetric interleaver. This is mainly because the other prototype was constructed using interleavers designed for 100G/200G applications and thus has less steep rising and falling edges.

43 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

Figure 2.25 shows the measured PDL of the LCoS-based TAI. These PDL profiles are also measured by the Agilent 81910A photonic all-parameter analyser. The plotted PDL figures are the average value of the maximum PDL within a defined clear channel passband across all DWDM channels within the operation spectrum. The defined clear channel passbands are centred on the standard ITU-T grid. The PDL is below 0.1 dB within a 25 GHz clear channel passband when the interleaver passband is tuned to 50 GHz (50%). The measured results show that for any specific clear channel passband, the PDL reduces as the interleaved passband ratio increases. For any specific interleaved passband ratio, the PDL increases as the defined clear channel passband widens. But when the interleaved passband significantly exceeds the defined clear channel passband, the variation of PDL is small. Another observation is that when the clear channel passband is defined to be the same as the interleaver passband, the PDL remains at about 0.65 dB. These results show that the LCoS-based TAI has stable and small polarization dependency, thus it is suitable to be used in the DWDM transmission system.

7

r Clear channel passband 6 10G 20G 30G 40G 5 50G 60G 70G 4 3 2 1 channel passband (dB) passband channel Max PDL within the Max PDL within clea 0 0 20 40 60 80 100 Interleaver passband ratio (%)

Figure 2.25 PDL of the LCoS-based TAI at different interleaved ratios with respect to different clear channel passbands.

Because the WaveShaper is a bench-top instrument designed for other laboratory applications and not optimized for interleaving function, the configuration time is slow. It takes several seconds to change from one asymmetric interleaving setting to another one. Within this period, the actual switching operation of the LCoS engine takes only below 20 ms, while most of the time is for electronic processing. If 44 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

the LCoS-based optical processor is specifically designed for TAI application, the pixel map calculation for complex functions (such as filter passband contouring and dispersion compensation) can be eliminated and the electronic processing time can be significantly shortened to several tens of ms due to simplified process.

2.2.4 Comparison between the Two TAI Designs Experimental verifications for both proposed TAI design methods can deliver the interleaving ratio asymmetry and full range tunability features required for the TAI. They can both be implemented in the DWDM network application. However there are some differences between these two methods in the aspects of optical characteristics, tuning performance and implementation issues. a. Insertion Loss For the TAI constructed using the cascaded symmetric interleavers method, the peak insertion loss figures of the passbands are about 1.25 to 1.5 dB for the two interleaver components respectively. The combined TAI has peak insertion loss of about 3 dB, including the connectors at both ends and between two interleaver stages. For product implementation, these two interleaver components will be integrated inside one package and no fibre connectors and adapters are required in between. This can reduce the insertion loss further. 1.5 dB maximum insertion loss for the TAI without connectors can be easily achieved. The TAI constructed using the optical processor method has higher insertion loss value (the measured excess insertion loss is 3.8 dB). This is due to the requirement to disperse, process and recombine the light. The typical insertion loss figure for commercial LCoS-based optical processors is about 5-6 dB. Therefore the first design has lower insertion loss. b. Polarization Dependent Loss Figure 2.26 shows the PDL profiles of one passband of the 50G/100G TAI constructed using the two proposed methods with both TAIs set to the symmetric 50%:50% interleaving ratio. The maximum PDL values within the 37.5 GHz passband centred at the ITU-T grid are 0.19 dB and 0.27 dB respectively. Similar results are obtained from other channels.

45 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

(a) (b) Figure 2.26 PDL profiles (red) of TAI constructed using : (a) Cascaded symmetric interleavers method; and (b) LCoS-based optical processor method. (The grey curves are the passband profiles for reference). c. Chromatic Dispersion Figure 2.27 shows the chromatic dispersion (CD) profiles of one passband of the 50G.100G TAI constructed using the two proposed methods at symmetric 50%:50% interleaving ratio. The CD range for the TAI constructed using the cascaded symmetric interleaver method is between -40 and +51 ps/nm within the 37.5 GHz passband centred at the ITU-T grid. For the TAI constructed using the optical processor method, the CD range is between -8.4 and +12.4 ps/nm within the same range. These figures show that the LCoS-based method causes less chromatic dispersion.

(a) (b) Figure 2.27 CD profiles (red) of TAI constructed using : (a) Cascaded symmetric interleavers method; and (b) LCoS-based optical processor method. (The grey curves are the passband profiles for reference).

46 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

d. Differential Group Delay Figure 2.28 shows the differential group delay (DGD) profiles of one passband of the 50G/100G TAI constructed using the two proposed methods at symmetric 50%:50% interleaving ratio. DGD is the same as the first order PMD. Averaging the DGD values across a spectrum will provide the PMD figure within that spectrum. For the observed channel, the maximum DGD values within the 37.5 GHz passband centred at the ITU-T grid are 0.15 ps and 1.4 ps respectively. Averaging the DGD values within this spectrum gives 0.06 ps and 0.52 ps PMD respectively. Similar results are obtained from other channels. These results show that LCoS-based method has larger (worse) PMD.

(a) (b) Figure 2.28 DGD profiles (red) of TAI constructed using : (a) Cascaded symmetric interleavers method; and (b) LCoS-based optical processor method. (The grey curves are the passband profiles for reference). e. Tuning Speed In terms of the tuning speed, LCoS-based optical processor method has potentially faster configuration time than the cascaded symmetric interleavers method which requires thermal or mechanical tuning of components, if the pixel map calculation is simplified for the TAI application only. f. Passband Profiles In terms of the interleaver output passband profile, the LCoS-based optical processor method allows the user to modify to a certain degree, such as changing the passband slope and even adding some equalization functions, while the passband slopes of the TAI based on the cascaded symmetric interleavers method cannot be changed.

47 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

The optical processor also allows the asymmetry to be implemented on only a portion of the spectrum, or different asymmetry ratios implemented on the same DWDM system. Even though there is no much application to this feature right now, it might be useful if a network provider chooses to upgrade the DWDM system in the order of transmission band, or to merge two different DWDM systems. This feature cannot be provided by the cascaded symmetric interleavers method. g. Control Complexity On the other hand, the lack of such flexibility also means that the control of the cascaded symmetric interleavers method is much simpler. It does not require electronic microprocessor in the TAI, unlike in the LCoS-based optical processor. h. Hardware Size The cascaded symmetric interleavers method also has the size advantage due to the simpler design. It does not require the long free space optical path to fully disperse the DWDM spectrum, as the LCoS-based optical processor does. i. Cost Due to the simpler control and optics, the TAI based on the cascaded symmetric interleavers method has lower hardware cost and operation cost.

2.3 Experimental Demonstration of TAI 2.3.1 Simulation of TAI for 10G/40G Hybrid System Upgrade The performance of TAI in 10G/40G hybrid DWDM system upgrade application is firstly studied by simulation, which allows the study of various modulation formats. a. Simulation Tool Commercial optical simulation software package from VPIsystems is used for the DWDM transmission simulation. VPIsystems’ software products are widely used in optical communication research and development for modelling and verification. They also help the network operators, equipment vendors and equipment designers in network planning, design automation and optimisation. In this study, a VPIsystems

48 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

product called VPItransmissionMaker (VPI-TMM) is used. It offers the ability to design advanced systems including novel modulation schemes, PMD compensation, Raman amplification, partial regeneration, adaptive dispersion compensation, optical channel monitoring and power flattering. It has a large library of photonic and electronic modules supporting various system concepts. Its graphical interface allows the users to build and simulate many topologies using Sample and Block modes, both in unidirectional and bi-directional simulations. It also has advanced multiple-signal representations and numerical models for speed and accuracy. Besides using the standard interleaver data provided by the component library of the simulation software, the optical performance data measured on the actual TAI prototype are used for the simulation. This will ensure that the simulation is more accurate and the results will be able to reflect the impacts from the actual characteristics of the device. The VPI-TMM software offers various parameters for transmission analysis, such as the quality factor (commonly known as the Q-factor), eye diagram, eye closure penalty, and so on. For this study, vertical eye closure value at the receiver is firstly used. It is defined as the vertical opening in the centre of an eye diagram and indicates the amount of voltage available to sample the signal. This is a good indicator of the combined signal degradation caused by the effects such as crosstalk and inter-symbol interference. To single out the signal distortion due to the interleaver effect alone, other noises in the optical link are turned off. Timing jitter that leads to horizontal eye closure is also assumed to be zero. To quantify the vertical eye closure level, the eye closure value (ECV) is calculated. This is a relative value and therefore use arbitrary unit. Higher figure indicates larger eye opening and better signal quality. However, the vertical eye closure alone cannot provide all the information about the signal quality. Q-factor is also recorded for each simulation scenario in the analysis. Besides the vertical signal level, the Q-factor also includes the noise information, which it assumes to be a normal distribution. It is used to predict the minimum bit error rate (BER) and the threshold voltage at that time. b. Simulation Model and Parameters Figure 2.29 shows the 10G/40G DWDM transmission model designed for the simulation. The optical link consists of six DWDM channels in C-band with 50 GHz channel spacing. Among these 6 channels, three 10 Gbit/s channels and three 40 Gbit/s 49 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

channels are alternating. The centre frequencies for the 10 Gbit/s channels are 192.9 THz, 193.0 THz and 193.1 THz respectively. The centre frequencies for the 40 Gbit/s channels are 192.95 THz, 193.05 THz and 193.15 THz respectively. To minimize the crosstalk between adjacent channels, orthogonal polarization states are assigned between 2 neighbouring channels.

100 GHz 10G receiver AWG (Even) 100 GHz 3x 10G AWG (Even) channels

TAI at the receiver

3x 40G 100 GHz TAI at the channels AWG (Odd) transmitter 100 GHz AWG (Odd) 40G receiver Figure 2.29 Simulation model of the 10G/40G DWDM transmission system.

NRZ OOK is used for the 10 Gbit/s channels. This is the standard modulation format for 10 Gbit/s transmission in most of the current DWDM networks. Because of the sufficient bandwidth available for 10 Gbit/s signal, advanced modulation schemes are not required. Various modulation schemes are used for the 40 Gbit/s channels, including basic NRZ OOK, 33% and 67% RZ-DPSK, 33% and 67% RZ-DQPSK, and duobinary. These schemes are widely studied in recent years for 40 Gbit/s transmission networks because of their relaxed optical signal-to-noise ratio (OSNR) requirements and/or high spectral efficiency [87]. The DWDM signals from the 10 Gbit/s channels and the 40 Gbit/s channels are multiplexed using 100 GHz AWG filter respectively and then combined using the TAI. Actual data measured from an NEL AWG multiplexer device are used in the simulation. The measured 3 dB passband width from the peak is about 84 GHz. For the TAI, the interleaving ratio between two output ports is varied from 50%:50% to 80%:20%, in steps of 10%. In the cases of asymmetric interleaving, the wider passbands are assigned to the 40 Gbit/s channels. The optical spectrum of the interleaved signal can be 50 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

observed (as shown on Figure 2.30). Back to back transmission is used in the simulation to exclude other influences in the optical transmission so that the contributing effect of the interleaver can be singled out. At the receiver end, an interleaver (de-interleaver) with the same characteristics as the input interleaver is used to separate 10 Gbit/s and 40 Gbit/s channels. They are then demultiplexed by another

10G: NRZ, 40G: NRZ 10G: NRZ, 40G: CS-RZ

10G: NRZ, 40G: 33% RZ-DPSK 10G: NRZ, 40G: 67% RZ-DPSK

10G: NRZ, 40G: 33% RZ-DQPSK 10G: NRZ, 40G: 67% RZ-DQPSK

10G: NRZ, 40G: Duobinary

Figure 2.30 Simulated optical spectra with various types of modulation schemes. 51 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

NEL AWG filter into individual DWDM channels. The eye pattern at each output channel is recorded, and the Q-factor and eye closure value are calculated. c. Simulation Results and Analysis Among the six DWDM channels used in the simulation, the results at two centre channels are analysed, including a 10 Gbit/s channel centred at 193.0 THz and a 40

12

10 10G NRZ NRZ 8 CS-RZ 33% DPSK 67% DPSK 6 33% DQPSK I 33% DQPSK Q 4 67% DQPSK I 67% DQPSK Q Duobinary Eye closure(a.u.) value Eye 2

0 40 50 60 70 80 90 Passband interleaving ratio for 40G channel (%)

Figure 2.31 Eye closure values of signals with different modulation schemes as a function of interleaving ratio for the 40 Gbit/s channel.

16 180

14 160 NRZ 140 12 CS-RZ 120 33% DPSK 10 67% DPSK 100 33% DQPSK I 8 80 33% DQPSK Q 6 67% DQPSK I 60 10G Q Q 10G (dB) factor 67% DQPSK Q 40G Ch Q factor (dB) Ch factor Q 40G 4 Duobinary 40 10G NRZ 2 20

0 0 40 50 60 70 80 90 Passband interleaving ratio for 40G channel (%)

Figure 2.32 Q-factor values of signals with different modulation schemes as a function of interleaving ratio for the 40 Gbit/s channel.

52 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

Gbit/s channel centred at 193.05 THz. The side channels are neglected because they do not have sufficient DWDM channels at both sides to emulate performance in multi- channel DWDM systems. The measured relative ECVs and Q-factor values of 10 Gbit/s channel and 40 Gbit/s channels are summarized and plotted on Figures 2.31 and 2.32. The results show that as the interleaver passband width increases, the eye closure value for the 40 Gbit/s channel also increases and the Q-factor improves. This is observed for all modulation schemes, including the more bandwidth efficient schemes such as duobinary and DQPSK. For these two schemes, the benefit of wider interleaver passband diminishes after 70% ratio, because the filtering effect is no longer the main contributing factor to the signal distortion. On the other hand, the narrowing of the interleaver passband does not cause any signal degradation effect on the 10 Gbit/s NRZ- OOK signal until the passband width falls below 30%. Below are some conclusions obtained from the simulation: • Interleaver is necessary for combining the alternating 100 GHz spaced 10 Gbit/s DWDM signals and 100 GHz spaced 40 Gbit/s DWDM signals in the channel density doubling upgrade, particularly when the modulation scheme for the 40 Gbit/s channel does not have good bandwidth efficiency. This is because of the inter-channel crosstalk between the neighbouring channels. Between the 10 Gbit/s and 40 Gbit/s channels, the 10 Gbit/s channels suffer high level of distortion from the crosstalk. Having interleaver significantly improves the signal quality of the 10 Gbit/s channels. As for the 40 Gbit/s channels, there are mixed results from the insertion of interleaver. • While reducing the inter-channel crosstalk, the interleaver also brings filtering effect on each DWDM channel, this again is more significant for 40 Gbit/s modulation scheme does not have good bandwidth efficiency. • Asymmetric interleaver passband ratio between the 40 Gbit/s and 10 Gbit/s channels leads to better signal quality at the receiver for all modulation schemes, including those with better bandwidth efficiency such as DQPSK and duobinary. • The 40 Gbit/s signal quality improves as its interleaver passband widens, while 10 Gbit/s signal does not suffer any degradation until its passband is narrowed to below 30%. Therefore the best splitting ratio is around 70%:30% between the 40 Gbit/s channels and the 10 Gbit/s channels. (This ratio will vary slightly with the different modulation schemes studied.) The eye patterns of the received signal for selected

53 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

modulation schemes with 50%:50% versus 70%:30% interleaving ratios are shown in Table 2.1.

Table 2.1 Eye patterns, ECVs, and Q-factors for different modulation schemes at 50%:50% and 30%:70% interleaving ratios.

TAI interleaving 50%:50% 70%:30% ratio Mod. Eye ECV Q-factor ECV Q-factor Bit rate Eye diagram scheme diagram (a.u.) (dB) (a.u.) (dB)

NRZ- 4.92 5.31 8.04 9.23 OOK

CS-RZ 1.91 3.23 8.81 5.34

33% 5.42 5.83 7.09 4.51 DPSK

40 66% 5.42 4.96 7.78 5.08 Gbit/s DPSK

33% 4.90 7.85 7.28 11.08 DQPSK

66% 4.78 8.22 6.13 12.79 DQPSK

DB 5.83 6.31 7.56 12.79

10 NRZ- 10.1 153.55 10.2 28.79 Gbit/s OOK

54 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

• Due to the adjacent channel isolation provided by the interleaver, the quality of the 10 Gbit/s signals does not vary regardless of the modulation scheme used for the 40 Gbit/s channels. • Different 40 Gbit/s modulation schemes receive different levels of signal quality improvement from the asymmetric characteristic of the interleaver. For signals that have wider spectral width and thus are more sensitive to the filtering effect at the interleaver (such as NRZ, CS-RZ and DPSK, particularly CS-RZ), the asymmetric interleaver brings up the greatest improvement. For signals with better spectral efficiency and are less sensitive to the filtering effect by nature (such as DQPSK and duobinary, particularly 67% RZ-DQPSK), the original signal qualities are better even with symmetric interleaver, therefore the level of improvement from the asymmetric interleaver is limited.

2.3.2 Experimental Demonstration of TAI for 40G/100G Hybrid System Upgrade In the past 2-3 years, 100 Gbit/s transmission technologies have been widely studied and test in the field [8-9]. Network upgrades from 40 Gbit/s systems to 100 Gbit/s are thus expected to take place in the next few years. Different from the 10G/40G hybrid system, both the 40 Gbit/s channels and the 100 Gbit/s channels use advanced modulation formats other than the conventional NRZ-OOK. For the 100 Gbit/s channels, more advanced modulation formats and multiplexing schemes, such as polarization multiplexed QPSK, are used to achieve spectral efficiency greater than 2 bit/s/Hz so that the 100 Gbit/s signal can be accommodated in a 50 GHz spectrum. Here I experimentally studied how TAI improves the overall system performance despite different modulation formats and signal bandwidths among the DWDM channels in the 40G/100G hybrid system [88]. a. Experimental Setup The experimental setup is shown on Figure 2.33. To form the Odd Channels, four Emcore TTX19900 external cavity laser (ECL) diodes with frequencies ranging from 193.15 THz to 193.45 THz and linewidth smaller than 100 kHz are combined using an AWG multiplexer with 100 GHz channel spacing and a 50 GHz offset from the ITU-T 100 GHz grid. All four DWDM channels are modulated in the same way for 40 Gbit/s transmission. Here, both 42.8 Gbit/s NRZ-DPSK and 42.8 Gbit/s RZ-DPSK

55 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

100 GHz AWG with 50GHz offset LD 1 LD 3 40G 40G or LD 5 modulator Odd 100G LD 7 50G/ Receiver 100G EDFA TOF VOA LD 2 TAI OSA LD 4 100G Even LD 6 modulator 120Gb/s PDM-RZ-8PSK LD 8 -10 100 GHz -20 AWG (a) -30 42.8 Gb/s NRZ-DPSK -40 Intensity (dB) 70%:30% Resolution: -50 30%:70% 0.1 nm 1549 1550 1551 1552 Wavelength (nm)

Figure 2.33 Experimental setup of TAI in 40G/100G hybrid system. Inset (a): Spectrum of hybrid 42.8 Gb/s NRZ-DPSK and 112 Gb/s PDM-RZ-8PSK signal. modulation schemes are studied separately. In the NRZ-DPSK modulator, a Fujitsu

FTM7938EZ-A LiNbO3 Mach-Zehnder modulator (MZM) in a push-pull configuration biased at the null point with ±V4 drive swing for both arms is used to generate the DPSK-modulated signal, where the required 42.8 Gbit/s electrical NRZ signal is obtained by using electrical multiplexing of four 10.7 Gbit/s data using Centellax MS4S1V2M 4×1 multiplexer. The RZ-DPSK modulator has an additional dual drive Fujitsu MZM after the first MZM, which is driven by a full symbol-rate clock (single- ended drive, biased at 1.5 V4) to generate 50% RZ pulses. For the Even Channels, additional four Encore ECL diodes with frequencies from 193.10 to 193.40 THz are combined by an ITU-T 100 GHz grid centered AWG multiplexer and modulated together with the 100 Gbit/s signal. Here, I study both 112 Gbit/s Polarization Division Multiplexed (PDM)-RZ-QPSK and 120 Gbit/s PDM-RZ-8PSK modulation schemes. The PDM-RZ-QPSK modulator consists of one Fujitsu FTM7938EZ-A MZM, two

EOSpace PM-DV5-40-PFU-PFU-LV 40 Gbit/s LiNbO3 phase modulators (PM), an Amonics AEDFA-PM-23-B polarization-maintaining EDFA, and a polarization

56 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

multiplexing unit. The two phase modulators are each driven by a 28 Gbit/s data stream to provide 0/4 and 0/0.54 phase modulation respectively. The MZM is driven by a 28 GHz clock to carve out 50%-duty-cycle RZ pulses. The 28 Gbit/s data is obtained by time-multiplexing four 7 Gbit/s PRBS signals (each with a pattern length of 211-1). The two 28 Gbit/s data signals are de-correlated by introducing a differential bit delay, producing a pattern length of 213-4 for the 28 GBaud signal. The polarization- multiplexing is achieved by dividing and recombining the signal after a 322 symbol delay using an Opto-Link polarization beam combiner. The 120 Gbit/s PDM-RZ-8PSK modulator contains an additional PM, with each driving data stream for the three phase modulators set to 20 Gbit/s, and the MZM driven by a 20 GHz clock. The Odd Channels and the Even Channels are combined using the LCoS-based TAI described eavlier, with the interleaving ratio changing from 20%:80% to 80%:20%. The inset of Figure 2.33 shows the WDM spectra for the interleaved 42.8 Gbit/s NRZ- DPSK and 112 Gbit/s PDM-RZ-8PSK signals at 70%:30% 30%:70% interleaving ratios measured on Ando AQ6317B OSA at 0.1 nm resolution bandwidth. The filtering effect caused by the narrowing of the interleaver passband can be observed. Same as the simulation, only back-to-back transmission is studied to single out the effect of the TAI. An Amonics AEDFA-DWDM-23-B EDFA is used to amplify the interleaved signal. A Santec OTF series manual tunable optical filter with 1 nm passband width is tuned to the channel under study. An Eigenlight Series 400 variable optical attenuator (VOA) is used to vary the OSNR of the received signal, which is measured by the OSA. An Optoplex delay interferometer-based DPSK demodulator and a pair of u2t BPDV2020R balanced receivers are used to receive the 42.8 Gbit/s NRZ-DPSK and the RZ-DPSK signals. An Anritsu MP1764A bit error rate tester (BERT) is used to calculate the bit error rate. For the 100G signals, a digital coherent receiver is used. It consists of an Optoplex polarization diverse 90-degree optical hybrid, another Emcore ECL as the local oscillator, and four single-ended photo detectors from u2t. The sampling and digitization functions are achieved with a Tektronix DSA 71604 4-channel real time sampling scope with a 50 GSa/s sampling rate and 16 GHz analog bandwidth. The captured data is then post-processed using a desktop computer to calculate the BER. The full description for the receiver structure and DSP algorithms used for post-processing are discussed in more detail in [17].

57 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

b. Experiment Results and Analysis By adjusting the VOA, the receiving signal’s OSNR level is set to deliver 2×10-3 BER for each modulation scheme at the various interleaving ratios. Figure 2.34 shows the OSNR requirements as a function of interleaver passband width. The results show that the 42.8 Gbit/s NRZ-DPSK signal starts to degrade when the passband is narrower than 65 GHz (65% interleaving ratio in the 50GHz/100GHz interleaver) and has about 3 dB OSNR penalty if the passband is reduced to 50 GHz (which is the case of the conventional symmetric interleaver). The signal quality deteriorates significantly when the passband narrows further. The 42.8 Gbit/s RZ-DPSK signal shows a similar trend, but has about 3 dB better OSNR tolerance compared to the NRZ-DPSK signal. This is expected since the 50% duty cycle RZ signal doubles the receiver sensitivity, when the average optical power remains the same. However RZ-DPSK signal quality starts to degrade when the passband is narrower than 80 GHz because its signal spectrum is wider. These results show that these two 40G modulation schemes require an interleaver passband that is wider than 50% to obtain good filtering performance.

42 .8 Gb /s NRZ- DPSK 30 42 .8 Gb /s RZ- DPSK 28 11 2 Gb/ s PDM-RZ- QPSK 120 Gb/s PDM-RZ-8PSK 26 24 22 20 18 16 14 20 40 60 80 100

OSNRrequirement for 2e-3 BE(dB) R Interleaver pass band width ( GHz) Figure 2.34 OSNR requirement as a function of interleaver passband width.

The 112 Gbit/s PDM-RZ-QPSK signal demonstrates a flatter OSNR tolerance behavior. There is only a 1 dB OSNR penalty between the optimum passband (at 50 GHz) and the worst cases when the passband varies from 25% to 100% (all pass for the 100G channels, and blocking the 40G channels). The signal begins to degrade when the passband narrows down to 20 GHz, resulting in an OSNR penalty of 2 dB compared to the case of a 25 GHz passband. This happens because the PDM-RZ-QPSK modulation 58 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

scheme has high spectral efficiency. These results also show that there is a flat region between the 30 GHz to 35 GHz passband range. The reason behind this effect is that after a 30 GHz optical filter, the ratio of the signal power that can be detected by the oscilloscope is increased because of the limited bandwidth of the oscilloscope compared to the case of a 35 GHz filter, so the OSNR requirement is reduced. However, the filtering effect is also increased, so the OSNR requirement will be increased. The coexistence of these two opposite effects makes the OSNR requirements for the 30 and 35 GHz passbands close. As the passband widens beyond 50 GHz, the signal quality degrades slightly due to crosstalk from neighboring channels. The 120 Gbit/s PDM- RZ-8PSK signal shows a similar trend but has higher OSNR requirements. Moreover, the flat region is now between 35 GHz and 40 GHz. The signal degradation starts to occur when the passband is narrower than 35 GHz. These results show that the 100G PDM-RZ-mPSK system can tolerate interleaver passbands narrower than 50%. Using this information, when the network is upgraded from a 40G to a hybrid 40G/100G system, the interleaving ratio can be tuned using the TAI to deliver the optimum passbands between the odd and even channels. For example, if the odd channels are modulated with 42.8 Gbit/s NRZ-DPSK signals and the even channels with 112 Gbit/s PDM-RZ-QPSK signals, a 65%:35% interleaving ratio would provide the best combined transmission performance. With the continuous tuning capability demonstrated here, the interleaving ratio can also be adjusted further as the network is upgraded, such as to 50%:50% when all channels are upgraded to 100 Gbit/s. Unlike the conventional interleaver which is completely passive, the TAI requires active control to set the target asymmetry ratio. This will potentially increase the complexity of the transmission systems. However, since the asymmetry ratio adjustment is only required during the network upgrades, which occur only once every several years, the additional control complexity caused by TAI to the network operator is minimum.

2.4 Conclusions In this chapter, I proposed a novel tunable asymmetric interleaver and implemented it using two different methods. It is the first optical interleaver to allow arbitrary asymmetric interleaving ratios with continuous and in-service tuning capability. Two implementation methods are proposed and experimentally

59 )%, 0)%, 0 .*% +((%),! )%,%1%, .*% +((%),! )%,%1%,

demonstrated. The first method uses two cascading symmetric interleavers, and the second method uses a programmable optical processor. The optical characteristics and performance of the two prototypes were analysed and compared, both prototypes met the targeted performance and are suitable for application in the actual DWDM system. The results from theoretical analysis, simulations, and experiments demonstrate that this new device can allow smooth and future-proof upgrade for the DWDM transmission system from 10 Gbit/s to 40 Gbit/s to 100 Gbit/s or higher. This work has produced two papers in major international optical communication conferences [80, 88]. A US patent has recently been issued [81].

60 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

)%, )%, 222      

           

  6 6  7 8

Inter-symbol interference (ISI) is a transmission impairment that is caused by narrowing of the filter passband. It becomes more significant as the DWDM channel spacing narrows and data baud rate increases. In this chapter, I propose a novel colourless optical intra-channel equalizer which can mitigate the ISI in all channels simultaneously. The principle is to design the periodic filter so that the overall filter profile becomes raised cosine. A prototype is designed and implemented using thin-flim F-P interferometer. Experiments show that this equalizer widens the passband width by 20%, and improves the opening of the received signal’s eye by 36~40%. Therefore it can effectively mitigate ISI in the DWDM transmission link. Since this equalizer is just a passive athermal filter and only one unit is required in the entire transmission link, this solution is cost effective too.

3.1 Background 3.1.1 ISI in DWDM Transmission As discussed in the previous chapter, one common way to increase the DWDM data capacity in each fibre is to reduce the channel spacing and thus increase the number of DWDM channels within the same transmission band. The typical DWDM channel spacing is changing from 100 GHz a decade ago to 50 GHz at present, and 25 GHz channel spacing system are being explored and proposed [3, 17]. This creates a filter narrowing problem. When the bit rate per channel in DWDM systems is 10 Gbit/s or lower, the optical signal spectral width is much smaller than the ITU channel spacing of 50 GHz. But when the per-channel bit rate is increased to 40 Gbit/s, the optical spectral width for the 33% duty cycle RZ-DPSK (a popular modulation format for 40

61 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

Gbit/s system due to its balance of system complexity and 3 dB receiver sensitivity improvement [89, 90]) signal is about 60 GHz for 3 dB bandwidth or 100 GHz for 10 dB bandwidth. The optical multiplexing/demultiplexing elements designed for 50 GHz spacing DWDM applications, such as the arrayed waveguide gratings and optical interleavers, can cause strong optical filtering effect to the 40 Gbit/s DPSK signals. The 50 GHz filtering effect leads to the broadening of the 40 Gbit/s optical signals, which results in the extension of signal energy into the time slots of neighbouring bits. This phenomenon is known as the ISI. The ISI is intrinsic due to the principle of Fourier transform: pulses of a certain ideal shape have a constant time bandwidth product. Generally speaking, narrowing down the spectral width will cause pulse spreading in the time domain. The ISI can cause dramatic increase of signal bit error rate. As illustrated on Figure 2.5, optical AWG multiplexers and interleaver are used to combine multiple optical channels into a 50 GHz-spaced DWDM signal to be transmitted in a single fibre. When the channel is upgraded from 10 Gbit/s NRZ-OOK to 40 Gbit/s DPSK in the 50 GHz-spaced DWDM systems using such architecture, the 40 Gbit/s DPSK signals suffer from strong filtering effects, which results in the signal pulse broadening. At the receiver end, they are no longer distinguishable as well- defined pulses. Instead, the energy from a broadened pulse can leak into the neighbouring bit periods, and cause ISI. Below are more detailed theoretical explanations. A digital modulated signal can be expressed as = − v(t) B In g(t nT ) (3.1) n where In represents the discrete information-bearing sequence of symbols and g(t) is the signal pulse. A baseband communication channel with strong filtering effect can be characterized as a band-limited channel with low-pass frequency response C(f). Its equivalent low pass impulse response is expressed as c(t). If a digital modulated signal is transmitted over a band pass channel, the received signal becomes

+∞ r (t) = v(τ )c(t −τ )dτ + z(t) (3.2) l O−∞ where z(t) is the additive noise. The signal term can also be represented in the frequency domain as V ( f ) ⋅C( f ) , where V(f) is the Fourier transform of v(t). If the channel is band-limited to W Hz, then C(f) = 0 for | f |>W. As a consequence, any frequency components in V(f) above | f | = W will not be passed by

62 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

the channel (or suffer very large attenuation in reality). Within the bandwidth of the channel, we may express the frequency response C(f) as C( f ) =| C( f ) | e jθ ( f ) (3.3) where |C(f)| is the amplitude-response characteristic and θ(f) is the phase-response characteristic. A channel is defined as non-distorting or ideal if the amplitude response |C(f)| is constant for all |f |≤ W and θ(f) is a linear function of frequency. On the other hand, if |C(f)| is not constant for all |f |≤ W , the channel distorts the transmitted signal V(f) in amplitude. And if θ(f) is not linear, the channel distorts the signal V(f) in delay [91]. As a result of the amplitude and delay distortion caused by the non-ideal channel frequency-response characteristic C(f), a sequence of pulses transmitted through the channel at rates comparable to the bandwidth W are spread and overlapped, and thus generating the ISI. As an example of the effect of ISI caused by optical filtering on 40 Gbit/s signals, the optical spectra and eye diagrams of a 43 Gbit/s (the bit rate for OC-768 with Forward Error Correction) DPSK signals are simulated under different optical filtering cases with AWGs and optical interleavers. The AWG filter is a Gaussian type, and its transfer function is

C 2n S D C f − f S T T (f ) = exp − ln 2D c T (3.4) D D f T T E E g U U where fc is the central frequency, and the 3 dB bandwidth is 2fg. In this simulation n is set to 1 for first-order filters. With higher orders of n, the flatness of the passing band can be increased. The intensity transmission curve of the optical interleaver for the simulation is shown in Figure 3.1, which is based on the characteristics of real devices. It has periodic passing bands which have flat top and relatively sharp edges.

Figure 3.1 Measured transmission characteristics of an optical interleaver.

63 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

Figure 3.2 shows the simulated optical spectra and eye diagrams for 43 Gbit/s 33% RZ-DPSK signals. The 33% RZ-DPSK signal has spectral widths of 60 GHz at 3dB, 80 GHz at 5 dB, and 100 GHz at 10 dB. Without any optical filtering, the signal eye diagram has a clear eye opening, as shown in Figure 3.2(a). After the 43 Gbit/s

(a) RZ-DPSK signal without optical filtering

(b) RZ-DPSK signal passing through multiplexing elements (200 GHz Gaussian shape AWG, and 200G/100G optical interleaver) in 100 GHz DWDM system

(c) RZ-DPSK signal passing through multiplexing elements (100 GHz Gaussian shape AWG, and 100G/50G optical interleaver) in 50 GHz DWDM system

Figure 3.2 Simulated optical spectra (left) and eye diagrams (right) of 43 Gbit/s 33% RZ-DPSK signals under different filtering conditions.

64 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

signal passes through an AWG optical multiplexer with 140 GHz 3 dB bandwidth and an 200G/100G optical interleaver in a 100 GHz channel spacing DWDM system, Figure 3.2(b) shows the signal with small degradations due to ISI noise. When the signal passes through similar multiplexing elements in a 50 GHz channel spacing system, the signal suffers serious degradations due to strong ISI (Figure 3.2(c)), since the channel bandwidth limitation is very close to the signal bit rate. The first-order Gaussian transmission characteristic of AWGs does not have a flattop passing band which adds extra non-ideal signal distortions. These results show that ISI has been one of the most challenging issues for high- speed transmission, such as 40G DPSK transmission over 50 GHz-spacing DWDM systems. Another source of filter narrowing effect that leads to the ISI problem is cascading switching elements, such as ROADM or OXC, in the transmission link. As the number of such switching elements increases in the transmission link, the passband of the combined cascading filter narrows [26, 92]. Figure 3.3 shows the narrowing of an ideal flat-top ROADM filter as the number of cascade increases from 1 to 20. In practical scenario, each filter passband is not so ideal and there might be some frequency offset in each device, this will result in even narrower combined passband. Figures 3.4(a)-(d) illustrate the received signal eye diagrams of a 40 Gbit/s DPSK signal as the cascade of 50 GHz-spaced WSS-based OXC nodes increases from 0 to 12 for a single channel. It can be clearly observed that the signal qualify becomes seriously degraded as the number of switching node increases. This ISI impairment significantly limits the transmission distance and thus needs to be mitigated in the next generation DWDM system.

N=1

N=20

Figure 3.3 Passband narrowing by the cascading ROADM filters.

65 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

(a) N=0 (b) N=4

(c) N=8 (d) N=12 Figure 3.4 Eye closure of 40G DPSK signal caused by ISI at different number of cascading 50 GHz WSS filters in the transmission link.

3.1.2 ISI Mitigation Methods There have been various schemes proposed for reducing ISI in band limited channels, including design of partial response signals, modulation codes for spectral shaping, side-band spectral filtering, optical and electrical equalizations, etc. Each of them has its technical and practical advantages, however also has some drawbacks, especially in high-speed DWDM network applications. a. Increase Signal’s Spectral Efficiency In optical transmissions, duobinary (DB) modulation format has been shown to be able to reduce the optical signal spectral width to be around the signal bit rate, for example, about 40 GHz for 40 Gbit/s optical DB signals. However, many simulations and experimental results have shown the DB has limited transmission performance due to its poor tolerance to the nonlinear effects in optical fibre. Optical DQPSK modulation is another modulation scheme which can support long-haul transmission with high spectral efficiency. However, the applications of optical DQPSK modulation

66 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

could be limited due to the relative high complexity of DQPSK transmitters and receivers and small tolerance to laser frequency drift. b. Digital Coding Techniques In digital communications, the power spectral density of digital signals can be controlled and shaped by selecting the transmitted signal pulse and by introducing correlation through coding, which can serve the purposes of tolerating channel distortion and noise, increasing clock components for the ease of synchronization, or reducing the DC components, etc. As an example, run length-limited codes have been demonstrated to reduce signal spectral width for transmission in band limited channels [91]. Although many of the coding schemes can be very powerful to enhance the signal transmission performance with the help of sophisticated electronic signal processing, their implementations are still technically challenging or very expensive at high speed (40 Gbit/s or beyond) for applications in optical communications. c. Pre-filtering For optical band limited channels, some side band pre-filtering methods have been proposed and demonstrated to modify the optical signal spectral width to fit into the channel bandwidth. The common optical signal pre-filtering methods include single-side-band (SSB) filtering [93] and vestigial-side-band (VSB) filtering [94], which can reduce the optical spectral width of the signal by half. The disadvantages for optical SSB or VSB pre-filtering techniques are the increased complexities and compromised transmission performance. d. Electronic Signal Equalization Signal equalization has been widely studied as a powerful tool for mitigating the signal ISI caused by various cases including limited channel bandwidth. It is commonly performed at the electronic domain after the optical signal detection, and is typically based on feed-forward equalizers (FFEs) and/or decision feedback equalizers (DFEs). It has been demonstrated that the electronic signal equalization can increase the inherent tolerance of a receiver to low-pass filtering, as well as other types of impairments such as CD, PMD, and differential mode dispersion (DMD) [95-97]. However, even though such technique is flexible and efficient, its cost increases quickly with the increase of data speed. 67 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

In this chapter, I propose and investigate an optical equalization (OEQ) method as a low cost solution to mitigate ISI caused by filter narrowing in DWDM systems, since the hardware cost of the optical equalization method does not increase with the data rate. A novel OEQ device that can simultaneously suppress ISI in multiple DWDM channels is designed and demonstrated.

3.2 Development and Implementation of Colourless Optical Intra-Channel Equalizer 3.2.1 Operation Principle a. Signal Equalization Using Linear Filters The underlying principle of signal equalization using filter is based on the theorem of Nyquist criteria which states that [91]: the pulse s(t ) satisfies

I1 if l = 0 s(lt) = J (3.5) K0 if l ≠ 0 if and only if the transform S( f ) satisfies

1 ∞ C n S B SD f + T = 1 f ≤ 1/ 2T (3.6) T n=−∞ E T U When the signal pulses satisfy the Nyquist criteria, there is no ISI with proper settings of sampling time. Some popular Nyquist pulses are those whose Fourier transforms follow the shape of raised-cosine. Therefore, the ideal transfer function for a band limited channel is raised-cosine, which will not cause strong ISI for the received signals. If the target transfer function is H ( f ) but the actual channel has a transfer function of H′( f ) which is different fromH ( f ) , a simple method is to cascade with the channel an equalization filter which has a transfer function equal to H ( f )/ H′( f ) , as shown in Figure 3.5. Here the actual channel transfer function H′( f ) appears in the denominator of the equalization filter. Therefore, there will be noise amplification at frequencies at which H′( f ) is small, which will degrade performance. The advantage of such equalization filter is its relative ease of implementation while many other advanced equalization methods rely on complicated algorithm and expensive high- speed electronics.

68 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

Actual Equalization Channel Filter H ( f ) H ′( f ) H ′( f )

Figure 3.5 Equalization filter for ISI suppression.

In optical DWDM transmission systems as shown on Figure 2.6, AWGs and optical interleavers are major factors causing channel bandwidth narrowing. The design of an optical equalization filter can be based on the combined filtering characteristics of the AWGs and interleavers. With an optical equalizer, the overall filtering characteristics should be close to the shape of Raised-cosine. The basic principle is shown in Figure 3.6. In this example, the combination of a Gaussian profile AWG and a flat-top interleaver in the transmission link produces an overall filtering profile with narrow passband (topi right profile). The corresponding OEQ is designed to have a dip top (bottom left profile) to equalize this narrow filtering shape. The resultant profile (bottom right profile) has a raised-cosine profile with wider passband.

AWG Interleaver Overall filtering

+

-1.0 -0.5 0 0.5 1.0 -1.0 -0.5 0 0.5 1.0 -1.0 -0.5 0 0.5 1.0 Frequency Frequency Frequency

Overall filtering after Optical equalizer equalization

+

-1.0 -0.5 0 0.5 1.0 -1.0 -0.5 0 0.5 1.0 Frequency Frequency Figure 3.6 Optical equalization for optical DWDM systems.

The equalization filter can be placed at both the transmitter and the receiver to compensate the filtering from optical multiplexer and demultiplexer, respectively. Without considering the addition of noise, a single equalization filter can be placed at either the transmitter or the receiver to compensate the filtering from both the optical

69 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

multiplexer and demultiplexer. With an OEQ filter included in the channel, the received 43 Gbit/s optical DPSK signal is shown in Figure 3.7. Compared with Figure 3.2(c), the equalized eye diagram in Figure 3.7 has much smaller amplitude jitter at the central sampling point. With good timing control and synchronization, the signal in Figure 3.7 can achieve good BER performance.

Figure 3.7 Optical spectrum and eye diagram of 43 Gbit/s 33% RZ-DPSK signals passing through optical multiplexer, demultiplexer and OEQ filter in a 50 GHz-spaced DWDM system. b. Colourless OEQ for ISI Suppression in Multiple DWDM Channels For practical applications, it is desirable to have a single optical equalization filter which can suppress ISI in multiple DWDM channels. This will significantly reduce the device inventory when applied in the transponders with different wavelengths. And as the DWDM transponders evolves and becomes fully wavelength- tunable at the transmitter and receiver, simultaneous equalization of multiple channels becomes a necessary requirement. Because this device can operate regardless of the channel wavelength (colour), it is called a colourless optical equalizer. The previous analysis shows the transmission curve within the passband of each DWDM channel is critical for an optical equalization filter. Therefore, a periodic comb filter, such as shown in Figure 3.8 can be used for simultaneous suppression of ISI in multiple DWDM channels. Within a passband around an ITU grid, the transmission curve should be designed for the compensation of the filtering from optical multiplexer and demultiplexer to generate an overall Raised-cosine transmission curve. The optical comb filters shown in Figure 3.8 can be made with Mach-Zehnder interferometers [98], Fabry-Perot (FP) interferometers [99], fibre Bragg gratings [100], optical loop mirrors[101], etc.

70 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

0

-5

-10 Insertion Loss (dB) f f f ITU grids N−1 N N+1

Figure 3.8 A periodic comb filter which can be used as an optical equalization filter for simultaneous ISI suppression in multiple DWDM channels.

The periodic profile of the colourless OEQ not only reduces device inventory, it also reduces the number of devices in the system significantly. As shown on Figure 3.9, instead of having one OEQ at each channel (Figure 3.9(a)), a single unit can be used for the entire DWDM link (Figure 3.9(b)). This will reduce the hardware cost even further.

Data λ Transmitter OEQ Receiver 1 in 1 D Odd M channels E

… U M … X U Data λ X Receiver N-1 in N-1 Transmitter OEQ ROADM … Data λ 2 in 2 Transmitter Repeater OEQ Receiver Interleaver Interleaver D M Drop Add E U M

… Or cross … X Even U Data channels λ X N in N Transmitter OEQ Receiver

(a) Placing one non-colourless OEQ in each DWDM channel

Data λ Transmitter Receiver 1 in 1 D Odd M channels E

… U M … X U Data λ X Receiver N-1 in N-1 Transmitter ROADM … OEQ Data λ 2 in 2 Transmitter Repeater Receiver Interleaver Interleaver D M Drop Add E U M

… Or cross … X Even U Data channels λ X N in N Transmitter Receiver (b) Placing a single colourless OEQ for all DWDM channels Figure 3.9 Reducing OEQ quantity by using colourless OEQ for all channels.

71 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

The term “optical equalizer” has also been used to describe an optical device or equipment to balance the power level among multiple DWDM channels. To distinguish the meaning, this type of equalizer is called optical inter-channel equalizer or simply channel equalizer, while the optical equalizer investigated in this work is called optical intra-channel equalizer since the equalization is performed within each DWDM channel, even though the colourless device can equalizer multiple channels simultaneously. In the rest of this thesis, the term OEQ refers specifically to the intra- channel optical equalizer.

3.2.2 Device Design a. Basic Design: Fabry-Perot Interferometer Among various technologies to implement the colourless OEQ, Fabry-Perot (FP) interferometer method is selected for the device design due to its relative simple principle and the maturity of fabrication technique. This technology is especially suitable for colourless equalization since it produces periodic output profile. When this period (that is, the FSR) is adjusted to be equal to the DWDM channel spacing, the same equalization operation can be imposed on all channels. The FP interferometer (or called etalon) consists of a plane-parallel plate of thickness l and index n that is surrounded by a medium of index n0, as shown in Figure 3.10. At the boundaries of the resonator, the electrical field is transmitted as well as reflected. The transmitted wave Et and the reflected wave Er are [102] ′ = tt Et f − f Ei (3.7) − j 2π C 1− r′2e FSR and

f − f − j 2π C ′ ′ FSR = + tt r e Er (r f − f )Ei (3.8) − j 2π C 1− r′2e FSR where the FSR is c FSR = (3.9) 2nl cosθ r is the reflection coefficient, t is the transmission coefficient for waves incident from n0 toward n, and r′ and t′are the corresponding coefficients for waves travelling from n toward n0. fC is the central frequency.

72 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

Mirror 1 Mirror 2 n0 n n0

Transmitted wave Et Reflected wave Er

θ l Input Ein

Figure 3.10 Model of a Fabry-Perot interferometer.

For symmetric FP interferometers, the following relationships exist: r′ = −r , R = r 2 = r′2 , and T = tt′ . For lossless mirrors, one can get R +T =1 from conservation-of-energy relation. Therefore, Et and Er can be re-written as

= T Et f − f Ei (3.10) − j2π C 1− Re FSR

f − f − j2π C − FSR = (1 e ) R Er f − f Ei (3.11) − j2π C 1− Re FSR And the light intensity transmission coefficients for the transmitted and reflected light are I E E * T 2 t = t t = (3.12) * − I E E 2 2 f f i i i T + 4Rsin (π C ) FSR f − f 4R sin 2 (π C ) I E E* r = r r = FSR (3.13) * − I E E 2 2 f f i i i T + 4R sin (π C ) FSR Figure 3.11 shows the light intensity transmission curve and phase change for different settings of mirror transmission coefficient. The depth of the transmission coefficient dip at the centre of ITU-T grid wavelengths is 1.7 dB, 3.5 dB and 5.4 dB for mirror transmission coefficients of 0.9, 0.8 and 0.7, respectively. With smaller mirror reflectivity, there is smaller phase change for signals passing through the filter.

73 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

0.12 T=0.9, FSR=50 GHz T=0.9, FSR=50 GHz 0.0 0.08

0.04 -0.5

0.00 -1.0 Phase (rad) Phase -0.04

-1.5 -0.08 Intensity transmission Coefficient (dB) Coefficient transmission Intensity

-2.0 -0.12 -100 -50 0 50 100 -100 -50 0 50 100 Frequency (f-f ) (GHZ) Frequency (f-f ) (GHZ) ITU ITU (a) T=0.9

0.24 T=0.8, FSR=50 GHz T=0.8, FSR=50 GHz 0 0.18

0.12 -1 0.06

0.00 -2

Phase (rad) -0.06

-3 -0.12

Intensity transmission Coefficient (dB) Coefficient transmission Intensity -0.18

-4 -0.24 -100 -50 0 50 100 -100 -50 0 50 100 Frequency (f-f ) (GHZ) Frequency (f-f ) (GHZ) ITU ITU (b) T=0.8

T=0.7, FSR=50 GHz 0.32 T=0.7, FSR=50 GHz 0 0.24

-1 0.16

-2 0.08 0.00 -3 -0.08 Phase (rad) Phase -4 -0.16

-5 -0.24 Intensity transmission Coefficient (dB) Coefficient transmission Intensity -0.32 -6 -100 -50 0 50 100 -100 -50 0 50 100 Frequency (f-f ) (GHZ) Frequency (f-f ) (GHZ) ITU ITU (c) T=0.7 Figure 3.11 Light intensity transmission curve and phase change under different mirror = − transmission coefficients (FSR=50 GHz, fc fITU 25GHz ). b. Parameter Design and Verification Since the proposed colourless intra-channel OEQ is used to compensate for the spectral distortion caused by the passive multiplexing and demultiplexing devices (such

74 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

as AWG multiplexers and demultiplexers, interleavers and de-interleavers) in the optical transmission path, its design needs to be customized for each transmission link with different multiplexing and demultiplexing elements. The optical characteristics of the DWDM signal, in particular its spectral profile, will also influence the optimal design of the OEQ. These optical characteristics are mainly determined by the bit rate, modulation format, and multiplexing scheme. As a passive device, the optical profile of the OEQ needs to be pre-designed before the device fabrication since it cannot be modified after the device is made. In this section, a colourless intra-channel OEQ for 40G DPSK system is designed as an example. The actual signal bit rate is 42.8 Gbit/s, which includes 7% overhead for Reed Solomon FEC codes. The transmission link between the optical transmitters and the receivers include a 100 GHz AWG multiplexer with Gaussian profile (3 dB bandwidth about 70 GHz), a 50G/100G flat-top optical interleaver, a 50G/100G flat-top optical de-interleaver, and another Gaussian 100 GHz AWG as the demultiplexer. The profiles of these devices are based on real commercial devices. The intensity transmission and group delay curves of the AWG and optical interleaver used in the simulation are shown in Figures 3.12(a) and 3.12(b) respectively. Based on these device data, the filter profile of the OEQ is calculated using the method described earlier. It has a periodic profile as shown on Figure 3.11(c), with a dip depth of about 5.4 dB at the centre of each 50 GHz ITU-T grid. This dip depth corresponds to a transmission coefficient of about 0.7 in FP interferometer.

(a) AWG (b) Optical interleaver Figure 3.12 Optical characteristics of optical devices used in the simulation: (a) AWG multiplexer/demultiplexer and (b) optical interleaver/de-interleaver.

75 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

Optical equalization filter Figure 3.13 VPI Simulation model for DWDM transmission link with colourless intra- channel OEQ.

The performance of the designed intra-channel OEQ is verified using VPI TMM simulation software. The simulation setup is shown in Figure 3.13. The signals are 33% duty cycle RZ-DPSK with 42.8 Gbit/s data rate. Three 100 GHz-spaced channels at 188.40 THz, 188.50 THz and 188.60 THz frequencies are multiplexed to form the Even Channels, and another three 100 GHz-spaced channels at 188.35 THz, 188.45 THz and 188.55 THz are multiplexed to form the Odd Channels. The two groups are combined using the 50G/100G interleaver. The fibre link contains six spans of standard single mode fibre (SMF) with total length around 500 km. In the simulation, the transmission link is first removed to obtain the back-to-back results, and then added to obtain the transmission performance. At the receiver end, a 50G/100G de-interleaver separates the Odd Channels and the Even Channels. The designed 50 GHz-FSR colourless intra-channel OEQ filter is placed before the Odd Channels after the de- interleaver. The Odd Channels and Even Channels are then separated by respective AWG demultiplexers. At each output, the eye diagrams of all three Odd Channels are recorded and the received signal’s Q-factors are calculated. Figure 3.14 shows the simulated eye diagrams of the received 42.8 Gbit/s RZ- DPSK signals. Without optical equalization, the Q-factors for the back-to-back signals are 13.8 dB, 14.4 dB, and 13.8 dB respectively. With one colourless OEQ to simultaneously suppress the ISI in all the three channels, the Q-factors are increased to be 19.8 dB, 22.9 dB, and 23.1 dB respectively. The improvement ranges from 6 dB to almost 10 dB. These figures show that the signal Q-factors can be increased by more than 6 dB under the back-to-back condition.

76 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

(f= 188.35 THz, Q=13.8 dB) f=188.45 THz, Q=14.4 dB) (f=188.55 THz, Q=13.8 dB) (a) Back-to-back without optical equalization filter

(f= 188.35 THz, Q=19.8 dB) (f=188.45 THz, Q=22.9 dB) (f=188.55 THz, Q=23.1 dB) (b) Back-to-back with optical equalization filter

(f= 188.35 THz, Q=12.3 dB) (f=188.45 THz, Q=12.5 dB) (f=188.55 THz, Q=12.1 dB) (c) After 6 spans of transmission without optical equalization filter

(f= 188.35 THz, Q=15.4 dB) (f=188.45 THz, Q=15.4 dB) (f=188.55 THz, Q=15.2 dB) (d) After 6 spans of transmission with optical equalization filter

Figure 3.14 Simulated eye diagrams of the received 42.8 Gbit/s RZ DPSK signals.

77 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

After 500 km transmission, the signal Q-factors for the three DWDM channels are 12.3 dB, 12.5 dB and 12.1 dB respectively. With the OEQ, the signal Q factors are increased by about 3 dB to be 15.4 dB, 15.4 dB, and 15.2 dB respectively. One can also observe that the eye diagrams become asymmetrical when the OEQ is placed (both back-to-back case and the transmission case). This is due to the additional dispersion caused by the FP interferometer type OEQ. As expected from the theoretical analysis and calculation, the simulations with parameters from existing networks demonstrate that the signal ISI can be effectively suppressed with the proposed OEQ. The colourless intra-channel OEQ is a practical, easy-to-implement scheme of mitigating the ISI issues across multiple DWDM channels simultaneously. In order to get the best performance, the shape of the transfer curve of an optical equalization filter should be designed based on the filtering effects the signal experienced. Some drawbacks of such device are: it introduced additional loss to the signal due to its dip at the centre of the channel, and the extra CD caused by the FP interferometer. c. Design Optimisations To optimise the design of the intra-channel OEQ, several parameters and variations are investigated here. The robustness of the device is also studied. These studies are conducted through simulation using VPI TMM software. Similar to the studies above, 42.8 Gbit/s 33% RZ DPSK signal is used, and both the back-to-back performance and the performance after transmitting over 500 km SMF fibre are measured. The per-channel launching power is swept from -4 dBm to +6 dBm in 1 dBm increments, which covers the normal operation range.

I. Dip Depth Variation The first parameter to study is the depth of the OEQ passband dip. While maintaining the period passband shape as shown on Figure 3.8, the depth at the centre is set from 60% to 140% of the ideal depth in increment steps of 20%. In this design, the ideal depth has 5.23 dB insertion loss. The improvement of Q-factor comparing to the case without OEQ is plotted against the dip depth on Figure 3.15. The insertion loss is also plotted.

78 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

8 7 Dip insertion loss 6 5 4 3 Q-factor improvement

improvement (dB) improvement 2 1 Dip insertion loss and Q-factor Q-factor and loss insertion Dip 0 50% 60% 70% 80% 90% 100% 110% 120% 130% 140% 150% OEQ relative dip depth

Figure 3.15 Effect of OEQ dip depth.

The results show that the optimum OEQ dip is not exactly at the theoretical calculated value, but is about 20% higher. However within ±20% variation from the theoretical dip level, the difference in Q-factor is less than 0.5 dB, which shows that the Q-factor improvement is not very sensitive to the depth variation. Also, as expected, increasing the dip depth leads to larger insertion loss for the signal, therefore the trade- off between transmission performance improvement and OSNR penalty degradation needs to be considered in field application. The data on Figure 3.15 are for 500 km SMF transmission at 4 dBm per channel power. The results are similar for back-to- back simulation and transmission at other channel power levels.

II. Passband Profile Symmetry Next, the effect of OEQ passband symmetry is studied. Two extreme cases are considered. The first one is the ideal case where the dip-shaped passband has perfect symmetry, and the second one only has OEQ dip profile at one side. In other words, it only contains half of the OEQ on the frequency axis. The passband profiles for these two cases are shown on Figure 3.16. These two profiles have the same depth and dip slope. Only single OEQ period is studied because the transmission function outside the 50 GHz range does not matter for the channel under study.

79 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

1 50 GHz 1 0 0 -1 -1 -2 -2 -3 -3 -4 -4

Insertion loss (dB) -5 Insertion loss (dB) -5 -6 -6 188.4 188.45 188.5 188.55 188.6 188.4 188.45 188.5 188.55 188.6 Frequency (THz) Frequency (THz)

Figure 3.16 Passband profiles for the (a) Symmetric OEQ; and (2) Half OEQ.

The transmission performance for different OEQ profiles are shown on Figure 3.17 and compared with the transmission performance without OEQ. These results show that having the worst symmetry (that is, one side of the OEQ is completely removed) will induce up to 1.3 dB penalty. However even with only half of the OEQ, there is still almost 2 dB improvement compared to the system without OEQ.

18 17 16 15 14 With symmetric OEQ 13 With half OEQ 12 Without OEQ Q-factor (dB) Q-factor 11 10 9 8 -4-20246 Per channel power (dBm)

Figure 3.17 Effect of OEQ passband profile symmetry.

III. Effect of Frequency Offset Another practical parameter is the frequency offset of the OEQ. In actual transmission system, the WDM signal wavelength might drift from the standard ITU-T grid, and the passive components, such as interleaver and multiplexer might not be centred exactly on the ITU-T grid either. The effect caused by these frequency offset is studied here. 80 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

In the simulation, the centre frequency of the OEQ is shifted from -10 GHz to +10 GHz with respect to the actual ITU-T grid. The Q-factors at various frequency offset conditions are measured and shown on Figure 3.18. The results show that frequency offset will degrade the performance of OEQ. However if the frequency offset is within ±5 GHz, the penalty is less than 1 dB.

17

16

15

14 Q-factor (dB) 13

12 -10 -5 0 5 10 Frequency offset (GHz)

Figure 3.18 Effect of OEQ frequency offset.

IV. Location of Equalization Multi-channel OEQ can be placed at the transmitter side or the receiver side of a DWDM system. If such transmission system were linear and noiseless, it would not make any difference to put it at the transmitter or the receiver side. However, because noise is always present in real systems and fibre nonlinearity exists, the locations of optical equalization filter are not necessarily equivalent. The equalization filter can even be factored, with part in the transmitter and part in the receiver. How the OEQ location effects the equalization performance is studied here. In the simulations where OEQ is placed at either the transmitter end or the receiver end only, full OEQ dip depth is applied. In the simulation where the equalization is distributed at both the transmitter and receiver ends, each OEQ has half of the dip depth. Figure 3.19 shows the simulation results. These results show that the difference among the OEQ locations is small, especially when the per-channel optical power is low. When the optical power is higher, such as 6 dBm per channel, placing the OEQ at the transmitter end will lead to slightly better performance of about 0.4 dB. This is 81 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

likely caused by the fact that the optical signal power is slightly lower if the OEQ is placed at the transmitter end (it contributes 5.23 dB additional insertion loss at the peak), therefore reduces the nonlinear effect at high power cases. In the cases where the optical power is lower, fibre nonlinear effect is not significant, therefore the difference caused by the OEQ loss is small. Distributing the OEQ between transmitter and receiver produces intermediate results.

17

16

15

14 OEQ at Rx OEQ at Tx 13 OEQ at Tx and Rx 12 Without OEQ Q-factor (dB) Q-factor 11

10

9 -4 -2 0 2 4 6 Per channel power (dBm)

Figure 3.19 Effect of OEQ location in the transmission link.

V. Optimum Power The study results above, such as on Figures 3.17 and 3.19, show that the insertion of OEQ in the transmission link does not shift the optimum transmission power for the WDM signals. In the case of 42.8 Gbit/s 33% RZ-QPSK signal, the optimum power remains at 4 dBm per channel before and after adding OEQ into the transmission link, regardless of the OEQ passband profile symmetry, centre frequency offset, and physical location. This ensures that the system setting and engineering rules do not need to be modified when implementing the colourless OEQ in the transmission system. These results also show that the colourless optical intra-channel equalizer is robust against manufacturing imperfection, such as having non-perfect passband profile, slightly larger or smaller dip depth (within ±20% variation) or certain degree (±5 GHz) of frequency shift. Therefore it is practical for implementation in the field.

82 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

3.2.3 Colourless OEQ Prototype and Characterization Based on the FP interferometer technology, a colourless optical intra-channel equalizer prototype is designed and fabricated. The FSR for the OEQ is 50 GHz. The passband is designed to equalize a 100 GHz AWG multiplexer and a 50 GHz/100 GHz optical interleaver in the transmission link. The green dashed curve on Figure 3.20(a) is the calculated ideal passband for equalizing such passive components, however due to the characteristics of FP interferometer and the capability of the fabrication process, the actual prototype is designed with an approximation profile, as shown on the blue solid

(a) Comparison between theoretical and actual passband profiles

14.00 150.00

12.00 100.00 10.00

8.00 50.00 6.00

4.00 0.00

2.00 CD (ps/nm) -50.00

IL (dB) (ps) & GD 0.00 0.00 25.00 50.00 75.00 100.00 125.00 150.00 175.00 200.00 -2.00 -100.00 -4.00

-6.00 -150.00 Frequency (GHz) Insertion loss Group delay CD

(b) Insertion loss, group delay and chromatic dispersion profiles Figure 3.20 Designed profile of a colourless OEQ with 50 GHz FSR.

83 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

curve on the figure. Based on this design and equations in the previous section, the group delay and chromatic dispersion profiles can be calculated as shown on Figure 3.20(b). The mechanical drawing and the photo of the constructed FP interferometer- based colourless OEQ are shown on Figure 3.21. Since the OEQ is essentially a single- input and single-output dielectric thin film filter, it is very compact. The dimension of the main body is about 18 mm × 16 mm × 12 mm. The total length including the fibre holder and fibre boots is within 80 mm.

80mm

Figure 3.21 Mechanical drawing and photo of the constructed colourless OEQ prototype.

The optical characteristics of the constructed prototype are measured using Agilent 81910A photonic all-parameter analyser. The measured spectrum ranges from 1520 nm to 1570 nm, in steps of 5 pm each. The resolution bandwidth setting for the phase measurement is 50 pm. For the loss measurements (including insertion loss and PDL), each measurement is measured three times to take average. For phase measurements (including group delay and DGD), each measurement is repeated 30 times due to the higher sensitivity to noise. The averaged values are calculated as the final result. Figure 3.22 shows the insertion loss profile. The passband has the periodic shape with 50 GHz FSR. The centre of each period is on ITU-T 50 GHz grid and has a dip of about 3 dB. Figure 3.23 shows the measured polarization dependent loss, group delay, chromatic dispersion and differential group delay profiles of the constructed OEQ prototype. Only a few periods are selected to allow better observation of the properties. The characteristics at the remaining section of the spectrum are similar. The group delay and CD profiles agree with the design on Figure 3.19. These results show that the 84 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

colourless intra-channel OEQ has small PDL (<0.4 dB) and DGD (<0.1 ps), and the CD level is also within ±35 ps/nm. Therefore it is suitable for actual application in transmission links.

Figure 3.22 Measured insertion loss profile of the constructed colourless OEQ prototype; Inset: zoomed-in a selected spectrum.

(a) (b)

(c) (d)

Figure 3.23 Measured optical characteristics of the colourless OEQ prototype; (a) PDL; (b) group delay; (c) CD; (d) DGD. (The grey curves are the passband profiles for reference).

85 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

3.3 Experimental Demonstration 3.3.1 Passband Widening Measurement Before measuring the ISI suppression effect on the actual WDM signals, the passband widening performance of the fabricated OEQ prototype is measured experimentally.

Tunable Optical Broadband 50G/100G bandpass OEQ spectrum ASE source interleaver filter analyzer

Figure 3.24 Experiment setup for the passband widening experiment.

The experiment setup is shown on Figure 3.24. An Agilent 83438A Erbium broadband amplified spontaneous emission (ASE) source is used as the signal source. There are three passive components used in experiment, representing the typical filtering elements in the WDM transmission link. The first is an Avanex PowerMux C- band 50G/100G optical interleaver. It has a 3 dB passband width of 0.33 nm and peak insertion loss figure of 1.8 dB. The second component is a Santec OTF-300 manual tunable filter, which is used to emulate the optical multiplexer/demultiplexer in the transmission link. Its peak insertion loss is 4.0 dB, and the 3 dB passband width is 0.30 nm. The third component is the fabricated colourless OEQ prototype. These three components are connected in series. The spectra of different filtering configurations are measured using an Ando AQ6317B OSA. The resolution bandwidth settings for all measurements are 0.1 nm. The power spectrum of the broadband ASE source is subtracted from the measured optical spectra using the internal mathematical function of the OSA to obtain the absolute insertion loss values at various configurations. These absolute insertion loss curves are shown on Figure 3.25(a). Figure 3.25(b) shows the normalized passband spectra after compensating for the peak insertion loss for easy comparison of passband widths. The spectral widths of the accumulated passband under these filtering conditions are measured from the obtained spectra. With only the 50G/100G optical interleaver (the orange curve), the measured 3 dB passband of the spectrum is about 0.32 nm, which is narrower than the ideal 50 GHz channel width of about 0.4 nm. When the bandpass filter is added, the passband is further narrowed to 0.29 nm (the green curve).

86 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

But when the OEQ prototype with periodic dip on ITU-T grids (the dark blue curve) is inserted in the transmission link, the resultant passband is widened (the light blue curve) to 0.35 nm as the passband peak is suppressed by the dip of the OEQ filter. In terms of frequency, the OEQ widens the 3 dB passband from 36 GHz to 43.3 GHz. This represents a 20% passband widening.

0

-5

-10 Interleaver -15 Interleaver + filter -20 OEQ Interleaver + filter + OEQ -25 Insertion loss (dB)

-30

-35 1549 1549.5 1550 1550.5 1551 Wavelength (nm) (a)

0

-1

-2 Interleaver

-3 Interleaver + filter

Interleaver + filter + OEQ -4

-5 Normalized insertion loss (dB) -6 1549.8 1550 1550.2 1550.4 Wavelength (nm) (b) Figure 3.25 Optical spectra of the filtering elements with and without OEQ. (a) Absolute power levels; (b) Normalized power levels.

The same filter passband widening can be measured for 0.5 dB passbands, which is defined as the passband width of the signal at 0.5 dB lower than the passband peak. The measured results show that OEQ increases the 0.5 dB passband from 0.16 nm to 0.25 nm (or 19.5 GHz to 30.8 GHz). This is a 58% increase in the passband

87 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

width. These results clearly confirm the expectation that the OEQ can practically expand the passband width of a transmission channel. There is a trade-off that the insertion of the OEQ device leads to larger insertion loss at the optical signal (about 3 dB at the ITU-T grid centre in this case). This can be compensated by increasing the optical amplification level in the transmission link.

3.3.2 Back-to-Back Experiment A simple experiment is then designed to verify the performance of the colourless intra-channel optical equalizer. Only single channel is used, and the signal does not transmit over transmission fibre. The experimental setup is shown on Figure 3.26.

RZ-DPSK modulator Tunable DFB Laser 50G/100G MZM1 MZM2 bandpass diode interleaver filter

42.8G 21.4G data clock

DPSK Balanced Sampling VOA EDFA OEQ demodulator receiver scope

42.8G Data 21.4G Clock

Data Data

4x1 Mux 10.7G 10.7G Frequency doubler Tunable RF phase shifter Tunable RF RF splitter phase shifter

10.7G Data Data 10.7G Inset: 42.8 Gbit/s Clock data and 21.4 Gbit/s PPG clock generation RF splitter

Figure 3.26 42.8 Gbit/s RZ-DPSK back-to-back OEQ experiment setup.

88 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

At the transmitter end, an Alcatel A1905LMI CW DFB laser with a linewidth of about 2 MHz is used as the signal source. This laser is controlled by a Newport 8016 modular laser controller to set the output power level and wavelength through varying the driving current and temperature. At driving current of 82.9 mA and temperature of 29.3°C, the laser output is set to 10 dBm and the wavelength is set to 1550.12 nm (193.4 THz). The CW laser signal is then modulated by a 42.8 Gbit/s RZ-DPSK signal. A Fujitsu FTM7937EZ Ti:LiNbO3 dual drive Mach-Zehnder modulator (MZM) with modulation speed of up to 43 Gb/s in a push-pull configuration biased at the null point with ±V4 of 5 V drive swing for both arms is used to generate the DPSK-modulated signal. The required 42.8 Gbit/s electrical NRZ signal is obtained by using an SHF 404 electrical multiplexer to multiplex electrical four 10.7 Gbit/s PRBS signals, which are generated from the DATA and DATA outputs of an Agilent 70843C pulse pattern generator (PPG) with appropriate splitting and electrical delay control (see Inset of Figure 3.26). The PRBS pattern length of the combined signal is 231-1. Another MZM in a push-pull configuration driven at twice the drive voltage and half the symbol-rate clock is then connected to act as a RZ pulse carver and generate the 33% duty cycle RZ pulses. The Avanex PowerMux C-band 50G/100G optical interleaver and the Santec OTF-300 manual tunable filter measured previously are inserted after the transmitter to introduce strong channel filtering effect in the transmission link. An Eigenlight Series 400 VOA and an Amonics AEDFA-DWDM-23-B EDFA are used to vary the OSNR of the received signal for BER measurement. The constructed OEQ prototype with about 3 dB dips at ITU-T grids is placed before the DPSK receiver. At the receiver side, an Optoplex Michelson interferometer-based DPSK demodulator with one bit optical delay (~23.3 ps) is used, and its constructive and destructive outputs are detected by a pair of u2t BPRV2125 balanced receivers with a 3 dB cut-off frequency of 31 GHz. The output electrical signal is then sent to an Agilent Infinium 86100C sampling oscilloscope with an 86109B 50 GHz bandwidth optical/electrical test head to measure the signal eye diagram. Figure 3.27 shows the signal degradation due to the filter narrowing by the optical interleaver and the bandpass filter. These eye diagrams are captured by connecting the optical signal at various points in the transmission link to the 50 GHz optical test head of module 86109B in the Infinium sampling scope. The vertical scale is 2.5 mW per division, and the horizontal scale is 5 ps per division. The accumulating 89 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

time to capture each eye diagram is about 30 seconds. Figure 3.27(a) is the good 42.8 Gbit/s 33% RZ-DPSK signal at the output of the modulator without filtering. When the signal passes through the 50G/100G interleaver, the signal is distorted due to the filtering effect on the channel, causing inter-symbol interference distortion (Figure 3.27(b)). Also, the eye diagram is no longer symmetric, this is caused by the dispersion from the interleaver. The interference continues as the signal passes through the subsequent 0.3 nm bandpass filter (Figure 3.27(c)).

(a) (b) (c)

Figure 3.27 Signal degradation due to filtering effect. (a) At the output of RZ-DPSK modulator; (b) after 50G/100G interleaver; (c) after the 50G/100G interleaver and the 0.3 nm bandpass filter.

The received signals at the balanced optical receiver output are shown on Figure 3.28. These eye diagrams are captured by connecting the output of the u2t photoreceiver directly to the 50 GHz electrical test head of module 86109B in the Infinium sampling scope. The vertical scale is 50 mV per division. Figure 3.28(a) is the received signal when the filtering elements and OEQ are not present in the system. It displays an ideal symmetric open eye diagram, with the vertical eye opening value of about 200 mV. With the insertion of the 50G/100G interleaver and 0.3 nm bandpass

(a) (b) (c)

Figure 3.28 Eye opening improvement by OEQ. (a) Without filtering elements in the transmission link; (b) after adding the 50G/100G interleaver and the 0.3 nm bandpass filter; (c) after adding the OEQ to the transmission link with the 50G/100G interleaver and the 0.3 nm bandpass filter. 90 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

filter, the eye is significantly closed due to filtering effect (Figure 3.28(b)). The vertical eye opening is reduced to about 70 mV. When the OEQ is added, the eye opening is improved to about 98 mV vertical eye opening (Figure 3.28(c)). This is a 40% improvement. These results from the back-to-back experiment demonstrate the ISI mitigation benefit of the designed OEQ device.

3.3.3 Transmission Experiment In the next step, the performance of colourless OEQ is tested on a commercial DWDM transmission system. The system is the NEC SPWV 160 DWDM system, which supports 50 GHz channel spacing DWDM transmissions. The modulation format used in its 40G transponders is also RZ-DPSK. The output power level of the transponder is about -6 dBm. The DWDM signals are transmitted through 5 spans of standard SMF-28 single mode fibre, the span length is 80 km. The span loss is about 23 dB, which comes from the transmission fibre attenuation, connection loss, and the insertion loss of the passive components at every span. Each span contains a dispersion compensation module which is a spool of dispersion compensation fibre (DCF) that can compensate for most of the chromatic dispersion in the transmission fibre. Including the DCFs, the total fibre length is > 400 km. There is a transmitting EDFA module at the output of the transmitter node, an in-line amplifier at every span, and a receiving EDFA module at the input of the receiver. All these amplifiers have internal gain equalization function across the transmission band. The transmission link does not contain any 50G/100G interleaver, therefore the filtering elements are the optical multiplexer at the transmitter end and the optical demultiplexer at the receiver end. Both of them are 50 GHz AWG with Gaussian passband profiles. The colourless intra- channel OEQ prototype is placed before the input of the transponder at the receiving end. At the receiver of the 40G transponder, there is an electronic dispersion compensator (EDC) circuit after the dual PIN balanced receiver to compensate for any residual chromatic dispersion in the transmission link. Its reference voltage level can be adjusted. The received signal eye diagrams are recorded using Agilent Infinium 86100C sampling oscilloscope, and the BER values are measured by a real-time bit error counting software written for SPWV 160 testing purpose. Since this is a commercial system, most of the parameters cannot be tuned, except for the reference voltage of the

91 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

EDC circuit. Therefore the BER performances under different EDC reference voltage values are measured and compared.

-2 -2.5 -3 -3.5 -4 w/o OEQ -4.5 w/ OEQ -5 Log (BER) -5.5 -6 -6.5 -7 -500 -400 -300 -200 -100 0 Vref of EDC (mV)

(a) (b)

Figure 3.29 Measured BER without and with the colourless OEQ under different EDC reference voltage settings after 5 span transmission over SPWV160 system. Inset (a) Eye diagram for the received signal at the optimised EDC reference voltage without the OEQ; Inset (b) Eye diagram for the received signal at the optimised EDC reference voltage without the OEQ.

Figure 3.29 shows the measured BER performance under different EDC reference voltage settings for the system without and with the OEQ prototype. The logarithm of BER is plotted as the Y-axis. It can be observed that the optimum EDC reference voltage is about 275 mV for both cases. The system with the insertion loss OEQ always produces better BER performance than the system without the OEQ. Across the measured range, adding OEQ in the system leads to at least 0.5 dB improvement in the log(BER). The maximum improvement occurs around the optimised EDF reference voltage setting. Under this setting, the measured BER for the transmission link without the colourless OEQ is about 6.3×10-5; when the OEQ is added at the receiver side, BER is reduced to about 3.1×10-7. In other words, about 1.3 order

92 )%, 2)%, 2 ##.,%++ )! ),##.,%++ )! ),##.,%++ )! ),% 4.5%,% 4.5%,% 4.5%,

improvement in the BER performance is achieved by adding the OEQ device. The eye diagrams at 275 mV EDF reference voltage for both cases are shown on the insets of Figure 3.28. Without the OEQ, the vertical eye opening of the received signal is about 55 mV (Inset (a)). It is improved to about 75 mV after adding the colourless OEQ (Inset(b)). This is an improvement of 36%.

3.4 Conclusions In this chapter, I proposed a low cost yet effective optical equalization device to mitigate ISI caused by filter narrowing in the DWDM transmission link. It uses a colourless intra-channel OEQ to restore the overall passband to a raised-cosine profile. With a periodic equalizing filter profile at an FSR equal to the DWDM channel spacing, one single device can simultaneously equalize all DWDM channels. A colourless OEQ prototype has been fabricated using thin film dielectric F-P interferometer technology. Simulation showed that this device can improve the received signal performance by 3 to 10 dB, and demonstrated that it is robust against manufacturing imperfection. Experiments of the OEQ prototype in the commercial DWDM system verified that this device can widen the passband of the over all channel by 20%, and expand the eye opening by 36% to 40%. This brought 1.3 order of magnitude of improvement in the BER performance. This device can also be integrated with other passive optical devices in the transmission link, such as DPSK demodulator and optical interleaver, to further reduce size, optical insertion loss, and cost, and improve tolerance to temperature change. Due to commercial sensitivity of this technology, this work was not published, but three patent applications are pending [103-105].

93 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

)%, )%, 999      

         

        

   

In this chapter, I propose and demonstrate a novel flexible band tunable filter. In contrast to the single degree (centre wavelength) tuning function of the conventional tunable filters, this filter offers two-degree tuning operation in both the centre wavelength and the passband width. A prototype of the FBTF is designed and fabricated using cascading tunable edge filters based on dielectric thin-film structure and mechanical tuning method. This prototype demonstrates various flexible passband tuning functions. Using this tunable filter as a key building block, a ROADM node architecture is proposed. It has the benefit of giving priority to the express/through channels. It does not require large number of switching/tuning elements as in the conventional ROADMs, therefore is low cost. Also, the modular architecture makes the node expandable and allows for easy customisation and “paid-as-you-grow” investment strategy. Experiments in a two-ring four-node testbed demonstrated that this proposed ROADM node can deliver all the required add/drop/cross-connect functions with little signal degradation. Therefore this FBTF-based ROADM node is suitable for application in metro WDM networks.

4.1 Background In the recent years, ROADMs have becoming an important element in DWDM optical networks. Comparing with conventional fixed wavelength optical add/drop multiplexers (OADMs) that require manual configuration, ROADMs significantly

94 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

reduce network operators’ operating expense such as truck rolls and man hours. ROADM-based systems also enhance the network flexibility and can better handle the ever growing Internet-based traffic [38, 39]. Currently the common architectures for ROADM include demultiplexer-switch- multiplexer architecture, broadcast-and-select architecture and wavelength selective switch-based architecture [39, 40, 106] as described below; • Demultiplexer-switch-multiplexer (DSM) architecture: All DWDM channels are fully separated and each channel has a 2×2 optical switch (or two 1×2 switches) that selects whether this channel is dropped/added or passed through to the output via the multiplexer (Figure 4.1(a)). This is a straightforward design, however for a DWDM system with n channels, it requires n fixed wavelength transponders to be pre-installed, and therefore it is a costly and inflexible approach. Another DSM approach is to install a multi-port fibre switch (Figure 4.1(b)). In this case, the dropped channels can be associated with any transponders, reducing the number and the total cost of transponders (if less than 100% of add/drop is required). However the multi-port switch is also costly, and its technology is not as mature as smaller degree switches. Also, having a single switch with large port-count represents a single point of failure in the node design. Another design variation is to use banded DSM architecture [107]. It divides the spectrum into bands and uses a smaller switch element on each band (Figure 4.1(c)). This approach reduces the risk of single point of failure and avoids the high start-up cost, because those wavebands that do not require add/drop do not need to be fully demultiplexed and do not require optical switches. Its disadvantages include higher optical loss and packaging and interconnection costs.

Multi-port switches 2x2 switches

Input…… …… Output Input Output Input Output

Demultip lexer Multip lexer Demultiplexer Multiplexer TP ND TPND …… TP ND TP ND TP ND TP ND 1 2 m Cascaded Cascaded λ λ … λ 1 2 n demultiplexers Multi-port multiplexers Transponders switches Transponders (a) (b) (c) Figure 4.1 DSM-based ROADM architectures: (a) using 2×2 switches; (b) using multi- port fibre switch; (c) banded DSM.

95 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

• Broadcast and select (B&S) architecture: As shown on Figure 4.2, in the B&S architecture, the entire input spectrum is split into two paths, one sent to the output (through path) and the other to the drop port (drop path). A wavelength blocker (WB) or dynamic channel equalizer is used in the through path to block those channels that are being dropped, and at the same time provides per- wavelength power balancing. On the Drop path, the dropped channels are separated and sent to individual transponders at the Drop ports. In order to provide full flexibility, tunable filters are required. For a node with dropping capability up to n channels, the first implementation is to use n units of 3-port tunable filters in cascade to select the n drop channels (Figure 4.2(a)). This approach has the disadvantage of unbalanced optical power among these dropped channels. Also low cost yet efficient 3-port tunable filter is not widely available yet. The second implementation uses a 1:n splitter with a 2-port tunable filter at each of the n splitter outputs (Figure 4.2(b)). This method suffers from higher insertion loss. Another alternative is to demultiplex all channels at the Drop port and use a large-scale fibre switch to assign the required channels to the transponders. This is an expensive solution. For the add channels, tunable or fixed lasers are used along with couplers. B&S architecture has been viewed to be a good candidate for metro as well as long haul applications and has received significant amount of research efforts [107- 110].

 #"   , #  #"   , #

    $#( ; $#( ; +# ,> , #  #"  +# ,> , # =,  '      +# G  +# '   G G  +# -"#  # +# #

G 2  -"#  2 

  (a) (b) Figure 4.2 B&S-based ROADM architectures: (a) using cascaded filters; (b) using splitter and filters.

• WSS-based architecture: It uses a WSS with high port count (typically 1×9). The WSS is used for dropping channels, and optical coupler or another WSS is used for adding channels (Figure 4.3) [40-42]. This architecture provides

96 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

excellent flexibility since any channel can be dropped at any Drop port. A 1×9 WSS at drop section can drop up to 8 individual channels. If more channels need to be dropped, the excess drop channels can be combined at a Drop port and additional WSS or demultiplexer connects to this port in cascade. An alternative is to use the WSS in both directions and use optical 3-port circulators to select the incoming and outgoing signals [111]. Such architecture has slightly higher optical loss, however it delivers the wavelength selection function at both the input and the output ends with only single WSS, thus reduces the cost and hardware size while eliminating the wavelength contention problem and improving the signal quality such as reduce the insertion loss and increase the extinction ratio.

 , #

     G   G ; (;  

 G 2   2  (a) (b) Figure 4.3 WSS-based ROADM architecture: (a) using unidirectional WSS; (b) using bidirectional WSS.

These architectures described above allow each WDM channel to be operated individually such as passing through to the output, dropping at local node, or being regenerated. For higher degree ROADM nodes, cross connect function can also be performed on individual channels. Therefore they provide large degree of flexibility for WDM network applications. However for certain portion of intermediate nodes, this degree of flexibility is not required. In these nodes, most of the traffic are through traffic and does not require local add, drop or regeneration. Furthermore, the remaining channels that require local operations often have adjacent centre frequencies and therefore can be grouped together and have group operation. Therefore these nodes do not require full add/drop operation on all individual channels and can be replaced with lower cost ROADM nodes [112]. In this chapter I introduce a low cost ROADM node architecture that does not require full demultiplexing and multiplexing of the WDM channels, and it operates at both individual wavelength and grouped waveband level. It also provides flexibility to 97 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

reconfigure the ROADM node remotely and is capable for in-service network upgrade for the future. Besides reducing the equipment cost, this ROADM architecture also delivers better optical performance. A key to this architecture is a novel optical filter that offers tunability at its centre wavelength as well as the passband width. It is called the flexible band tunable filter (FBTF). This chapter will start by introducing this device, followed by description of the low cost ROADM node architecture and its modules. The performance of the device prototype and network experimental results will be presented and the benefit of this ROADM architecture will be discussed.

4.2 Development and Implementation of Flexible Band Tunable Filter 4.2.1 Flexible Band Tunable Filter and Its Functions An optical filter is a passive optical device to select a certain portion of optical spectrum from a wider spectrum at the input. Of all devices, optical filter is one of the most important passive devices used in the WDM optical networks. They are needed wherever signals with different wavelengths propagating in a fibre must be multiplexed or demultiplexed in a WDM network [113]. Most of the optical filters in the current DWDM network have fixed passband. However as the network traffic becomes more dynamic and reconfigurable demands rises, tunable filters are started to be deployed in the network in application such as the B&S ROADM nodes described earlier. Conventional tunable filters allow the user to tune the centre wavelength or frequency of the passband in either a discrete or continuous fashion. In other words, they offer one degree of freedom in the tunability aspect. Here I propose a novel optical filter that offers an additional degree of tunability, namely on the passband width. It is thus also called the flexible band tunable filter [114]. Besides the centre wavelength/frequency and the passband width, the two degrees of tunability can also refer to the tunability at both rising and falling edges of the selected passband. With this device, any single channel or multiple of adjacent channels (waveband) can be filtered out, or it can also allow the whole spectrum to be reflected with no optical signal at the transmit port. Some possibilities are illustrated on Figure 4.4.

98 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

1 2 3 4 5 6 7 8 Select Ch 4 λ

1 2 3 4 5 6 7 8 Select Ch 2-6 λ

1 2 3 4 5 6 7 8 Select Ch 6,7 λ

1 2 3 4 5 6 7 8 Select None λ

Filter passband

Figure 4.4 Examples of the possible passband set by the FBTF.

4.2.2 Design of FBTF The basic design of this novel BFTF is to use two stages of wide band tunable filters to form the rising and falling edges of the filter. Figure 4.5 illustrates this idea. It is comprised of a pair of 3-port tunable edge filters (TEFs) in series (Figure 4.5(a)), which behave as high-pass and low-pass filters in the optical domain, setting the upper and lower edges of the FBTF passband respectively. The overlapping section of these two TEFs’ passbands becomes the passband of the FBTF and exits to the Drop port (Figure 4.5(b)). The passing-through bands from these two TEFs are combined and sent to the Through (or Express) port. Therefore the FBTF is a 3-port device, and each of the incoming channels goes either go to the Drop port or the Through port, and none of them is blocked by the device. In actual device fabrication, the low pass optical tunable edge filter is not actually a low pass filter that allows the entire lower wavelength spectrum to be pass and reflects the entire spectrum with higher wavelength. Instead it is essentially a bandpass filter too, however it has a very wide passband and the rising edge of the passband has a very low wavelength (Figure 4.5(c)), therefore within the operating spectral section (usually within the range of 1520 to 1620 nm) it behaves like a low pass filter as shown on Figure 4.5(b). Similarly, the high wavelength tunable edge filter is essentially a bandpass filter with very wide passband, whose rising edge falls within the operating spectrum while its falling edge falls outside. 99 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

In B Low-pass TEF (a) A Through C Combiner High-pass TEF

Drop

“Low pass filter” “High pass filter”

(b) 1 2 3 4 5 6 7 8 λ Flexible band tunable filter Operating spectrum

Low λ TEF High λ TEF

(c) 1 2 3 4 5 6 7 8 λ Flexible band tunable filter Operating spectrum Figure 4.5 The operation principles of flexible band tunable filter: (a) design schematic; (b) spectra of passbands using tunable edge filters; (c) actual filtering elements.

Figure 4.6 shows the spectrum at each point in the schematic. An 8-channel example is used here. If Channels 6 and 7 need to be dropped, the low wavelength TEF will be tuned to the position that its falling edge is between Channel 7 and Channel 8, so that its passband includes Channels 1 to 7 (point A in Figure 4.6(a)), and Channel 8 is reflected/rejected (point B). At the same time, the high wavelength TEF is tuned to the position that its rising edge falls between Channel 5 and Channel 6, so that all channels higher than Channel 5 within the operation spectrum will be dropped. In this case, these channels include Channels 6 and 7, which are the final drop channels of the flexible band tunable filter. The rest of the channels are reflected/rejected, which are Channels 1 to 5 in this case (point C). Both the reflected bands at point B and point C are then combined using a 1×2 coupler and sent to the final reflection port, which contains all channels except the dropped Channels 6 and 7.

100 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

1 2 3 4 5 6 7 8 In λ

1 2 3 4 5 6 7 A λ

8 B λ

Transmit/ 6 7 Drop λ

1 2 3 4 5 C λ

1 2 3 4 5 8 Reflection λ Operating spectrum Figure 4.6 Example of WDM channel spectra at various points within a FBTF.

It should be noted that in Figure 4.5(a) and the following schematics, the drop/transmit path is shown on the same side as the input path (left hand side on Figure 4.5(a)), while the reflection path is shown at the opposite side (right hand side on Figure 4.5(a)). This is opposite to the conventional notation for filters where the drop path is shown at the opposite side. It is deliberately drawn in this way so that it is consistent with the conventional notation for a ROADM, since ROADM is a main application for the flexible band tunable filter, and will be discussed in more details in sections below. Besides using 1×2 coupler at the reflection port, another low wavelength TEF can also be used to combine the two reflection bands from two TEFs at the input (Figure 4.7(a)). This approach is more costly due to the additional TEF required. However it produces better optical performance. Firstly, the manufacturer specified insertion loss figures for the drop and reflection paths for a TEF are between 1.8 and 2.3 dB. Comparing to the 3.3 dB insertion loss through a 1×2 coupler, the TEF at output approach can lead to 1 to 1.5 dB loss reduction. More importantly, the isolation at the output reflection port is improved significantly with the additional TEF at the output reflection port. This is because all the reflected channels go through two filters in

101 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

cascade before reaching the output, which means that they are filtered twice. Having a good isolation performance will ensure that the residual signals of unwanted channels do not affect the output signals. It is therefore very important to obtain good isolation figure for the flexible band tunable filter, especially due to the fact that the current isolation level achieved by single stage TEF is only 20 to 30 dB. During the operation, the two low wavelength TEFs at the input port and reflection port are tuned simultaneously with same tuning positions. The wavelength spectrum at each point is shown at Figure 4.7(b).

B In Reflect/Express Low λ TEF Low λ TEF A (a) C

High λ TEF

Transmit/Drop

1 2 3 4 5 6 7 8 In λ

1 2 3 4 5 6 7 A λ

8 (b) B λ

Transmit/ 6 7 Drop λ

1 2 3 4 5 C λ

1 2 3 4 5 8 Reflection λ Operating spectrum

Figure 4.7 FBTF using additional low λ TEF: (a) Device schematic; (b) WDM channel spectra example.

Optical attenuators can be placed at the output of two TEFs (points B and C in Figures 4.5(a) and 4.6(a)) to balance the optical power level of these two bands. Tilt 102 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

equalizers can also be used to provide better channel equalization since the FBTF solution does not fully demultiplex the input DWDM signals into individual channels, unlike in wavelength blocker or wavelength selective switch. In the FBTF, the two TEFs operate independently, and the tuning of the FBTF is hitless. This means that only the channels that are required to switch between the Drop port and the Through port will be affected during the tuning operation, the transmission of all the other channels at both sides will not be affected.

4.2.3 FFBTF Prototype a. Dielectric Thin Film F-P Interferometric Filter The key elements of the FBTF, namely the high pass and low pass TEFs, can be realized by using F-P interferometers, as described in Section 3.2.2. Thin film filter (TFF) is based on a micro-optic technology that uses different types of dielectric material to construct the mirror and cavity of the FP interferometer. Figure 4.8 is the schematic of a typical thin film FP interferometric filter. A stack of materials of different refractive indices is used as the mirror. The stack is made up of very thin layers of material with alternating refractive indices. The typical high refractive index material used is SiO2 (refractive index about 1.46). The typical low refractive index materials are TiO2 and Ti2O5 (refractive indices are 2.3 and 2.1 respectively). The thickness of each layer is 1/4 of the specified wavelength to be reflected. These layers alternate about 10 times and achieve reflectivity of about 90%. The resonator cavity of the FP interferometer is also made up of dielectric material, instead of an air gap. Its length is about twice the specified wavelength.

Figure 4.8 Schematic of a thin film FP interferometric filter [115, p. 233].

103 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

Through precision design and multiple layer deposition, ultra wide passband with steep passband edges can be achieved. By setting one of the passband edges within the operation spectrum while keeping the other edge outside, the resulted filter passband profile behaves as a high-pass or low-pass optical filter. b. Tuning Mechanism There are various methods to tune the passband centre wavelength of an optical filter, such as acousto-optic tuning, electro-optic tuning, and so on. For a single stage thin film FP interferometric filters, according to equation (3.9), the FSR is determined by three factors, namely the refractive index of the material in the resonant cavity n, the length of the cavity l and the incident angle of the input light θ. Among these three factors, the refractive index and the length of the cavity are more difficult to change, so the most convenient method is to change the incident light angle. This can be achieved by rotating the collimator of the input fibre or rotating the thin film stack (Figure 4.9).

Through channels l

n All input θ λ channels drop Dropped Angle Tuning channels

Figure 4.9 Mechanical angle tuning for optical FP interferometric filter.

In the wavelength domain, the FSR of the FP interferometric filter can be expressed as: λ2 FSR = (4.1) 2× n×l cosθ It can be noticed that in the wavelength domain the FSR is no longer a constant across the optical spectrum, but depends on the resonant wavelength λ, which can be calculated by: 2× n×l cosθ λ = (4.2) m

104 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

where m is an arbitrary integer 1, 2, 3… Equation (4.2) shows that if refractive index n and cavity length l are fixed, the centre wavelength of the filter is proportional to the cosine of the incident angle (m is set at a fixed value due to the cascading of filters in series). It means: λ = λ θ 0 cos (4.3)

Here (0 is the centre wavelength of the filter when the incident light is normal to the thin film surface. It is a constant with the value of: 2× n×l λ = (4.4) 0 m Comparing to other filtering technologies, the FP interferometric method (including the thin film filter) can achieve widest tuning range of up to 500 nm (Table 4.1) and therefore is suitable for applications in fibre optic networks. Also, since the F- P filter technology is mature and there are large numbers of device manufacturers working on thin film devices, constructing the FBTF using thin film FP filter technology has very low risk of lacking suppliers and supports.

Table 4.1 Comparison of different technologies for tunable filter [116].

c. Prototype Construction A FBTF prototype is constructed using two thin film-based TEFs with mechanical angle tuning. A coupler is used to combine the two passing-through bands at the Express port. This filter is designed for standard ITU-T 200 GHz spacing DWDM system with tuning range across the full C-band. The reason why 200 GHz is 105 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

selected instead of the more common 100 GHz channel spacing is that it is more difficult to produce TEF with wide passband and sharp filter edge using the thin film technology. 200 GHz system allows wider gap between two adjacent channels. To achieve mechanical tuning, a stepper motor with precision step sizes is attached to the bulk optic material with the dielectric thin film coating (Figure 4.10). During the tuning process, the bulk optic material is tuned physically to change the angle with respect to the incident light, instead of physically rotating the angle of the incident beam.

Stepper motor Dielectric thin film coating

Input and Electrical output fibres control interface Figure 4.10 Internal structure of a TEF in the FBTF prototype.

The mechanical tuning by stepper motor provides a benefit to the FBTF, namely the latching operation. With the stepper motor, the physical position of the filter is maintained even after the power is turned off, therefore the wavelength selection can be kept without applying constant electrical power.

Figure 4.11 GUI of the FBTF control software in LabVIEW.

106 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

The control software is written using G language on the LabVIEW platform developed by National Instrument. The graphical user interface (GUI) is shown on Figure 4.11. Its GUI allows the user to tune each TEF in real time through dragging the spectrum bars, and it shows the spectral positions of both TEFs graphically. RS-232 serial interface is used to communicate between the PC and the stepper motor controller circuits inside the FBTF. The software also links and controls an Agilent 81640A sweep laser source and an Agilent 81634A photodetector through GPIB connection to allow instant measurement of the spectrum at each point of interest. The measured spectra are shown on the waveform graph.

4.2.4 FBTF Prototype Characterization Firstly, a passband from about 1532 nm to about 1544 nm is selected by tuning two TEFs to positions 4019 and 1062 respectively. These figures indicate the precise tuning step number so the tuning positions are repeatable. This band covers about eight DWDM channels at 200 GHz channel spacing. Figure 4.12 shows the relative insertion loss spectra of all points of interest measured by Agilent 81910A photonic all-parameter analyser. Each individual insertion loss profile is shown at Figure 4.13. The input light is split into two sections through the short wavelength TEF: the spectrum with wavelength longer than the selected value (about 1545.9 nm) is sent to the 1×2 coupler (Point B in Figure 4.5(a)), while the shorter wavelength section is sent to the input of

A B C Through Drop Through

Figure 4.12 Example of measured insertion loss profile. 107 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

Point A Point B

Point C Drop

Through

Figure 4.13 Measured insertion loss profile at various points of FBTF. the second TEF (Point A). The second TEF (long wavelength TEF) then further splits this spectrum into two sections: the section with wavelength shorter than the selected value (about 1529.2 nm) is sent to the second input of the 1×2 coupler (Point C), while the rest is the drop band. The 1×2 coupler combines the spectrum sections at Point B and C to form the express/through output. Based on these measured insertion loss profiles, the maximum insertion loss at Point A is and Point B, point C and the Drop port are 1.55 dB, 2.54 dB, 4.56 dB, and 3.28 dB respectively. For the Through port, the insertion loss figure is different at both sides (8.4 dB and 6.0 dB respectively). As discussed earlier, this difference can be balanced with inserting appropriate optical attenuators. Up to 0.75 dB loss ripple can be

108 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

observed near the filter edges. This is an intrinsic problem that comes with filter coating layer design, especially for filter with very wide band and steep passband edge such as the TEFs. But this is within the acceptable insertion loss uniformity range for WDM systems. The passband edges of the resultant filter are steep enough to avoid any skipping channel.

Drop

Through Through C B A

(a)

Drop Through Through

(b)

Through Through Drop

(c) Figure 4.14 Measured PDL, DGD and CD at various ports.

Other optical parameters of the FBTF prototype are also measured. The results show that the PDL levels at the output of the TEF (Points A and B) are less than 0.1 dB (Figure 4.14(a)). This is resulted from a polarization balancing mechanism inside the 109 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

TEF, where the light travels back in the opposite direction to cancel the difference between two polarizations caused by the optical path. The maximum DGD figures are 1.1 ps and 0.8 ps respectively for Drop and Through ports (Figure 4.14(b)), while the CD levels are within ±3 ps/nm at all passbands (Figure 4.14(c)). These results show that the filter delivers noteworthy overall optical performances. The FBTF is also tuned to other positions, and the same measurement steps taken. The insertion loss spectra of all measurement points at three example positions are shown on Figures 4.15, with the tuning settings of the two TEFs indicated. These figures demonstrate the continuous passband width and centre wavelength tuning capability of the FBTF. The measured optical performances at these tuning positions are similar to those obtained at the first position.

4.3 Experimental Demonstration of Low Cost Expandable ROADM Node Based on FBTF 4.3.1 ROADM Node Architecture a. Motivations The two-degree tuning capability offered by the FBTF enables the flexible selection of channels to be added and/or dropped in an optical switching node while not requiring full demultiplexing of the WDM signal by an optical demultiplexer, wavelength blocker, or wavelength-selective switch. Therefore one of the main applications for the FBTF is a low cost yet flexible ROADM node. There are two major motivations for the FBTF-based low cost ROADM node architecture. Firstly, most of the traffic in the currently deployed optical switching nodes in the field does not require local processing (such as add, drop or regeneration), instead these channels are passed to the subsequent node transparently (called the Express channels or the Through channels). Since they usually travel a further distance in the WDM network, they are more sensitive to the optical loss in the transmission path compared with the locally added or dropped channels. Even though there are optical amplifiers to compensate for these optical losses, more amplifiers will lead to larger noise and worse OSNR, which in turn degrades the signal quality. Therefore it is desirable to give priority to these Express/Through channels. Secondly, often the channels travelling to the same destination (such as dropped, crossed or passed through) are adjacent to one another. It thus provides an opportunity to group them together

110 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

A B C Through Drop Through

(TEF 1: at 5473, TEF 2: at 1118)

A B C Through Drop Through

(TEF 1: at 5242, TEF 2: at 2037)

B C Through Drop Through

(TEF 1: at 4746, TEF 2: at 1587) Figure 4.15 More examples of centre wavelength and passband width tuning by FBTF.

111 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

and process them using a single or fewer switching elements, instead of fully demultiplexing them and use one or more optical switches for each individual WDM channel. Besides reducing the hardware requirement, group switching also reduces the filtering effect from demultiplexing individual channels and thus reduces ISI. b. Node Architecture With the use of the novel FBTF device, a low cost ROADM node that provides priority to Express channels and allows grouping of WDM channels travelling to the same destination can be designed and realised [117]. It has a modular architecture with a stackable feature. At the top of the stack, an Express Module is placed to select the channels for local operation and send the remaining Express/Through channels to the output. The remaining channels are sent to the cascade down port for local processing. This is achieved by using the 3-port FBTF. If no local operation is required for any input channel, this Express Module will be the only module at the node. When the needs for local add/drop or regeneration rise, corresponding modules can be added through the reserved cascade ports. For those channels that are processed locally, the respective processed channels (such as the added channels or channels crossed from other paths) are sent to the Express Module through the cascade up port (or called the return port) after the local processing, and these channels are added to the Express/Through channels to form the ROADM node output. Each added module (such as the Tunable Add/Drop Module or the Regeneration Module) also reserves cascade ports for subsequent module addition. This modular stackable architecture with upgrade through reserved cascading ports allows the future upgrades to be performed in-service without affecting the existing traffic. c. Modules for the ROADM Node Besides the Express Module, a number of modules with different functionalities are available for the stackable ROADM node to provide users the flexibility to configure the node according to the traffic condition and capital budget. Some key modules are:

• Express Module: It uses a 3-port FBTF at the input to separate the Through (Express) channels and the channels that require local processing. The locally 112 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

In Out

FBTF FBTF (a) Express Module

Cascade down Cascade up In Out TBF TBF (b) Optical ? ? Add/Drop Drops Adds Module CL DMUX/MUX Cascade down Cascade up

In 1In 2 Out 1 Out 2 FBFSW FBF (c) Cross- Connect Module Cascade Cascade Cascade Cascade down 1 down 2 up 1 up 2 In Out

FBTF FBTF (d) Return FBTF: Flexible band Module tunable filter TBF: Tunable band filter Cascade down Cascade up FBF: Fixed band filter CL: Colourless In Out DMUX: Demultiplexer FBTFOEO/AMP FBTF MUX: Multiplexer (e) SW: Optical switch Regeneration OEO: Optical-electrical- Module optical regenerator Cascade down Cascade up AMP: Optical amplifier

Figure 4.16 Schematics of some key functional modules for the stackable ROADM node.

processing channels are sent through a cascade down port to the subsequent module, while the passing-through channels are sent to another filter at the output end to be combined with the returned channels from the cascade up port after local processing. • Optical Add/Drop Module: This module has a filter to select the channel(s) that need to be dropped at the local node, and another filter for the corresponding add channels [118]. There are several variations to this module. Firstly, if a waveband of multiple contiguous channels is required to be dropped, a 113 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

waveband filter is used, and a demultiplexer is added to separate these channels. Otherwise a single channel filter is used and no further demultiplexing is required. Secondly, if the wavelength of the dropped channel or waveband is not fixed, a tunable channel filter or a tunable waveband filter is used instead of a static filter. In the cases of tunable waveband dropping, the demultiplexer should be a colourless unit, such as a cyclic AWG demultiplexer or 1:n optical de-interleaver (where n is the number of WDM channels in the dropped waveband), so that any contiguous n channels can be separated. Thirdly, the optical filter and/or the optical multiplexer at the add side can be replaced with a coupler to reduce the hardware cost. However the tradeoff is larger insertion loss and more crosstalk from the neighbouring channels. • Optical Cross-Connect (OXC) Module: If the ROADM is placed between two WDM rings or at a multi-degree node, an OXC Module can be added to switch the WDM channel or waveband between multiple paths. Small scale optical switch (such as 2×2 switch) is used for the cross-connect function. Similar to the Add/Drop Module, there are options to select single channel or multiple channels using different filter, and to select tunable or fixed wavelength operation. • Return Module: In the cases that the channels that require local operation are not contiguous, a Return Module can be used to pick up the isolated passing-through channel(s) and return them to the output. It has the similar configuration as the Express Module, where two FBTFs are connected in series to separate the channels-of-interest and the remaining cascading channels. Therefore they can be built using the same design. • Regeneration Module: This module uses a FBTF to pick up the channels that require regeneration and perform it through optical-electrical-optical (OEO) subsystem before returning it to the output via another FBTF. It can also use optical amplifier to boost the signal power of these particular channels without converting the signal to the electrical domain. Wavelength conversion can be performed if required, however the wavelength of the add filters needs to be tuned accordingly. • Other possible modules include optical equalization module, multicasting module, signal grooming module.

114 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

The schematics of some key functional modules are shown on Figure 4.16. It should be noted that only one example is used for each module. For example, only the flexible waveband add/drop example is shown for the Add/Drop Module. Also, only the components for the main function in each module are showed. Other non-essential or common components, such as the optical monitoring tap and photodetector, the optical amplifier, the variable optical attenuator, and the control circuitry, are not included. d. Node Features and Benefits The FBTF-based stackable ROADM node has several features: • The selection of the express channels and local operation channels are done in a “hitless” manner. In other words, only the related channels that need to change the destination will experience the tuning effect, while other channels (the existing traffic) will not be affected during the operation. This allows in-service system upgrade. • Since the node uses a modular structure with cascade port at each module, the users do not need to install the full system up front. Instead they only need to install the modules that are required at present. When future upgrade is needed, additional module(s) can then be installed without disturbing the existing modules. This allows a “pay-as-you-grow” investment strategy. • Besides providing flexible reconfiguration of optical add/drop operation as a single degree ROADM, it can be easily upgraded to a multi-degree ROADM node (or called a Wavelength Cross-Connect node) through adding OXC Modules. • There is a broad selection of modules to tailor for different requirements in any specific application. For example, the Add/Drop Module can be either flexible or tunable, and drop either single channel or a band of multiple channels. • Since it is fully transparent, the switching operation is bit-rate independent, modulation format independent, and protocol independent. For example, it can perform add/drop or wavelength routing on a 10 Gbit/s OOK system, or a 40 Gbit/s RZ-DPSK system, or a 100 Gbit/s PM-RZ-QPSK system, or any mixed line rate system. Therefore it is future proof. This is helpful especially since the

115 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

channel rate in the commercial WDM system keeps increasing and new advanced modulation formats are emerging constantly. • Unlike most ROADM nodes that uses fixed-grid optical demultiplexer to select or separate the channels, this ROADM can operate on any channel grid due to the continuous tuning capability of FBTF. For example, it can operate on 50 GHz-spaced system, or 100 GHz-spaced system. It can even operate on non- standard channel grids. • Because the express channels and the cross-connect channels do not need to be demultiplexed into individual WDM channels inside the ROADM node (unlike most of the other ROADM architectures), these WDM channel do not suffer from filtering effect in the node, therefore the signal degradation due to ISI is reduced (refer to Chapter 3 earlier) and the quality of the received signal is improved. This can be considered as a type of waveband switching [119]. • Compared with other ROADMA node architectures described earlier, this node occupies more compact footprint due to fewer hardware element and not requiring wavelength-selective switch. It also requires simpler control and consumes less electrical power. Most importantly, the hardware cost is lower.

There are some limitations to this low cost ROADM architecture. Firstly, it can only accept a finite number of modules in cascade due to the optical power budget limit. For example, if the ROADM node requires simultaneous add/drop of eight spectrally separated wavebands, it will requires eight Add/Drop Modules to be connected below the top Express Module. This will impose about 30 dB loss at the WDM channels dropped at the lowest module of the stack. A solution is to insert optical amplifiers in the middle of the stack to elevate the power level of the signals at the lower portion modules. The second limitation is that the Express Module can only select contiguous channels for local processing. If the channels that require local processing are separated in multiple spectrally discontinuous bands, one or more Return Modules will be required. Due to these limitations, this ROADM node architecture is not suitable for some applications, such as the examples described above. However, most ROADM nodes in

116 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

the field do not require add/drop or cross-connect on large number of spectrally separated bands simultaneously, and therefore this architecture is applicable.

4.3.2 FBTF-Based ROADM Node Prototype and Network Testbed a. Network Testbed In order to test the performance of the proposed FBTF-based ROADM node, a network testbed is designed and constructed. It includes two rings and four nodes (Figure 4.17).

Node 2: 1-D 40km SMF & ROADM matching DCF Express TOAD

Node 1: Tx Ring 1 Terminal Rx 40km SMF & Node 3: 2-D matching ROADM DCF 40km SMF & Express matching DCF TOAD OXC

Node 4: Tx Ring 2 Terminal Rx

Figure 4.17 Experimental WDM network testbed with stackable ROADM nodes.

The first ring consists of three nodes. The first node is a terminal node (Node 1), which provides multiple WDM signals to be tested in the network. These signals include a WDM channel carrying OC-192 SONET signal at 10 Gbit/s data rate, three channels carrying OC-48 SONET signals at 2.5 Gbit/s data rate through WDM transponders, and a channel carrying analogue video signal stream. Both the transmitter and the receiver for each signal are placed at the same node.

117 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

Node 2 is a single degree ROADM node. It has two modules in stack configuration. At the top of the stack is an Express Module, which selects the WDM channels to be processed locally. A Tunable Optical Add/Drop (TOAD) Module is connected through its cascade ports. This module has a four channel tunable band filter with colourless demultiplexer, so it can add/drop any four contiguous WDM channels. Node 3 is a two-degree ROADM node, because it is placed at the intersection between the first ring and the second ring. It consists of three modules. The top two modules are the Express Module and 4-channel TOAD Module, which are the same as Node 2. The third module is a fixed 4-channel OXC Module, which crosses four pre- determined channels between two rings. In the first ring, there is a 40 km standard SMF-28 fibre with corresponding dispersion compensation fibre (DCF) between any two adjacent nodes. Optical amplifiers are placed before and after the transmission fibres. The second ring contains only two nodes. Besides Node 3, this ring has a terminal node (Node 4). This node contains two sets of test signal transmitters and receiver. The first is an OC-48 SONET transponder, and the other is another analogue video source and receiver. These two channels can be crossed with the OC-48 signal and the video signal in Node 1 of the first ring. This network testbed enables the testing and demonstration of several key ROADM functions such as express channel selection, tunable optical add/drop, cross- connect between ROADM degrees, and mixed line rate, mixed protocol switching. This operation spectrum range for this testbed is from 1531.12 nm to 1561.42 nm, which consists of 20 channels at the C-band with 200 GHz spacing. The corresponded frequency range is from 192.0 THz to 195.8 THz. The 200 GHz channel spacing is used because of the FBTF prototypes, the key device in the ROADM architecture, operate on 200 GHz grid. b. Design and Construction of Modules Three types of modules are designed and fabricated for the two ROADM nodes, namely the Express Module, the TOAD Module, and the OXC Module. They are the most common modules in the proposed ROADM node architecture, therefore this testbed can demonstrate and test most of the common functions in the proposed ROADM architecture.

118 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

I. Express Module Figure 4.18(a) shows the schematic of the Express Module. It has the similar configuration as shown in Figure 4.16(a), with the exception that the second FBTF on the output side is replaced by an optical coupler to reduce the hardware cost and size. Optical power monitors are also added at various points of the circuit.

Input Output TEF

FBTF Coupler/splitter

VOA + tap + PD

Ta p + P D Cascade down Cascade up (a)

Tap+PDs

Couplers Power circuit

Tunable edge filter (High wavelength) VOA+tap+PDs

Tunable edge filter (Low wavelength)

Controller

(b) Figure 4.18 The Express Module prototype (a) Schematic, (b) Photo.

At the Input Port of the module, a Santec IPD-1 single channel integrated tapped photodetector (PD) is placed to monitor the input optical power level. It contains a 3%

119 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

tap splitter and a photodetector. It has a small footprint (21 mm length and 3.5 mm outer diameter) and low loss of 0.5 dB, therefore it is suitable to be integrated in the ROADM modules to monitor the signal level and provide loss of signal alert. After the tap PD, the remaining 97% of the input light is sent to the FBTF, which is consist of two TEFs – a low wavelength TEF and a high wavelength TEF. These TEFs are designed in house and fabricated using the technologies discussed in Section 4.2.3 above. Customized blue anodised metal enclosures are built to provide better protection from shock and fibre damage during module assembling. The two TEFs are connected in cascade, delivering the filtered band (the Drop channels) to the Cascade Down Port of the module to be processed locally in the subsequent modules. The remaining channels (the Through/Express channels) are combined using a 50:50 optical coupler. Two JDSU VCB voltage-controlled VOAs are inserted before the coupler input to eliminate the power imbalance between the left and right Through channels as discussed earlier. These two VOAs also have integrated tap and PDs to monitor the optical power levels of the input and output signals. The selected Through channels are combined with the locally processed channels returned from the Cascade Up Port through another 50:50 optical coupler. Another Santec IPD-1 integrated tap PD is placed to monitor the output signal before sending it to the Output Port of the module. The fibre connections between these optical components are fusion spliced instead of using fibre connectors to reduce the insertion loss and the hardware size. Figure 4.18(b) is the photo of the constructed Express Module prototype without the front panel and cover. A common printed circuit board (PCB) is designed for all the three types of modules in the prototype. The common elements include (1) a 250-pin connector and guide pin to connect the module to the back plane of the module shelf; (2) a power circuit that converts the 48 V DC power from the module shelf to the various voltage levels required for the components in the module (such as 3.3 V DC, 5 V DC, and 18 V DC); (3) a RCM2100 RabbitCore micro-controller module for receive the instruction command from computer and control individual components, as well as to collect the component information and report to the computer; (4) an Ethernet connector which acts at the interface between the computer and individual module; and (5) three light emitting diodes (LEDs) to indicate the status of the current module, and an LED array to graphically show the channels selected by the current module. More details of the control mechanism, computer interface, and LED indicator will be discussed in the later sections. The PCB also provides space and holders for winding 120 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

the extra fibres, as well as holders for small components such as couplers and fusion splice sleeves. The average insertion loss figures of the Express port and the Drop port of the fabricated TEF prototypes are 1.8 dB and 2.3 dB respectively. Based on the commercial product specifications, the insertion loss of a 50:50 coupler is 3.4 dB, the insertion loss of the Santec IPD-1 tap PD is 0.5 dB, and the insertion loss of the JDSU VCB VOA is 1.2 dB without attenuation. Therefore the insertion loss of the Cascade Down channels is 5.1 dB, and it is up to 12.1 dB for the Express channels after power balance by the VOAs. The Cascade Up channels experience 3.9 dB insertion loss in this module. It should be noted that even though these figure seem to show that the Express channels experience higher loss than the remaining channel that are processed locally, these locally processed channels actually experience higher loss in the ROADM node because they have more loss in the subsequent modules, which adds to the 5.1 dB and 3.9 dB figure here. Compared with other common ROADM designs and products, the 12.1 dB loss for the Express channels is low, especially considering that it includes the power monitoring and power balancing components in the optical path. The insertion loss figures here do not include the connector loss, which typically adds up to 0.2 dB per connection because the small form factor MU connectors are used.

II. TOAD Module The schematic of the TOAD Module is shown on Figure 4.19(a). It also has a Santec IPD-1 tap PD at the Input Port to monitor the incoming power level. It then has a tunable band filter to select the drop channels. The tunable band filter is also designed in house and fabricated using the same stepper motor-based mechanical tuning method as the TEFs. This is a 4-skip-0 filter, which means that it has a passband width of about 800 GHz which covers four adjacent 200 GHz WDM channels. All the remaining 200 GHz-spaced channels are sent to its reflection port and not “skipped”. Since the dropped channels are tunable, a colourless demultiplexer is built to separate these four channels. It consists of three interleavers. The first interleaver is an 800G/400G interleaver, and each of its two outputs is connected to a 400G/200G interleaver. With this design, any four adjacent channels can be separated and distributed to the four Drop Ports. Optoplex interleavers are used to build this colourless demultiplexer. The remaining channels are sent to the subsequent module through the Cascade Down Port. For the add side, only one Add Port is provided to simplify the design. The added 121 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

signal is combined with the signals from the Cascade Up Port by an optical coupler and sent to the Output Port after going through another power monitor. If more than one channel needs to be added, another optical coupler can be placed outside of the module to combine them before sending them to the Add Port. Colourless multiplexer is not required at the add side. Optical coupler is a low cost solution which provides the same function, and it even allows non-adjacent channels to be added. Figure 4.19(b) shows the photo of the constructed TOAD Module prototype.

Input Output Tunable band filter

Colourless Interleaver Demux Coupler/splitter

Tap + PD

Cascade down Drop 1-4 Add Cascade up (a)

Tap+PDs

Power circuit Interleavers

Tunable band filter

Controller

(b) Figure 4.19 The TOAD Module prototype (a) Schematic, (b) Photo.

122 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

The average insertion loss figures of the Express port and the Drop port of the fabricated tunable band filter prototypes are 1.9 dB and 2.3 dB respectively. The insertion loss of the colourless demultiplexer constructed by cascading Optoplex interleavers is 2.5 dB. Therefore the insertion loss of the Cascade Down channels and the Drop channels are 2.8 dB and 4.9 dB respectively. For the add port, both the Add channels and the Cascade Up channels experience 3.9 dB loss from the 50:50 coupler and the Santec power monitor.

III. OXC Module The schematic of the OXC Module prototype is shown on Figure 4.20(a). This is a 2×2 OXC Module, with fixed band switching. It has two Input Ports, each connects

Input 1 Input 2 Output 1 Output 2 Fixed band filter

VOA + tap + PD

Tap + PD

Optical switch Cascade Cascade Cascade Cascade down 1 down 2 up 1 up 2 (a)

Tap+PDs Fixed band filters

Power circuit

VOA+tap+PDs Optical switch

Controller

(b) Figure 4.20 The OXC Module prototype (a) Schematic, (b) Photo.

123 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

to a Santec IPD-1 power monitor. Each input signal is then sent to an AFOP fixed band 200 GHz 4-skip-0 filter which covers four 200 GHz grid channels with centre frequency ranging from 192.8 THz to 193.4 THz. The selected channels are sent to a JDSU SN 2×2 optical switch. If the cross-connect function is required, the SN switch will be set to the cross state, otherwise it will remain at the bar state. The unselected channels are sent to the Cascade Down Ports for operation by the subsequent modules. Two JDSU VCB VOAs with integrated tap and PD are placed before the SN switch to equalize the optical signal levels between the two inputs. At the output of the SN switch, another two AFOP fixed band filters with the same passband frequency are placed to combine the Cascade Up channels to the respective OXC channels before sending to the power monitor and then to the Output Ports. The photo of the constructed OXC Module prototype is shown on Figure 4.20(b). The average insertion loss figures of the Express port and the Drop port of the AFOP fixed band filter are 0.5 dB and 1.0 dB respectively. The insertion loss of the JDSU SN 2×2 optical switch is 0.9 dB. Therefore the insertion loss of the Cross- connect channels is 4.1 dB without additional attenuation from the VOAs. The insertion loss is 1.5 dB for both the Cascade Down and the Cascade Up paths.

IV. External Hardware After assembling the individual components onto the PCB for each module, a front panel is designed and attached to the PCB. Figure 4.21 shows the front panel of the TOAD Module, but it is essentially the same for all the other modules. On the top left side of the panel, there are three LEDs to indicate the operation status of the module. The first LED is a green LED. When turned on it shows that the module is powered and operating properly. The second LED is a red LED indicating that there is some error with the current module. The third LED is also a red LED, which indicates there is some faulty condition in the entire network. These LEDs are controlled by the micro-processor in each module, which performs routine status check on all components in the module and communicates with other modules through a centralized computer. Below the three Status LEDs is an array of 20 LEDs called the Channel LEDs. They provide a visual indicator that shows the corresponding channels being selected by the current module. These LEDs are bicolour ones which can display either green or red output colour. Figure 4.22 shows three examples of the change of Channel LED 124 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

status in an Express Module, where the green LEDs indicate the band of WDM channels being selected by the FBTF for local processing, and the red LEDs indicate the remaining Through channels. These LEDs are also controlled by the micro-controller based on the current module configuration.

Controller

Ethernet cable

Status LEDs Channel LEDs

Ejector Ethernet connector In/Out ports Figure 4.21 Front panel of the TOAD Module.

Selected Remaining channels channels

Figure 4.22 Example of the operation of Channel LEDs.

The front panel has three rows of optical connectors. MU connectors are used because it has a small form factor. Each row has two duplex MU adapters side by side,

125 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

making the total of 12 optical connectors for each module. Not all the 12 connectors are used, based on the actual module configuration. These fibre connectors are placed at about 60 degree angle to prevent laser damage to the operator’s eyes. Having such angle also makes the fibre routing inside the module easier. There is an RJ-45 Ethernet connector at the bottom of the panel. It links to the internal micro-controller to the centralized controller computer. There are also ejectors and fasteners on the front panel to secure the module onto the shelf. In this prototype, NEC SPWV-40 shelves are used. Each module occupies two slots on the shelf.

Figure 4.23 A complete prototype module with front panel and cover.

Node 2 Node 3

Figure 4.24 Two ROADM nodes with five modules in the NEC SPWV-40 shelf.

126 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

At the final step, a black metal cover is placed to enclose the components in each module. Figure 4.23 is the photo of a completed module. Figure 4.24 shows the five constructed modules plugged into a NEC SPWV-40 shelf. Node 2 consists of an Express Module and a TOAD Module, and Node 3 consists of an Express Module, a TOAD Module, and an OXC Module. Even through they are physically located in a same shelf in the experiment, there are 40 km SMF fibre, corresponding DCF and an optical amplifier between these two nodes. c. ROADM Control Configuration and Software I. Communication between Controller Computer and the ROADM Modules In the ROADM prototype, external computer control is set up instead of the on- shelf control and management as in the commercial switching node products, and the control commands and data are transmitted through TCP over Ethernet cable instead of through the shelf back plane. As described above, the back plane connection only provides the 48 V DC power. Each ROADM module has a RCM2100 RabbitCore micro-controller core module from Rabbit Inc., as shown in Figures 4.18-4.20. Each core module integrates a 22.1 MHz Rabbit 2000T microprocessor, 512k SRAM, 34 parallel user I/O, as well as an Ethernet connection. There is an Ethernet cable to extend it to the front panel of the module. Using an Ethernet switch, these modules and the centralized controller computer are linked and set under a same subnet. Transmission Control Protocol (TCP) is used to communicate between the computer and the micro-controllers in the ROADM modules. A set of commands is defined to allow the computer to read the status of each component in each module, and to control the operation of each component. After receiving these operation commands, the RabbitCore module translates them into appropriate settings of the module I/O pins. These voltage settings travel on the PCB to reach and control the corresponding components. On the other hand, when a read status command is executed, the statuses of these components, such as the output voltages of tap PDs, are received by the RabbitCore module through the I/O pins and translated into corresponding status data and sent to the controller computer via TCP. Here the central controller computer acts as the TCP client, and the individual modules acts as the TCP server. The server RabbitCore module is always open and waiting for call from the client. The client (the management software on the controller computer) calls each server and establishes a TCP connection in the beginning of the 127 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

control software program, and this connection does not close until the program ends. The TCP port number is set to be 8080. The command body is a string, which consists of two sections: the address section and data section. The address section contains two bytes (two characters). The first character indicates the component type, and the second character indicates the component number within the module. The data section has variable length between 1 byte and 7 bytes, depending on the type of component the command is sent to. Besides communicating between the controller computer and the individual devices, the RabbitCore micro-controller also automatically performs routine check of the status of each optical component that has electrical feedback (such as tap PDs, VOAs, and tunable filters) to identify any system or component fault, and displays the status on the Status LEDs. It also calculates the channels processed by the corresponding module (such as the channels selected by the FBTF in the Express Module, the add/drop channels selected by the filter in the Optical Add/Drop Module, and the cross-connect channels selected by the filter in the OXC Module) and sets the appropriate illuminating colour for each Channel LED.

II. Management Software The management software in the central controller computer is written using LabVIEW graphical language. It is custom designed for the constructed network testbed, and it has three levels of control: the network level, the node level, and the individual module level. Figure 4.25 is the network level management interface. It shows the schematic of the 4-node 2-ring network prototype. For each node, it can display the spectrum at the node output or select to view the details in the node. The displayed spectrum (such as the example shown in inset) is the target spectrum, and not the actual spectrum measured by OSA. When “Display details” is selected, the next level (that is, the node level) GUI is popped up in the screen. The node level management interface displays the modules presented in the node. For each module, the states of the three Status LEDs are shown, as well as the key component such as the FBTF for the Express Module and the filter for the Optical Add/Drop Module. Basic control of these key components is available at this level. For example, the user can select the express channels using the FBTF in the Express

128 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

Inset

Figure 4.25 Network level management interface. Inset: displayed target spectrum.

(a)

(b)

Figure 4.26 Node level management interfaces for (a) Node 2 and (b) Node 3. 129 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

Module, or change the state of the cross-connect switch in the OXC Module. Figure 4.26 shows the node level management interface for Node 2 and Node 3 in the ROADM testbed. In this interface, each node can accommodate up to eight modules. After configuring a node, the user can return to the upper level (that is, the network level) interface by clicking the “OK” button. Or the user can get access more detail control of each module by clicking the “Details” button. The module control interface will then pop up.

(a)

(b)

(c)

Figure 4.27 Module level management interface for (a) Express Module, (b) Tunable Optical Add/Drop Module, and (c) OXC Module. 130 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

Figure 4.27 are the module level interface for the Express Module, Tunable Add/Drop Module and the OXC Module constructed for the ROADM prototype. As shown, this level of GUI provides more detail information than in the node level interface. These details include the schematic of the module configuration, the wavelength and frequency of the channels (instead of only the channel number), as well as the real time readings from the optical power monitors. It also indicates the node number and module number. There is also an “OK” button for user to close this GUI and return to the node level interface.

4.3.3 FBTF-Based ROADM Experiment and Results a. Experiment setup Figure 4.28 shows the testbed configuration. Node 1 is a terminal node which contains five test channels (Inset). The first three are from OC-48 SONET transponders with data rate of 2.5 Gbit/s. They are the commercial SPWV C40 series product made by NEC. In the 20-channel 200 GHz-grid system, their channel numbers are Ch 2, Ch 3 and Ch 4, which correspond to frequencies of 192.2 THz, 192.4 THz and 192.6 THz respectively. An Anritsu MP1656A Portable STM-16 Analyser (SONET analyser) is used to provide the test client signal into the transponder. The transponder converts the signal to the line signal with WDM wavelength. The signal received by the transponder is sent to the SONET analyser to calculate the error rate.

Node 2: 1-D 40km SMF & E ROADM matching DCF Express F D TOAD G

C C Node 1: Tx Ring 1 Inset Terminal Rx B CPL 40km L A SMF & Node 3: 2-D matching Multiplexer T.L. Mod. Ethernet I ROADM DCF tester OC-192 V. Tx 40km SMF & Express H SONET OC-48 Tx Tx Tx T. L. matching DCF TOAD analyzer Camera SONET OC-48 TPND 1~3 OXC analyzer Rx Rx Rx P.D . P.D . K J

Demultiplexer

Tunable filter Node 4: Tx Ring 2 Terminal Rx

Figure 4.28 ROADM network testbed and experiment system.

131 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

The next test signal is an analogue video signal. The video source is a live video captured by a digital CCD camcorder (Canon GL1 model) and a modified video transmitter from NTK, and the optical source is a Santec ECL210 tunable laser with external modulation port. The video signal has a bandwidth of about 6 MHz since it is a standard definition video. It is sent to the NTK video transmitter through an S-video cable and modulated into RF signal with carrier frequency of about 60 MHz. This RF signal is used to modulate a Santec ECL210 tunable laser, which has a RF modulation range of 1 to 400 MHz. The last test signal in Node 1 is a Gigabit Ethernet signal, which is generated by Agilent N2X Multi-services Tester. The electrical signal is fed to a SDL IOAP-

MOD9170 10G LiNbO3 amplitude modulator, which modulates the output of an Alcatel A1905 DFB laser source tuned to the ITU-T channel grid. At the receiver side, the optical Gigabit Ethernet signal is received by HP 11982A amplified lightwave converter to convert back to electrical signal. The received electrical Gigabit Ethernet signal is then fed back to the Agilent N2X Multi-services Tester to calculate the error rate. Among these five test channels in Node 1, the OC-48 channels have fixed wavelengths. They are combined using an NEC SPWV C-40 multiplexer. The wavelength of other two signals, namely the analogue video signal and the Gigabit Ethernet signal, can be tuned. Therefore a three-port optical coupler is used to combine them with the multiplexed OC-48 signals. At the receiver end, a 3-port tunable optical filter from Optoplex is used to select the Gigabit Ethernet channel. The remaining channels are sent to the SPWV C-40 demultiplexer to be demultiplexed. Node 2 is the 1-D ROADM node with an Express Module and a TOAD Module as described above. A set of analogue video receiver is connected to the Drop Port of the TOAD Module. The main component is a NTK video receiver to convert the optical signal back to the baseband analogue video stream, which is displayed on an LCD monitor. Node 3 is the 2-D ROADM node with an Express Module, a TOAD Module and an OXC Module as described above. It also has a test signal as the added signal. This is a 10 Gbit/s OC-192 SONET signal. An Agilent SpectralBER 10GB/s Bit Error Rate Test System is used. Because the optical source of the SpectralBER analyser is not on the standard ITU-T WDM grid and is not tunable, I substitute it with a tunable source by cutting the fibre between the source laser and the modulator and splicing the output fibre pigtail of an Agilent 8168C tunable laser onto the modulator input. The receiver 132 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

side of the OC-192 SONET test signal is located at Node 1 (Inset of Figure 4.28). No modification is required because the photodetector can operate on any wavelength within the operation spectrum window. Node 4 is a simple terminal node which only has one set of analogue video transmitter and receiver. This video source is a video tester which generates different test patterns with different colours. Another Santec ECL210 tunable laser is used as the optical source for the video signal. Among these test signals, the SPWV OC-48 SONET transponders are located in NEC’s SPWV C-40 shelves, since they are part of the commercial SPWV C-40 product line. The other test signals and related instruments, including the OC-192 SONET signal source and Agilent SpectralBER analyser, digital oscilloscope, video camera, video tester, tunable lasers, video receivers, LCD monitors, Gigabit Ethernet transmitter and receiver, N2X Multi-services Tester, are located externally. The WDM multiplexers, demultiplexers, and optical amplifiers are also part of the NEC SPWV C- 40 system and are located in the SPWV C-40 shelves. The transmission fibre and dispersion compensation fibre are located externally. Other external items include a laptop as the centralized controller PC and an eight-port Ethernet switch for communication between the controller PC and each ROADM module.

Transmission fibres Video source 1 ASE Lasers for videos source OC-48 SPWV Digital scope transponders Node 2 Node 3 bay Video receivers SPWV shelf OSA 1

Controller PC Ethernet hub OC-48 SONET OSA 2 analyzer Amplifiers OC-192 SONET Lasers for OC- analyzer 192 signal Mux/Demux

Figure 4.29 Photo of the ROADM network testbed and experiment system.

133 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

The network prototype and the experiment system are shown on Figure 4.29. Each SPWV bay contains two shelves, the upper shelf is for transponder modules and the ROADM prototype modules, and the lower shelf contains optical amplifier and multiplexer/demultiplexer. Altogether three SPWV bays are used in this experiment, containing six SPWV shelves. b. Network Operation Experiment Firstly the network operation of the constructed ROADM prototype nodes is tested. In this experiment, the wavelengths of both video channels are tuned to 193.0 THz (Channel 6), the Gigabit Ethernet signal is tuned to 194.4 THz (Channel 13), and the OC-192 SONET channel is tuned to 193.2 THz (Channel 7). The Anritsu MP1656A SONET analyser is connected to OC-48 channel #1 (192.2 GHz). The FBTF in the Node 2 Express Module is tuned to pass a band that consists of Channels 5 to 8. The tunable band filter in the subsequent TOAD Module is also tuned to the same range. In Node 3, the Express Module selects Channels 5 to 12 for local process. The TOAD Module selects Channels 7 to 10 for add/drop and passes the remaining channels to the subsequent OXC Module. The fixed wavelength OXC Module has cross-connect function for Channels 5 to 8. The operation of the seven test channels are listed in Table 4.2 below. The three OC-48 SONET channels and the Gigabit Ethernet channel are the Express channels which travel through the entire Ring 1 and returns to Node 1; The video channel from Node 1 is dropped at Node 2; The OC-192 SONET channel is added from Node 3 and travels to Node 1 to be received; The video channel from Node 4 is either crossed to Ring 1 to reach Node 1, or remains at Ring 2 and goes back to Node 4, depending on the state of the OXC switch. This arrangement demonstrates express, add, drop and cross-connect operations in the proposed ROADM node. Figure 4.30 shows the optical signal spectra measured at various points in the testbed. They are measured by Ando AQ6317B OSA. The resolution bandwidth used in the measurements is 0.1 nm. The captions of the subplots correspond to the measurement points in Figure 4.28. Figure 4.30(a) and (b) illustrate the 3-channel OC-48 SONET signals before and after the first post-EDFA. The peak levels are about -24 dBm and -3 dBm respectively, therefore it can be obtained that the gain of the EDFA is about 21 dB. However the OSNR is reduced from > 40 dB before the amplifier to about 33 dB after amplification. 134 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

Table 4.2 Test channels and operations in the ROADM network operation experiment. Channel Frequency Signal Source Destination Operation No. (THz) OC-48 #1 2 192.2 Node 1 Node 1 Express OC-48 #2 3 192.4 Node 1 Node 1 Express OC-48 #3 4 192.6 Node 1 Node 1 Express Video #1 6 193.0 Node 1 Node 2 Drop Gigabit 13 194.4 Node 1 Node 1 Express Ethernet OC 192 7 193.2 Node 3 Node 1 Add Video #2 6 193.0 Node 4 Node 4 Cross

(a) (b)

(c) (d)

135 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

(e) (f)

(g) (h)

(i) (j)

(k) Figure 4.30 Measured optical signal spectra at various points in the ROADM experiment testbed.

136 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

Figures 4.30(c) shows the spectrum of the Gigabit Ethernet signal, analogue video signal, and the three OC-48 SONET signals combined by the optical coupler at the output of Node 1. Figure 4.30(d) is the same signal after transmitting through the first span of 40 km SMF and matching DCF, and after the pre-EDFA before the OADM Node 2. The signal level at the input of Node 2 is about -4 dBm to -9 dBm, and the OSNR is slightly reduced to about 30~31 dB after passing through another EDFA. Figure 4.30(e) is the spectrum at Node 2 output. Here the analogue video channel has been dropped, leaving only four channels. The insertion loss experienced by these four express channels is about 12 dB. This insertion loss figure is slightly lower than the expected value, mainly because the actual insertion loss figures of the optical components are better than the specified values, which are the worst levels that the manufacturers guarantee to meet or exceed. Figure 4.30(f) is the spectrum at the Node 2 TOAD Module Drop Port, which contains the dropped analogue video channel. The insertion loss experienced by this channel is about 9 dB, which is also slightly lower than the expected value calculated from the device specifications. Figure 4.30(g) is the spectrum at the input of Node 3, after passing through another EDFA and fibre span. The peak is about -10 dBm for the OC-48 SONET channels and -17 dBm for the Gigabit Ethernet channel. The OSNR is slightly reduced again to about 27~28 dB. Figure 4.30(h) is the spectrum at the output of Node 3. Again it could be observed that the express channels experience about 12 dB insertion loss from the node. In this node, the OC-192 SONET channel at 193.2 THz (corresponding to the wavelength of 1551.72 nm) is added. It has about 10 dB higher power than the OC-48 SONET channels. There is a small band at about 1546 nm. This is the filtered band reserved for a special control channel for SPWV C40 system. Since this experiment and testbed do not require actual SPWV system control, no signal is applied to this channel. However the ASE noise within this filtered band builds up and is thus shown in the measured spectrum, especially after going through more amplifiers. Figure 4.30(i) is the second analogue video signal generated at Node 4. It is sent to Node 3 for cross-connect operation with Ring 1. In this experiment the OXC switch is set to the bar state (that is, not crossed), so this video channel remains at Ring 2 and returns to Node 4. Figure 4.30(j) is the returned signal. The insertion loss after going through Node 3 is about 4 dB, which agrees with the expected value of 4.1 dB calculated from the device specifications.

137 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

Figure 4.30(k) is the signal transmitted back to Node 1. It contains the same five channels at Node 3 output after going through a post-EDFA, final span of fibre, and the pre-EDFA. The peak values for the OC-48 SONET channels, the Gigabit Ethernet signal, and the OC-192 SONET signal, are about -7~-8 dBm, -16 dBm, and +1 dBm respectively. The OSNR for the express OC-48 channels is about 25 dB. The network operation experiment above demonstrates that the constructed ROADM nodes are capable of delivering various functions at individual channel or band level, and the insertion loss figures agree with the expected values. The measured insertion loss figures are consistent with the expected values. These values show the through traffic priority feature of the proposed modular ROADM design based on the FBTF device. c. Signal Quality Experiment After confirming the function of the ROADM node, the qualities of the various test signals are measured. The switching experiment of the analogue signals show that the video streams can be switched between the two output video monitors by controlling the optical switch in the Node 3 OXC Module. No degradation is observed in the video quality. No observable delay is found during the switching transient period either. The Agilent N2X Multi-services Tester is connected between the Gigabit Ethernet transmitter and receiver at Terminal Node 1. Ethernet Test Signal function (ETHTest) provided by the N2X tester is used to perform one-way on-demand diagnostic tests. Data test patterns with pseudo-random bit sequence (PRBS) generated by the tester are fed through this link. No error is shown from the received signal during an observation period of 30 minutes. The BER of the OC-48 SONET signal is also tested at Terminal Node 1 before and after the transmission in Ring 1. The test signal uses pseudo-random OC-48 SONET frames generated by an Anritsu MP1656A analyser. B1 error rate was monitored for regeneration section bit error. A manual VOA is placed before the received signal to adjust the optical power launched into the SONET analyser, and a 50:50 splitter is used to monitor the real time optical power with a power meter. The OC-48 #3 channel with frequency of 192.7 THz is tested. Firstly the back-to-back performance is measured, then the signal is transmitted to Node 2 and dropped at the TOAD Module and the BER is measured, finally the signal is transmitted to Node 3 and 138 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

crossed to Ring 2, the BER at the OXC Module output is measured. Figure 4.31 shows the measured BER curves at these three measurement points at different received power levels. The results show that the power penalty caused by the Node 2 is less than 0.2 dB at 10-12 BER level. When passing through both Node 2 and Node 3, the power penalty is less than 0.4 dB at 10-12 BER level. These results show that the ROADM node imposes little degradation to the OC-48 SONET signal.

8R

8R

8R

8R

8R  

8R 11 . 11 .U 8R

8R

8R R8 R R8 R R8 R R8 R

VHV10VR ]Q1V` ^RI_

Figure 4.31 BER performance for the OC-48 SONET signal in the ROADM testbed.

The transmission quality of the OC-192 SONET signal is also evaluated. This test signal is generated by the Agilent SpectralBER 10G analyser and carries a payload of 231-1 PRBS. In the test, the SONET signal is launched in Node 1 with the frequency of 194.1 THz. The BER and eye diagram are measured at the back-to-back location (output of Node 1), Express Module output (output of Node 2), Node 2 TOAD Module Drop Port, and Node 3 OXC Module Output Port respectively. The VOA and integrated 50:50 splitter setting is also used. In addition, a manual optical band pass filter from Santec with 0.3 nm bandwidth is placed before the VOA to remove the noise outside the bandwidth of the SONET signal. Figure 4.32 (a) and (b) are the eye diagrams measured by Agilent Infinium 86100C sampling oscilloscope at the back-to-back position and the Node 2 Express Module output respectively. Both eye diagrams have the similar eye opening of about 15 mV and have similar eye profiles. This shows that the ROADM node has negligible influence on the signal quality of the Express channels.

139 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

(a)

(b)

Figure 4.32 Eye diagrams of the OC-192 SONET signal at (a) BTB position and (b) ROADM node Express Module output.

Figure 4.33 shows the measured BER curves at various measurement points. The vertical axis is the –log(BER) in logarithm scale, and the horizontal axis is the received power level. It can be obtained that the power penalty is only about 0.05 dB for the Express and Cross-connect Channels and about 0.04 dB for the Drop Channel. This demonstrates that the performance of the ROADM meets the requirement of the WDM network applications [120]. -log(BER)

Received Power (dBm) Figure 4.33 BER performance for the OC-192 SONET signal in the ROADM testbed.

140 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

The switching speeds of the nodes are also tested. The switching time of the OXC Module is less than 10 ms, and the tuning speed of the Express Module and the TOAD Module is 200 ms between adjacent channels and about 1 second from end to end of the operation wavelength range.

4.4 Conclusions In this chapter, I proposed and demonstrated a novel tunable optical filter, namely the FBTF. In contrast to the single degree (centre wavelength) tuning function of the conventional tunable filters, this filter allows two degree tuning operation in both the centre wavelength and the passband width. I then designed a device prototype and fabricated it using cascading tunable edge filters based on dielectric thin-film structure and mechanical tuning method. This prototype demonstrated various flexible passband tuning functions. Using this tunable filter as a key building block, a modular ROADM node architecture is proposed. This ROADM node architecture has the benefit of giving priority to the express/through channels. Because it does not require large number of switching/tuning elements such as in the DSM-based, B&S-based, or WSS-based ROADM nodes, it provides a low cost solution for reconfigurable add/drop in the metro WDM network applications. With its large module selection, the ROADM node can also deliver other functions such as OXC. The modular architecture makes the node expandable and allows for easy customisation and “paid-as-you-grow” investment strategy. To verify and demonstrate the function of such ROADM node, a testbed was constructed. It consists of two inter-connecting ring networks and four nodes. Among the four nodes, two are the terminal nodes for signal source and receiver, and the other two are the ROADM node prototypes with Express Modules, TOAD Modules and OXC Module. Different switching operations were successfully performed on the testbed and the signals’ eye diagrams and BER performances are measured. The results showed that the proposed ROADM node can deliver all the required add/drop/cross-connect functions with little signal degradation. Therefore this FBTF-based ROADM node is suitable for application in metro WDM networks.

141 )%, 9)%, 9 %:*% & .*% )%, -#, %:*% & .*% )%, -#, %:*% & .*% )%, -#, 

Besides the application in the modular ROADM node, the FBTF device can also be used in other WDM network applications such as non-uniform waveband aggregator and de-aggregator [121] and reconfigurable waveband cross-connect node [37]. This work has been published in three international conference papers [39, 117, 120]. Two US patents have been issued [114, 121], with one more pending [118].

142 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

)%, )%, ;;;     

   6   6

 

The next generation ROADM nodes need to provide colourless, directionless and contentionless switching capabilities. As a result, the conventional passive optical WDM demultiplexer can no longer be used in the ROADM node’s transponder aggregator. The transponder aggregator designs so far require wavelength selector, which adds cost, size and power consumption to the node. In this chapter, I propose a new design of transponder aggregator which does not require wavelength selector (or called filterless). Instead it uses the local oscillator at the colourless transponder to select the target channel from all dropped channels. Through theoretical analysis and experiments, I demonstrate that this filterless transponder aggregator can achieve less than 0.5 dB OSNR penalty between receiving a single channel and receiving 96 DWDM channels, provided that the balanced receivers have reasonable common mode rejection ratio characteristic and there is sufficient power difference between the local oscillator and the per-channel signal. As part of this work, a method to measure and adjust the common mode rejection ratio of a balanced photodetector is proposed and demonstrated. This filterless transponder aggregator could significantly reduce the cost, size, power consumption in the next generation ROADM node and improve reliability and performance.

5.1 Background 5.1.1 Colourless and Directionless ROADM As described in the previous chapter, ROADM node has been widely deployed in long haul and metro WDM networks in the past few years. It allows the flexible adding and dropping of any or all WDM channels at the wavelength layer. A multi-

143 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

degree ROADM node (a node with three degrees or higher) also provides cross- connection function of WDM signals among different paths. As the traffics of the global optical network become more dynamic and the network topologies evolve from ring to mesh or meshed ring, the current ROADM nodes exhibit some limitations, in particular. (1) the coloured transponder assignment issue where each transponder corresponds to a fixed wavelength and therefore all transponders need to be preinstalled (high capital expenditure) or manually provisioned during system reconfiguration and upgrade (high operation expenditure), and (2) the directed add/drop switching issue where the add/drop operation of each degree in the node is separate and the transponders cannot be shared among different degrees, which limits the network’s routing, restoration and rerouting capability. To overcome these limitations, colourless and directionless (CL&DL) multi- degree ROADM has attracted significant research interests lately [40-42, 122-1127]. In such ROADM, the add/drop ports are not wavelength specific and any channel from any input port can be dropped to any transponder connected to the node, and each transponder can be tuned to any DWDM channel. Similarly, each added channel can be switched to any output port, regardless of which input port the corresponding drop signal came from. These features allow full automation of wavelength assignment with pay-as-you-grow investment strategy, as well as more efficient sharing of transponders in a node among different paths and better protection scheme. The most straightforward method to achieve CL&DL switching is to fully demultiplex all the input channels from all input ports, and use a large dimension fibre switch to switch these individual input channels and individual newly added channels to respective output ports or drop ports [122, 123]. It requires a large fibre switch (also called spaced switch or photonics cross-connect) with the dimension of [(L+K)×N] × [(L+K)×N] where N is the node degree, K is the total number of DWDM channels from each input, and L is the maximum number of local add/drop channels per degree. This is not practical because the large dimension fibre switch is costly and it presents the potential problem of single source of failure. The common method is to have a dedicated subsystem to CL&DL switching operation. Here it is called the Transponder Aggregator (TA) (Figure 5.1). In a TA, all channels to be dropped locally are combined through an aggregation device such as wavelength-selective switch (WSS) or coupler. These aggregated drop channels are

144 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

sent to respective transponders through a channel separation unit. For the add side, the added signals are combined, then multicast and selected by appropriate output ports.

North ROADM WSS SPL Input from all Added signal West East degrees… to all… degrees ROADM ROADM

WSS SPL SPL WSS Aggregated drop channels Combined add signals

WSS SPL Channel separation CPL … … Transponder (a) TPND 1 … TPND n Aggregator Drop Add TPND … TPND

WSS: Wavelength-selective switch SPL: Optical splitter TPND: Transponder CPL: Optical coupler Figure 5.1 A 3-degree colourless and directionless ROADM node. Inset (a): Schematic of a transponder aggregator.

5.1.2 Wavelength Selection Methods in Transponder Aggregator Different methods and hardware configurations are used to perform channel separation and enable each colourless transponder to receive the correct WDM signal. In Method 1, the n aggregated drop channels are demultiplexed using an optical demultiplexer with fixed wavelength assignments, followed by an n×n fibre switch for channel selection (Figure 5.2(a)) [124, 125]. In Method 2, a 1×n WSS selects and sends each of the n drop channels to the respective output port, which connects to the targeted transponder (Figure 5.2(b)) [40, 41]. Since the WSS with port count higher than 1×9 is not commercially available yet, the drop signals can firstly be split into x parts using a 1:x optical splitter, and then use x units of standard WSS to separate them (Method 3, Figure 5.2(c)) [42]. Here x = FVn / 9 if 1×9 WSS’s are used. Method 4 uses 1:n optical splitter to broadcast the drop channels into n equal shares, and then uses an array of n tunable filters to select the channel for each transponder (Figure 5.2(d)) [126]. All these methods use some wavelength selector, such as demultiplexer, WSS, or optical filters. These devices are costly, as analysed in Ref. [122]. They also require more space due to complicated optics and control circuitry. Since they are active wavelength selection

145 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

operations involved, these channel separators need electrical control and thus consumes electrical power at various degrees.

Aggregated Aggregated Aggregated Aggregated drop channels drop channels drop channels drop channels

Fixed Demux 1xn high port 1:x splitter 1:n splitter … count WSS … … … Tunable Tunable nxn fibre switch … 1x9 WSS 1x9 WSS filter filter … … … T T T T T T T T P P P P P P P P N … N N … N N … N N … N D D D D D D D D 1 n 1 n 1 n 1 n (a) (b) (c) (d) Figure 5.2 Various channel selection methods in Transponder Aggregator: (a) Using fixed demultiplexer and fibre switch; (b) Using high port count WSS; (c) Using splitter and standard WSS; (d) Using splitter and tunable filter array.

All the TA designs so far require some wavelength selective element (or called the wavelength selector), such as WDM demultiplexer, WSS, and tunable filter array, to perform channel separation before the signals reach the transponders. In this chapter, a novel TA without wavelength selector is proposed and demonstrated experimentally. It simplifies the design and significantly reduces the hardware cost while maintaining similar performance.

5.2 Development of Transponder Aggregator without Wavelength Selector 5.2.1 Operation Principle Here a TA subsystem design without requiring wavelength selector is proposed. It is applicable for system with coherent receiver. In this method, the channel separation unit only contains a passive 1:n splitter, which splits the drop channels into n equal parts (Figure 5.3). This is similar to Method 4 above, however the tunable filters are not required to select one channel for each transponder, instead each transponder receives all of the n WDM channels. The channel selection is performed within the transponder through tuning the wavelength of the local oscillator (LO) laser in the

146 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

coherent receiver. This laser is tunable since the transponders are tunable in colourless ROADM.

Aggregated drop channels

1:n splitter …

T T P P N … N D D 1 n Figure 5.3 Proposed channel separation method in the TA without wavelength selector.

According to this method, the channel separation unit no longer needs to provide channel selection function, but instead simply splits the WDM signals into n multiple parts. Therefore this method does not require any costly wavelength selection hardware in the TA. This significantly reduces the hardware cost of the TA. It also makes the TA (and the overall ROADM node) more compact and has lower power consumption because the channel separation unit only consists of a passive splitter. As described above, a key requirement for this technique is that the system uses coherent receiver. Even though DWDM transmission systems with coherent receiver are more costly and complicated than systems with direct detection, it is common (almost always used) in transmission systems at per channel data rate of 100 Gbit/s and beyond. This includes the transmission of signals using single carrier modulation formats [3, 17, 128-131] (such as PDM-QPSK) and using multi-carrier modulation formats [5, 16, 132-134] (such as OFDM). Coherent system has also been proposed for 40 Gbit/s applications [135, 136]. The applications of these 40Gbit/s, 100 Gbit/s and beyond systems range from ultra long haul terrestrial application [5, 131, 134], long haul terrestrial application [17, 129], metro application [3, 16], access application [133], and submarine application [128, 135]. So in the near future the application of DWDM transmission system with coherent receiver will be widely spread and the deployment will increase and dominate the DWDM transmission market.

147 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

A slight trade-off of this design is that it has slightly larger optical loss than Methods 1 to 3 (but lower than Method 4 due to the removal of tunable filters). For a TA with 16 add/drop channels, the insertion loss in the channel separation unit for the proposed method is 14.4 dB, comparing to 8 dB, 6.5 dB, 9.5 dB, 17.4 dB respectively for the 4 existing methods. However this will have little effect to the system performance since this additional loss can be compensated by increasing the amplification level of the drop side amplifier, which already exists in the TA. Since in this method each coherent receiver accepts multiple WDM channels simultaneously without the wavelength selector in the TA, this receiver is also called a “filterless coherent receiver”. More detailed theoretical analysis of the filterless coherent detection technique in the TA without wavelength selector will be described below.

5.2.2 Theoretical Analysis In coherent detection system, the LO laser is tuned near the centre frequency of the channel of interest to demodulate the signal to electrical baseband via intradyne detection [137]. The total bandwidth downconverted is given by the photodetector bandwidth, which is typically in the tens of GHz. As the electrical baseband signal can be sampled, and digital signal processing (DSP) algorithms used to extract the channel of interest, coherent detection and DSP removes the need for optical filtering at the receiver. This enables WDM channels to be packed closer together, since digital filters can be designed with arbitrarily sharp cutoff. In coherent optical orthogonal frequency- division multiplexing (CO-OFDM), frequency sub-channels can in fact overlap while remaining digitally separable. The front-end of a coherent receiver is an optical hybrid combining the signal with the LO. Square-law photodetection follows. The output photocurrent consists of a desired signal-LO beating term corrupted by interferences arising from signal-signal and LO-LO beating. The interferences may be suppressed with balanced detection, where a pair of identical photodiodes is illuminated with the signal mixed with opposite phases of the LO [138]. Imprecision in the responsivities of the photodiodes or power imbalance in the optical hybrid reduces interference suppression. As the signal-signal interference scales with the number of WDM channels, the filterless receiver needs to be carefully designed to ensure the loss of performance is acceptable [139]. A key

148 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

parameter for such receiver is the signal-to-interference-and-noise ratio (SINR). A large SINR will ensure good receiver performance. In the following analysis, PDM-QPSK modulation format is considered, since it is the most common modulation format adopted for 100 Gbit/s transmission systems. In PDM-QPSK transmission, the four quadratures of a polarization-and-phase diversity coherent receiver have identical statistics. Hence, SINR can be inferred from analysing the powers of the signal, noise and interference terms in one of the quadrature. Consider the photocurrents in Figure 5.4:

2 =+11 ++ I++REjEIIin,,,, q LO q sh + th + (5.1) 22

2 =+++11 I−−Rj Ein,,,, q E LO q I sh − I th − (5.2) 22

* 1 22 IRREEΔ+−=+()Im () +() RRE +− −() + E + I ΔΔ ++ I I Δ (5.3) in,, q LO q2 LO , q in , q sh , th , adc , where Ein, q and ELOq, are the input and LO electric fields incident on the quadrature of interest; R+ and R− are the responsivities of the two photodiodes; and Ish,Δ , Ith,Δ and

Iadc,Δ are currents arising from shot noise, thermal noise and analogue-to-digital converter (ADC) noise; and IΔ is the output photocurrent.

U 

Ein, q I+ IΔ

V I− ELOq,

Figure 5.4 Canonical quadrature of a polarization- and phase-diversity coherent receiver. a. Signal

The first term of IΔ is the desired signal arising from beating between the input

− jft2π ()=+ () () k and LO electric fields. In WDM transmission, EtEtin,, q m qB Ete k , q is the km≠ summation of the electric fields of the channel of interest (channel m) and its ≈ neighbours. fkfksp is the frequency of the k-th channel with respect to the centre of

149 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

the band, and fsp is the channel spacing. Provided that care has been taken to avoid crosstalk between adjacent channels at the transmitter, the desired channel can be extracted with no distortion. Thus,

=+() * SRR+−Im () EEmq,, LOq (5.4)

The power of the signal current is:

2 1 2 SRRPP=+()+− (5.5) 2 ch,, q LO q

Assuming that all WDM channels are received with equal power Pch , and = = = Pch, qPN ch q and PLO, qPN LO q are the powers per quadrature ( Nq 4 ) at each channel for the signal and LO. b. Interference

The second term of IΔ is interference arising from LO-LO and signal-signal

2 beating. Provided the LO electric field is continuous-wave (CW), ELO, q is a DC signal and can be blocked with a DC-rejection filter having arbitrarily low stop-band bandwidth. Fluctuations of the LO electric field arising from relative intensity noise 1 (RIN) will result in an interference term ()R+−−ΔRP, whose power is: 2 LO

1 2 22≈−() () I RINRRP+− LO, q O RINfdf (5.6) 4 BWp where the integration is over the bandwidth of the photodetector, typically much greater than the RIN bandwidth. Ref. [140] showed that when a semiconductor laser is driven

FV−12 1 above saturation, GWO RIN() f df saturates at ~30 dB. Hence, one can assume: GW2π HXBWp

2621 2 − IRRP≈−()+−()210π × (5.7) RIN4 LO, q

For the signal-signal beating term, it can be assumed that Ein, q is an additive white Gaussian noise (AWGN) process whose total power NPch ch, q is evenly distributed

1 2 over a bandwidth of Nf. It can then be shown that ()RRE+−− has a triangular ch sp 2 in, q two-sided power spectral density (p.s.d.) of:

150 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

CS2 C S 1 2 Pch, q f Sf()=−() RR+−DT Nf D1 − T (5.8) IIsig sig DT chsp D T 4 EUfsp ENf ch sp U

Provided that Nch is large, the spectral triangle is essentially flat over the bandwidth Rs of the desired signal in (5.4). Thus, CS 1 2 R 22≈−() DTs IRRNPsig−+− sig ch ch, q DT (5.9) 4 EUfsp The total interference is given by:

22=+ 2 IIsig− sig I RIN (5.10)

< 3 Note that typically, PLO10 P ch , so even for filterless detection of a 100-channel

22>> WDM signal, IIsig− sig RIN . c. Noise

The total noise in IΔ is given by:

=+() * + ++ NRR+−Im () EEn,, q LO q I sh , Δ I th , Δ I adc , Δ (5.11)

The first term arises from amplified spontaneous emission (ASE) from optical amplifiers, and is additive white Gaussian noise (AWGN) with a two-sided p.s.d. of

P R =+1 ()2 ()= ch, q s SRRPPLOASE−+− nqLOq,,, where OSNR0.1 nm is the optical signal-to- 2 Pnq, 12.5GHz noise ratio (SNR) measured in a 0.1 nm optical bandwidth at 1550 nm. The second and third terms in (5.11) arise from shot noise of the photodetectors, and thermal noise of the receiver’s trans-impedance amplifier (TIA). These are AWGN

1 2kTamp with two-sided p.s.d. of SqRRPNP≈+()+−() + and Sf()= , IIth,,ΔΔ th LOq,, chchq IIth,,ΔΔ th 2 RL respectively; where q is the electronic charge, k is Boltzmann’s constant, Tamp is the effective temperature of the amplifier, and RL is the trans-impedance load. Finally, ADC noise encapsulates quantization effects, in addition to noise and nonlinear distortion of the ADC. ADC performance is usually quantified by the effective number of bits (ENOB) at a target dynamic range. Let the number of bits be B

Δ= B and the quantization step size be Vpp− adc 2 . The equivalent noise of the ADC referred to the photodiode output is given by:

151 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

Δ2 412()()−() σ 2 ≈ 22 BENOB (5.12) IIadc,,ΔΔ adc 2 RL Adding the noise terms together, we have: 2kT Δ2 2 =+11()2 +() +() + +amp +B−ENOB NRRPPRqRRPNPRR+−n,, q LO q s +− LO , q ch ,, q ch q s s 2 4 (5.13) 22 RL 3RL Typically, LO-ASE beat noise is much larger than the shot-noise and thermal- noise terms. d. Signal-to-Interference-and-Noise Ratio Using the signal, interference and noise variances derived in equations (5.5), (5.10) and (5.13), the SINR of the receiver is:

S 2 SINR = IN22+ PP = LO,, q ch q 2 CS CS2 12CSRR− R q 2kT Δ − PP+++++DT+−DTs NP2 () P NPRDTamp R 4BENOB LO,, q n q++DT ch ch , q LO , q ch ch , q s22 s 2 EURR+−EU fsp RR +− ()RR+−+ EU RL 3RL (5.14) If as expected, LO-ASE noise and signal-signal interference are the dominant terms, (5.14) can be simplified to: P ≈ ch, q SINR 2 1 CSRR− P2 CSR + +− ch, q DTs PNnq, DT ch DT 2 EURR+−+ P f LO, qEU sp (5.15) CS2 2 1 CSRR− P CSR ≈−SNRDT1 DT+− N ch, q DTs DT+ ch DT EU2PRRnq,,EU+− P LOqEU f sp

2 C R+ − R− S where CMRR = D T is the common-mode rejection ratio. It is observed that E R+ + R− U

interference scales linearly with CMRR and the number of WDM channels Nch , and inversely with the LO-to-per-channel-power ratio PLOP ch . The amount of performance degradation due to interference expressed in dB units is given by:

CS2 22 1 CSRR− PPCSR logDT 1−∝DT+−NNch,, qDTs ch q 10 DT+ chDT ch (5.16) EU2PRRnq,,,EU+− P LOqEU f sp P LOq In a special case where a single-ended detector is used instead of balanced detector, R− is 0, and thus CMRR = 1. In such case, the interference term is at its

152 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

maximum. In an ideal case, R+ and R− have the same value (completely balanced) and thus CMRR = 0, which bring the interference to the minimum.

5.2.3 Simulation Matlab simulations are performed to verify the power spectral density of signal- signal interference derived in (5.8). Figure 5.5 shows the simulation results obtained for the three systems with different number of channels, CMRR, LO-to-per-channel-power ratios, and channel frequency spacings. The blue curves show the simulated signal- signal interference, while the red curves show the theoretical spectrum. Good agreement between the two curves is observed.

.R.V: 1J$ -120 96×112 Gbit/s PDM-QPSK 1$J:CR1$J:CV: 1J$ 50 GHz channel spacing -140 (a) CMRR = −10 dB -160 P = 20 dBm (f) (dBm/Hz)(f) LO ^RIL+<_ yy Pch = −8 dBm S -180 ]VH `:CVJ%1 7 -4 -2 0 2 4 Freq. (Hz) 12 x 10

-120 22×112 Gbit/s PDM-QPSK -140 50 GHz channel spacing (b) -160 CMRR = −15 dB PLO = 14 dBm (f) (dBm/Hz)

^RIL+<_ -180 yy Pch = −8 dBm S ]VH `:CVJ%1 7 -1 -0.5 0 0.5 1 Freq. (Hz) 12 x 10

-120 35×112 Gbit/s PDM-QPSK 100 GHz channel spacing (c) -140 CMRR = −10 dB P = 14 dBm

(f) (dBm/Hz) LO ^RIL+<_

yy -160 Pch = 0 dBm S ]VH `:CVJ%1 7 -3 -2 -1 0 1 2 3 Freq. (Hz) 12 x 10

Figure 5.5 Spectrum of theoretical (red) and simulated (blue) signal-signal interference for three example systems.

The analytical results of Section 5.2.2 above are used to estimate the OSNR penalty of coherent WDM detection using the filterless architecture compared to the conventional architecture. This penalty is defined to be the excess OSNR needed to

153 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

achieve the same performance as single-( detection at a reference OSNR. Figure 5.6 shows OSNR penalty results of a 128 Gbit/s PDM-QPSK system where balanced detection was assumed with CMRR = −15 dB, which is the value provided in the specification of the commercial u2t BPDV2020R balanced receiver. The power difference between the LO and signal is varied by fixing the LO power at 12 dBm and setting the per-channel signal power at −10 dBm and −15 dBm respectively. Practical characteristics of the photodetector responsivity and ADC quantization noise are assumed. It is observed that OSNR penalty grows more rapidly with number of channels when using a higher received power per channel. This is an expected result from (5.15).

7 -15 dBm/ch 6 -10 dBm/ch

5

4

3

2 OSNR Penalty (dB) 1

0 0 20 40 60 80 100 No. of Channels

Figure 5.6 OSNR penalty for 50 GHz-spaced PDM-QPSK signals at different WDM channel numbers and different channel power levels.

In the next simulation, the signal power is fixed at −15 dBm per channel, while the CMRR is varied. The results are shown on Figure 5.7. It is observed that the OSNR penalty is disastrous for CMRR = 0 dB (single-ended detection), but remains at 1 dB even for the transmission of 96 WDM channels provided that the CMRR is greater than 15 dB. If the CMRR is improved to −20 dB, the OSNR penalty between 1 channel and 96 WCM channels is less than 0.5 dB. These results show that in order to achieve low OSNR penalty in the filterless coherent receiver with large number of WDM channels, balanced receiver with low

154 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

7 CMRR=0 dB 6 CMRR=-10 dB CMRR=-15 dB 5 CMRR=-20 dB 4

3

2 OSNR PenaltyOSNR (dB) 1

0 0 20 40 60 80 100 No. of Channels

Figure 5.7 OSNR penalty for 50 GHz-spaced PDM-QPSK signals at different WDM channel numbers and different CMRR settings.

CMRR (<−15 dB) should be used. Most of the commercial high-speed balanced photodetectors meet or exceed such CMRR level. The simulation results also show that larger power difference between the LO and per-channel signal also reduces the OSNR penalty, however the signal power should be sufficiently high to minimize the quantization error and error due to shot noise, and the total optical power at the receiver (combination of LO and signal power from all incoming channels) cannot exceed the saturation limit of the photo receivers. Overall, the simulation demonstrates that with product grade balanced detectors and appropriate LO and signal power settings, the proposed TA without wavelength selector causes little penalty to the receiver performance and thus is practical for application in the CL&DL ROADM nodes.

5.2.4 Design Software Based on the analytical model in Section 5.2.2, a design software is written using LabVIEW. The main processing is performed using Math Script from National Instrument. It has a graphical user interface (GUI) that allows the user to set the signal conditions (including baud rate, channel spacing, LO power, per-channel power) and the receiver conditions (including photodetector type, CMRR, insertion loss, amplifier type, ADC bits and ENOB). The software will produce the OSNR vs. BER curve for a

155 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

filterless coherent receiver under such conditions. It also allows the simultaneous analysis of multiple channel counts to compare the OSNR penalties under different WDM settings. Figure 5.8(a) is an example of the GUI. In this example, the log-scaled BER curves at OSNR from 5 dB to 30 dB are shown for four different channel count settings (1 channel, 16 channels, 40 channels and 96 channels). Under this condition (CMRR = -10 dB, 18 dB power difference between LO and per-channel signal), there is about 0.4 dB OSNR penalty at BER = 1×10-3 level when the channel number increases from 1 to 16. This figure increases to about ~3 dB when the number of channels increases to 96. In another example shown on Figure 5.8(b), the CMRR is improved to -15 dBm, and the power difference between LO and per-channel signal is increased to 20 dB. Under this condition, the OSNR penalty at BER = 1×10-3 level between single channel and 96 channel is only ~0.5 dB.

(a)

(b) Figure 5.8 Examples of the filterless ROADM receiver design software GUI.

156 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

5.3 Experimental Demonstration of Transponder Aggregator without Wavelength Selector 5.3.1 Experiment Setup To verify the theoretical analysis results, a 96-channel DWDM transmission system with 50 GHz channel spacing is set up. 96 is currently the highest channel count used in commercial C-band 50 GHz-spaced DWDM systems. The 96 channels here are from 191.50 THz to 196.25 THz, which correspond to a wavelength range of 1527.61 nm to 1565.50 nm. 

Figure 5.9 Experimental setup of the 96-channel filterless coherent receiver experiment.

The transmission testbed setup is shown on Figure 5.9. The WDM transmitter consists of two sets of 100 GHz-spaced 112 Gbit/s PDM-NRZ-QPSK channels, one is the Odd Channels and the other is the Even Channels. In each set, 48 Alcatel A1905LMI DFB lasers with 100 GHz spacing are combined using an athermal 100 GHz PM optical multiplexer from Agilecom. The multiplexed signal is modulated by a

Fujitsu FTM7961EX 40 Gbit/s LiNbO3 I-Q modulator operating at 28 Gbit/s rate. The 28 Gbit/s I and Q modulation data are obtained by electrically time multiplexing four PRBS data stream from an Agilent 7083C 12.5 GHz pulse pattern generator (PPG)

157 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

driven at 7 GHz clock rate by Agilent N5183A MXG analogue signal generator. A Sage Model 6705K DC to 26.5 GHz tunable electrical delay is placed between the 28 Gbit/s I and the Q data signals to de-correlate them. Each 7 Gbit/s PRBS signal has a pattern length of 211-1, therefore the combined 28 GBaud (56 Gbit/s) signal has a pattern length of 213-4. This 56 Gbit/s NRZ-QPSK signal is then passed through an optical polarization multiplexing stage. Here the signal is split into two parts using a polarization maintaining (PM) optical splitter from Opto-Link. A General Photonics VDL-001 PM manually variable optical delay line is placed in one path to de-correlate the signal from the other path. In the other path, an Eigenlight Series 400 PM variable optical attenuator with monitor is used to balance the power between the two paths. These two 56 Gbit/s optical signals are then combined using an Opto-Link OLCS polarization beam combiner to form a 112 Gbit/s PDM-NRZ-QPSK signal. This signal is amplified and sent to a 50 GHz/100 GHz interleaver from Optoplex to combine with the other set of 48 channel signals. Among the 96 channels, the 193.25 THz channel is used as the test channel. Its wavelength is 1551.32 nm. Because the DFB laser has a wide linewidth of ~2 MHz which causes large phase noise, the DFB laser for this test channel is replaced with a tunable external cavity laser (ECL) from Emcore for BER measurement. The linewidth of this Emcore TTX19900 ECL laser is ~100 kHz. The 96-channel 112 Gbit/s PDM-NRZ-QPSK test signal is then transmitted over three spans of standard SMF-28 single mode fibre. The span length is 80 km. An Amonics AEDFA-DWDM-23-B EDFA is used at each span to compensate for the fibre attenuation. This transmission link is not dispersion managed, which means that there is no dispersion compensation fibre/module (DCF/DCM) or tunable optical dispersion compensator (TODC) inserted in the link. Before the receiver, noise is added to the signal to allow the adjustment of OSNR for the received signal. This noise is generated by an ASE source from Agilent, and is then passed through a tunable optical filter with the passband of ~300 GHz. This filter is to reduce the amplifier gain tilt variation during the OSNR adjustment, since the Amonics EDFAs used in this experiment do not have good gain flattening capability. 300 GHz is sufficiently wide to cover the channel-of-interest and several neighbouring

158 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

channels. In the experiment, a CoAdna DCEB series dynamic gain equalizer (DGE) is used as the tunable optical filter. After adding the noise, the signal is amplified again and then passed through a 50 GHz WSS from Finisar. This WSS acts as the drop portion of a simple transponder aggregator without wavelength selector, as illustrated in Figure 5.1(a) and Figure 5.3. It selects the combination of drop channels and passes them to each transponder. At the digital coherent receiver, the signal from the WSS output (which contains multiple DWDM channels) is attenuated using an Eigenlight Series 400 manual VOA to obtain a specified per-channel power level. Another Emcore TTX19900 tunable ECL is used as the LO and is tuned to the channel of interest (193.25 THz). The LO signal is amplified using an Amonics AEDFA-PM-23-B PM-EDFA to reach its specified power level. The received WDM signal and the LO signal then enter a polarization diverse coherent mixer (also called a 90° optical hybrid) with four sets of balanced outputs. These four sets of data (X0°, X90°, Y0°, Y90°) are received by four u2t BPDV2020R balanced receivers respectively. The electrical signals from the receiver outputs are sampled and digitised using a Tektronix DSA 71604 4-channel Digital Serial Analyser (real time digital oscilloscope) with a 50 GSa/s sampling rate and 16 GHz analogue bandwidth. The captured data are then post-processed offline using a computer to calculate the BER and the Q-factor. The full description for the DSP algorithms used for post-processing is discussed in more detail in [17].

5.3.2 CMRR Measurement and Adjustment Technique The theoretical analysis above shows that two important parameters that affect the SINR in the filterless coherent receiver are the CMRR of the receiver and the difference between the LO power and per-channel signal power. Having small CMRR (as close to 0 as possible) and setting LO power much higher than the signal power will improve the SINR and therefore lead to less OSNR penalty between single channel and multiple channel receiving. Since in commercial transmission system the per-channel signal power needs to follow design engineering rules and cannot be easily modified, and the LO power cannot be set too high in order not to saturate the photodetector, the system improvement through optimising the power parameter is less feasible. Therefore reducing CMRR becomes a key to the system optimisation.

159 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

Before investigating how to reduce CMRR, two challenges need to be addressed. Firstly, to experimentally verify the effect of the CMRR of the balanced receiver, it is desirable to obtain multiple balanced receivers with different CMRR characteristics. However this is not practical due to the high cost and the unavailability of balanced receivers with specified CMRR figure. Secondly, the CMRR value of a balanced receiver is usually not provided by the manufacturer. Only the specified value is available (for example, u2t specified a maximum level of -15 dB CMRR in their balanced receivers). However it is important to be able to measure the CMRR accurately so that one can learn the correct relationship between the CMRR and the receiver performance. To meet these challenges, a technique to adjust and measure the CMRR of a balanced photodetector is proposed here. Firstly, the LO laser is turned off, so only the WDM signal reaches the coherent mixer. Even though there is no restriction in the number of WDM channels, the maximum number is recommended since it will increase the power level and minimize the effect of noise. Therefore in this experiment all 96 channels are sent to the coherent mixer. The average optical power at the eight output ports of the coherent mixer is measured using an optical power meter. It is denoted as P. Then one end of the balanced photodetector is disconnected. The root mean square (RMS) value of the acquired signal’s voltage is estimated from the acquired digital trace on the real-time oscilloscope. It is denoted as Vrms. From this value the V RMS current value can be calculated using I = RMS × 2 since the receiver has a RMS 50Ω matching 50Ω load. Assuming a linear relationship at the scope, the mean responsivity I of this photodiode can be calculated as R = RMS . Then the port is reconnected and 1 P other port is disconnected to measure the mean responsivity of the second photodetector

R2. The R1 and R2 are the effective values of mean R+ and mean R− described in the previous section and not the absolute photodetector response since the signal RMS is detected after the DC block and digitiser bandwidth filtering. This procedure is repeated for all four channels. In the next step, both ends of the balanced photodetector are connected and the RMS value of the acquired signal is measured for all channels. These values are used to

160 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

calculate the mean R+ − R− for each channel. Unlike the mean R+ + R− value which is calculated from adding the mean R+ and the mean R− values, the mean R+ − R− value is directly measured. One reasons is that this value is much smaller, calculating it from the mean R+ and the mean R− values will introduce more noise. Moreover, the measurement above is a quick scalar measurement, however in fact the R+ and the R− values are dependent on signal frequencies and signal timing offset. In other words, the responsivity is a complex function of frequency. Therefore one cannot treat them as scalar and simply do a direct subtraction to obtain the R+ − R− value. Frequency domain measurement of the CMRR will be described later. Using these figures, the mean CMRR can be calculated using the equation

2 C R+ − R− S CMRR = D T . E R+ + R− U The mean CMRR values of the set of four u2t BPDV2020R balanced receivers are measured using this method. The measured responsivity data and the calculated CMRR are shown in Table 5.1. These values show that the balance performance of these units exceed the specification level of -15 dB.

Table 5.1 CMRR calculation for four u2t balanced receivers. Channel 1 Channel 2 Channel 3 Channel 4

Mean R+ (a.u.) 0.0948 0.0901 0.0914 0.0886

Mean R− (a.u.) 0.0978 0.0923 0.0966 0.0966

Mean R+ − R− (a.u.) 0.0097 0.0074 0.0137 0.0115 Calculated CMRR (dB) -25.95 -27.84 -22.79 -24.13

To obtain different CMRR settings, one end of the balanced photodetector is manually adjusted to impose additional attenuation and the new mean R+ − R− value is measured. By visually observing the signal amplitude range on the real-time digital oscilloscope, the target R+ − R− value can be reached approximately. The adjusted

CMRR can then be calculated using the same equation with the new R+ − R− value.

The R+ and R− values remain unchanged. Table 5.2 shows the example of the same four u2t balanced receivers after some manual CMRR adjustment.

161 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

Table 5.2 CMRR calculation for four u2t balanced receivers after adjustment. Channel 1 Channel 2 Channel 3 Channel 4

Mean R+ (a.u.) 0.0948 0.0901 0.0914 0.0886

Mean R− (a.u.) 0.0978 0.0923 0.0966 0.0966

Mean R+ − R− (a.u.) 0.0302 0.0269 0.0251 0.0267 Calculated CMRR (dB) -16.09 -16.63 -17.50 -16.84

 0

-5

-10

-15

-20 CMRR (dB)CMRR -25 Channel 1 Channel 2 -30 Channel 3 Channel 4 -35 0 5 10 15 20 25 Frequency (GHz) (a) CMRR=-11.2dB 0

-5

-10

-15

-20 CMRR (dB)CMRR -25 Channel 1 Channel 2 -30 Channel 3 Channel 4 -35 0 5 10 15 20 25 Frequency (GHz) (b) CMRR=-15.3dB

0 Channel 1 -5 Channel 2 Channel 3 -10 Channel 4 -15

-20 CMRR (dB)CMRR -25

-30

-35 0 5 10 15 20 25 Frequency (GHz) (c) CMRR=-24.9dB

Figure 5.10 Example of CMRR frequency response under different adjustments.

162 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

In the above measurements, the mean (RMS) values of the receiver response are used. To better understand the response of the balanced receiver at different frequencies, a period of data is captured from the real-time digital oscilloscope for each photodetector setting, and the frequency response is analysed. Figure 5.10 shows three examples of the frequency response of adjusted CMRR. Example 3 (Figure 5.10(c)) is under the optimum setting (no manual attenuation). Since the Tektronix real time digital oscilloscope used in the measurement has a bandwidth limit of 16 GHz, the average CMRR is calculated from DC to 16 GHz. The average CMRR values for these three examples are 11.2 dB, 15.3 dB and 24.9 dB respectively. It can be observed that the typical CMRR response is flat across the DC to 16 GHz spectrum. However under the optimum CMRR condition, Channel 3 has higher (worse) CMRR at higher frequencies. This indicates some imperfection in the hardware. These three CMRR conditions are used in the filterless coherent receiver experiment below.

5.3.3 Experiment Results and Analysis In the experiment, the number of WDM channels reaching the filterless coherent receiver is selected using a Finisar 50 GHz WSS. Even though these WDM channels do not need to be adjacent, continuous bands of channels are used in the experiment, with the centre at the channel of interest (193.25 THz). Figure 5.11 shows the spectra of the received WDM signals when selecting 1 channel, 8 channels, 24 channel, 48 channels, 72 channels and all 96 channels respectively. The spectra of the LO laser is also shown. Ando AQ6317B OSA is used in this measurement. The resolution bandwidth is 0.1 nm. It can be noted that some channels have lower amplitude than the rest, and this non- uniformity varies over time. This is due to the imperfect alignment of the PM fibre axis in certain channels of the PM AWG multiplexer used at the transmitter end. This misalignment causes the fluctuation of power over time. This issue does not occur on the channels close to the channel of interest, therefore it should have little impact to the experiment result. The CMRR figures of the photodetectors are adjusted to the three values described above (-11.2 dB, -15.3 dB, and -24.9 dB), and measured the BER at OSNR range from about 14 dB to about 18 dB. This is the range where the BER performance is between 10-2 and 10-4. In this experiment, the signal power is set to about -0.5 dBm per channel, and the LO laser power is set to about 19.5 dBm. Therefore the power

163 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

difference between the LO and per-channel signal is 20 dB. This is a typical value in 100G DWDM network with standard single mode fibre. Three channel settings are measured: 1 channel, 24 channels, and 96 channels. They represent the conventional filtered receiver, filterless receiver with 25% dropping ratio, and filterless receiver with maximum (100%) drop ratio.

 -10 -10 -15 -15

-20 -20

-25 -25

-30 -30

-35 -35 Power (dBm) Power (dBm) -40 -40

-45 -45

-50 -50 1525 1535 1545 1555 1565 1525 1535 1545 1555 1565 Wavelength (nm) Wavelength (nm) (a) 1 channel (b) 8 channels

-10 -10

-15 -15

-20 -20

-25 -25

-30 -30

-35 -35 Power (dBm) Power (dBm) -40 -40

-45 -45

-50 -50 1525 1535 1545 1555 1565 1525 1535 1545 1555 1565 Wavelength (nm) Wavelength (nm) (c) 24 channels (d) 48 channels

-10 -10

-15 -15

-20 -20

-25 -25

-30 -30

-35 -35 Power (dBm) Power (dBm) -40 -40

-45 -45

-50 -50 1525 1535 1545 1555 1565 1525 1535 1545 1555 1565 Wavelength (nm) Wavelength (nm) (e) 72 channels (f) 96 channels

10

5

0

-5

-10

-15 Power (dBm) Power -20

-25

-30 1525 1535 1545 1555 1565 Wavelength (nm) (g) LO

Figure 5.11 Spectra of received WDM channels at different numbers and the LO.

164 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

Figure 5.12 shows the BER vs. OSNR plots under three different CMRR conditions. Since theoretically there is a linear behaviour between the OSNR and the BER plotted in logarithm scale at this BER region, linear fit is performed to the measured data. From the fitted curves the OSNR value for 10-3 BER are located for each channel setting. The OSNR penalties for higher channel settings are then calculated.

 -1 10 1-ch 24-ch -2 10 96-ch

-3 10 BER

-4 10

14 16 18 20 OSNR (dB) (a) CMRR= –11.2dB -1 10 1-ch 24-ch -2 10 96-ch

-3 10 BER

-4 10

14 16 18 20 OSNR (dB) (b) CMRR= –15.3dB -1 10 1-ch 24-ch -2 10 96-ch

-3 10 BER

-4 10

14 16 18 20 OSNR (dB) (c) CMRR= –24.9dB Figure 5.12 Measured BER vs. OSNR for different CMRR conditions.

165 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

Figure 5.13 shows the OSNR penalty under these three CMRR conditions. The results show that the filterless coherent receiver only experienced 0.32 dB penalty between single filtered channel and all 96 channel settings under the good CMRR condition of -24.9 dB. This value is well within the system tolerance. Even at border line CMRR condition (-15.3 dB), the penalty between single channel and 96 channels is only 0.71 dB. This is also acceptable. At poor CMRR condition (-11.2 dB), the penalty increases to 2.4 dB and is cannot be used in the existing transmission system. However, if the number of WDM channels at the receiver reduces to 24, the penalty is 0.91 dB and is within 1 dB tolerance range.

 2.5 CMRR=CMRR=-24.9dB –24.9dB CMRR=CMRR=-15.3dB –15.3dB 2 CMRR=CMRR=-11.2dB –11.2dB

1.5

1

OSNR Penalty (dB) 0.5

0 0 20 40 60 80 100 No. of Channels Figure 5.13 Measured (symbols) and theoretical (line) OSNR penalties for different CMRR conditions.

These results are compared with the theoretical values calculated based on the model described in Section 5.2.2 (the solid curves in Figure 5.13). Here the actual settings of the coherent receiver are used. For example, the bandwidth of the digitiser is set to 16 GHz, the ADC bit is 8, and the ENOB is 4.8. Good agreement between the theoretical expected value and the experiment results has been observed. The measured data have slightly larger OSNR penalties because the actual experimental transmission conditions are not ideal. The good agreement also shows that our CMRR measurement method has good accuracy. In Figure 5.14, the BER vs. OSNR data are plotted in a different way to compare the effect of CMRR under each channel setting. At the single channel setting, there is

166 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

only about 0.2 dB OSNR penalty between balanced receivers with good and poor CMRR characteristics. For the 24-channel system, the OSNR penalty when CMRR is - 15 dB or below is also small, however poor CMRR level of -11.2 dB leads to about 1 dB penalty. For the 96-channel system, the OSNR penalty increase more significantly with the deterioration of CMRR. It reaches about 2.5 dB between good CMRR and poor CMRR conditions.

 -1 10 CMRR=-24.9dBCMRR= –24.9dB CMRR=-15.3dBCMRR= –15.3dB -2 10 CMRR=-11.2dBCMRR= –11.2dB

-3 10 BER

-4 10

14 16 18 20 OSNR (dB) (a) 1 channel

-1 10 CMRR=-24.9dBCMRR= –24.9dB CMRR=-15.3dBCMRR= –15.3dB -2 10 CMRR=-11.2dBCMRR= –11.2dB

-3 10 BER

-4 10

14 16 18 20 OSNR (dB) (b) 24 channels

-1 10 CMRR=-24.9dBCMRR= –24.9dB CMRR=-15.3dBCMRR= –15.3dB -2 10 CMRR=-11.2dBCMRR= –11.2dB

-3 10 BER

-4 10

14 16 18 20 OSNR (dB) (c) 96 channels Figure 5.14 Measured BER vs. OSNR for different channel settings.

167 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

Finally, the performance with single-ended receiver is tested. Here the single- ended receiver is set up by simply disconnecting one end of each balanced photodetector. Three channel setting are studied: single channel, 8 channels, and 16 channels. The power levels of the LO and per-channel signal remain the same. About 2.5 dB penalty is observed at the 8-channel system, and the penalty increases to about 6.5 dB at the 16-channel system (Figure 5.15). This is similar to the results reported in [139]. It again demonstrates the effect of CMRR (which is 1 in this case) in the filterless coherent receiver scheme.

 7

6 SEDSED BDBD (CMRR=-11.2dB) (CMRR= –24.9dB) 5 BDBD (CMRR=-15.3dB) (CMRR= –15.3dB) BDBD (CMRR=-24.9dB) (CMRR= –11.2dB) 4

3

2 OSNR Penalty (dB) Penalty OSNR

1

0 0 20406080100 No. of Channels Figure 5.15 OSNR penalties for single-ended detector and balanced detectors with different CMRR conditions.

5.4 Conclusions In this chapter, I proposed, theoretically analysed and experimentally demonstrated a new design for transponder aggregator in a colourless, directionless and contentionless ROADM node. It does not require wavelength selector such as WSS, demultiplexer or tunable filter array. Instead it uses the local oscillator in the colourless coherent transponder to select the target channel from all the received WDM channels. Theoretical analysis and experiments demonstrate that this filterless transponder aggregator can achieve less than 0.5 dB OSNR penalty between receiving a single channel and receiving 96 DWDM channels, provided that balanced receivers have

168 )%, ;)%, ; ,+#&%, '',%')#, $ ,+#&%, '',%')#, $))#.) 1%%') %%!)#,#.) 1%%') %%!)#,#.) 1%%') %%!)#,

reasonable CMRR characteristic and there is sufficient power difference between the local oscillator and the per-channel signal. A method to measured and adjust CMRR in the balanced photodetector is also proposed and demonstrated. This filterless transponder aggregator has the potential to significantly reduce the cost, size, power consumption in the next generation ROADM node and improve reliability and performance. Due to commercial sensitivity of this technology, this work was not published, but two patent applications have been filed [141, 142].

169 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

)%, )%, <<<         

 6 7   

     

PMD is an important transmission impairment which distorts the signal quality and limits the transmission distance. Due to the random time-varying nature of PMD, to realise its compensation is very challenging. Existing PMD compensation schemes either require feedback loop with large number of iteration and random guessing or require feedforward loop with complex SOP calculation. In this chapter, I proposed and demonstrated a novel PMD compensation technique. It can measure and compensate PMD at all orders in real time over a wide range of spectrum. The main concept is to restore the pulse by restoring the spectrum. The technique includes two steps: measuring the PMD via spectral interference and phase retrieval, and compensating for the PMD using pulse shaping. The concept and theory for these two steps are discussed, and the performances are verified using experiment and simulation. The results show that this technique can be used to construct a PMD compensator subsystem to compensate PMD at all orders in real time, and therefore is suitable for application in WDM systems.

6.1 Background 6.1.1 PMD Issue in DWDM Transmission Polarization mode dispersion (PMD) is an optical phenomenon that affects signal quality during optical transmission. When propagating down an optical fibre, the optical signal undergoes changes in polarization due to uncontrollable physical changes in the fibre. Because light travels at slightly different velocities for different polarizations, the pulse shape is broadened over time and distorted. This pulse broadening phenomena is referred to as the PMD. The first order PMD can be

170 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

represented as a differential group delay (DGD) between two principal states of polarization (PSP) [143], which is the propagation time between the two orthogonal axes (planes) for the polarization. PMD is usually caused by environmental conditions such as physical stress, temperature variation and fibre imperfections. It is dynamic and varies over time. The individual factors that cause PMD cannot be measured or even observed in isolation, the phenomenon must be viewed as a constantly changing, unstable stochastic process. There are no known practical ways of eliminating the effect of this random and time varying process entirely. At lower bit rate transmission, PMD is not an important factor and is often neglected. However as the transport speed increases above 10 Gbit/s, such as at 40 Gbit/s and 100 Gbit/s, the impairment from PMD becomes a serious issue due to the short bit period and it limits the transmission distance. At these high bit rates, higher order PMDs (such as the second-order PMD, which is the variation of first order PMD with wavelength/frequency) also become important factors in system degradation. Furthermore, a large amount of the installed fibres exhibit PMD values that are several times that of current state-of-the-art fibres, making the PMD one of the most difficult issues in implementing next generation high bit rate transmission systems [144]. Besides transmission fibre, passive components in the signal path (such as filter, multiplexer, coupler) can also cause first order or higher order PMD. As the demand of network traffic bandwidth grows and DWDM network transmission bit rate increases, compensation for first order and higher order PMD has attracted strong research interests.

6.1.2 Current PMD Measurement and Compensation Methods In recent years a variety of schemes have been proposed to counter the effect of PMD, including electronic and optical dispersion compensation techniques. Electronic dispersion compensation (EDC) can be used successfully to compensate for pulse distortion caused by many factors, including PMD. However due to the electronic processing speed limit, EDC cannot handle high transmission rates where the bandwidth becomes large. Comparing to electronic dispersion compensation techniques, optical compensation is more suitable for application in high transmission rate system.

171 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

Therefore various optical compensation techniques have been investigated and employed. Some compensating schemes are based on feedback loops and complex algorithms to optimise the control parameters [145, 146]. These schemes have the advantage that they do not require the knowledge of the PMD parameters. Instead, they are based on monitoring the degree of polarization (DOP) and changing the state of polarization through a feedback loop in order to minimize an error signal. However, they are cumbersome and less practical for fast real-time compensation, because their algorithms rely on random guesses and iterative loops. More promising schemes are based on feedforward compensators, because they are faster and easier to implement. However, they require the knowledge of the fibre’s PMD parameters at any given time. For example, one proposed scheme measures the state of polarization (SOP) by passing the signal through an optical filer placed before a polarimeter, scanning the optical filter through the spectrum of interest and then using dispersive elements to compensate for PMD [147]. Measuring SOP under such scheme can prove to be difficult and time consuming, making it a challenge to use these devices for live compensation. In this chapter, a new feedforward optical method to achieve live measurement of the polarization dispersion using spectral interference and real-time PMD compensation using pulse shaping is proposed and demonstrated.

6.2 Feedforward Real-Time All-Order PMD Measurement and Compensation Technique 6.2.1 PMD Model and Theory Although single-mode fibres are theoretically designed to support only one mode of light propagation, the anisotropy in the geometry of real optical fibres and/or the environmental factors (e.g., the stress) results in two modes of propagation distinguished by their polarization. Because of the optical birefringence, the two modes travel with different group velocities. The random change of the birefringence leads to a random coupling between the modes [148]. This is the basis of PMD, which results in the distortion (i.e. broadening) of the input pulse and limits the transmission capacity of the optical fibre.

172 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

It is useful to consider the optical fibre as a series of successive optical waveplates, with their principal axis rotated one from another, as shown in Figure 6.1. At any given moment the waveplates have random orientation. Furthermore, they change their orientation on sub-millisecond time scale, mostly due to environmental changes (temperature, stress, vibrations, etc.). Each waveplate is characterized by a differential time delay between its fast and slow axis.  ω ω E0 ( ) E( )

Figure 6.1 The random birefringent fibre as a stack of randomly oriented waveplates.

The x and y axes are defined by the linearly polarized (along x, for example) input beam. After propagating through the waveplate system, the output electric field will be given by   ()ω = ()ω []()ω + ()ω E E0 xa yb (6.1) where a(2) and b(2) are complex coefficients which indicate how much phase has been acquired along the respective axis. The a and b coefficients are wavelength dependent, and they also vary in time in a random fashion. Because of the birefringence, the phase changes are asymmetric, leading to the pulse distortion and causing signal loss due to PMD. In this model, if we consider the x and y axes independently, the phase accumulated in each of them is independent of the other, and the spectral and/or power measurements along any or both of these axes will not indicate the pulse distortion information. If, however, the electric field along the x axis interferes with the electric field along the y axis, the phase difference between the two electric fields will be easily seen in the interference pattern. This spectral interference (i.e., the spectrum of the interference term) will provide the spectrum of the phase difference between the two distinct optical paths. We can define two optical paths A and B as two orthogonally polarized states travelling together through the fibre. They form a Mach-Zehnder interferometer, as depicted in Figure 6.2. These paths are brought together after propagation by a polarizer which defines the end of the interferometer. Both A and B paths experience phase accumulation for the x and y polarizations. An amplitude

173 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

measurement of the interference term will deliver the phase difference, hence the birefringence, between x and y for every wavelength within the pulse’s bandwidth.

Phase modulator

ϕy(ω) E E E(B) E(B) Light source

C     iϕ ωS DE A + E B e y T 2 E y y U (A) (A) E E Optical Spectrum   ϕ ω   Analyzer ( A i x + B ) Phase modulator Ex e Ex 2 ϕx(ω)

Figure 6.2 Mach-Zehnder interferometer where the phases along x and y lead to spectral modulation due to interference.

For two orthogonally polarized states, x and y, travelling together through the fibre via two optical paths A and B, the dispersive difference in phase leads to a modulation in spectra as follows:

= ( A) + (B) + ( A) (B) φ − ( A) (B) φ S S S 2 S x S x cos( x ) 2 S y S y cos( y ) (6.2) where Sx,y(A,B) are the power densities for the beams propagating along the A and B optical paths, for the x and y polarizations, and φx,y are the corresponding phases. By appropriately choosing the orthogonal axis, one can enhance the interference term in Equation (6.2). This can be done, by example, by placing a rotating polarizer before the optical spectrum analyser. Figure 6.3 shows an example of spectral interference from Ref. [149], obtained with a setup similar with Figure 6.2. The measured power spectrum shown by the smooth dotted black line is given by S = S ( A) + S (B) , which is the case when no phase has been introduced in either arm of the interferometer. As soon as the phase modulators introduce some phase difference between the arms of the interferometer, the spectrum analyser sees a spectrum which is modulated in frequency (solid blue line). The modulation is due to the interference term in Equation (6.2), which is calculated, and plotted as the red dotted line in Figure 6.3.

174 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

2500

2000

1500 S(nW)

1000

500 .

0 420 430 440 450 460 470 480 490 500 λ(nm) Figure 6.3 Spectral interference obtained in Ref. [149] from a Mach-Zehnder interferometer.

One can easily obtain from the spectral interference a spectrum of polarization changes. An example of inferring PMD from spectral interference is Ref. [150], where the PMD is related to the transmission spectrum measured through an analyser. As described in Ref. [151], there is a simple relationship between the spectral interference and the polarization change P, given by:

( A) (B) ()φ ()ω − ( A) (B) ()φ ()ω 2 S x S x cos x S y S y cos y P = (6.3) S Hence, one can infer the polarization change as a function of frequency simply by measuring the spectral interference between the two paths, A and B. In other words, by defining two orthogonal axes for the input and recording the spectrum of the output through a properly placed polarizer, we can measure the spectrum of the phase difference between the two axes, i.e., the birefringence spectrum. This way, the spectral interference will provide a measure of the polarization change as a function of wavelength. This is in fact all the information needed to be able to compensate for the PMD. Therefore by imposing on the optical pulses a phase opposite to the one measured by spectral interference, we can compensate for the phase accumulated in the fibre and recover the temporal shape which they had before entering the fibre. This can be observed as recovering the spectrum. When the accumulated phase variation is cancelled, the PMD in the transmission fibre is also compensated.

175 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

6.2.2 PMD Measurement and Compensation System Based on the above model and analysis, a novel PMD measurement and compensation scheme is developed [25]. The implementation of this scheme is schematically shown in Figure 6.4.

Fibre link Tap PMD compensator Pulse Tx Polarizer Rx shaper

Driver Polarizer

Spectrum Spectrum Phase measurement comparison extraction

Figure 6.4 Schematic of all-order PMD compensation scheme.

Before entering the PMD compensator, the pulses acquire unknown phases at each frequency while propagating along the fibre link, which leads to pulse distortion. At the PMD compensator, an optical splitter is used to tap out a small portion of the distorted input signal. This signal is passed through a polarizer to obtain the spectral interference as described in the theoretical model above. This polarizer is rotated to maximize the interference. The interference pattern is recorded through a spectrum measurement device, such as an OSA. It is then compared with the ideal spectrum with no PMD. The comparison result is transformed into a phase spectrum, which is then imposed with an opposite sign onto the signals by the pulse shaper to recover the spectrum. As shown, this is a single step feedforward scheme. It does not require multiple iterations to obtain the polarization state information, as required in the feedback schemes. The procedure is deterministic and does not require random guessing. Therefore it is fast and can perform the PMD compensation in real-time fashion. Also, while knowing the full Stokes parameters certainly allows for measuring the DGD as a measure of PMD [152], in this scheme the only information needed in order to be able to compensate for the PMD is how much phase has been accumulated at each

176 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

frequency. Therefore it does not require complex SOP computation process. In fact, simple analogue system is sufficient for the PMD compensation. Another important benefit of this scheme is that it can compensate PMD in all orders. The spectral interference data contain the information for PMD at all orders. Through appropriate phase retrieval technique, the PMD at all orders can be obtained. Since this scheme can compensate for different amount of PMD at different frequencies, it is suitable for WDM system application where the different optical channels might suffer from different amount of PMD due to different transmission paths between the source and destination. There are two main steps in the proposed scheme: (1) PMD measurement through phase retrieval, and (2) PMD compensation through pulse shaping. In the following sections, these two steps will be discussed separately in more details.

6.2.3 Step 1: PMD Measurement by All-Order Phase Retrieval In the case of linear phase changes, the PMD is exactly the DGD between the slow and the fast axis, and can be obtained directly from measuring the periodicity of the spectral interference pattern. This is the case of short-length regime considered in Ref. [150], when the PMD is given by: Δτ = 2πN / Δω (6.4) where N is the number of fringes measured over a spectral range Δω. In the general case when the phase changes are not linear in frequency, one can estimate quite accurately the PMD by extracting an average DGD from looking at the extrema and zero-level crossings of a spectral interference pattern [150]. While this method gives a good average estimate of PMD, it is not sufficient for the real-time compensation of PMD. Instead, a real-time measurement of the phase spectrum is necessary [153]. As mentioned above, the key issue in resolving the PMD problem using the proposed method is the retrieval of the phase spectrum φ(ω) from spectral interference signal s(ω) = S(ω)cos[ϕ(ω)], where S(ω) is the envelope function, and φ(ω) is the unwrapped phase function known with the ambiguity 2kπ (where k is an integer). Common approaches involve demodulation methods based on the Fourier transform [154] and the wavelet-transform [155]. Despite their usefulness for many applications (e.g., see [156]), potential drawbacks revolve around the fixed nature of the basis

177 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

functions which do not necessarily match the variations of the signals that naturally occur in non-stationary processes. Alternatively, within the limits defined by Bedrosian [157] and Nuttall [158] theorems, the Hilbert transform (HT) has found a number of useful applications to the problem of phase recovery. Motivated by the need to describe noisy non-stationary signals using adaptive data-dependent basis functions, Huang et al. [159] have developed a local and fully data-driven technique. Paired with the Hilbert transform, the Hilbert-Huang transform is composed of two main algorithms for filtering and analysing non-stationary multi- component signals. This algorithm is used here to process the spectral interference signal and phase retrieval [160]. The Hilbert-Huang transform employs an adaptive technique which decomposes the signal into a finite and often small number of intrinsic mode functions (IMFs) that have well-defined instantaneous attributes. A signal must satisfy two criteria to be an IMF: (1) the number of extrema and the number of zero crossings are either equal or differ at most by one; and (2) the mean of its upper and lower envelopes equals zero. The decomposition of the signal into IMFs, called sifting, is performed as follows: 1. Identify all the local extrema of the original signal;

2. Construct the lower (xL) and upper (xU) envelopes of the signals by interpolating the local maxima and local minima by a cubic spline method;

3. Subtract the mean, m1(2)=(xL(2)+ xU(2))/2, from the original signal to produce ω = ω − ω the first IMF (IMF1) component, i.e., h1 ( ). s( ) m1 ( ) ; 4. Calculate the first residual component by subtracting IMF1 from the original ω = ω − ω signal, r1( ) s( ) h1( ). ω The residue r1( ) is treated as a new signal subject to the sifting process described above, yielding the second IMF (IMF2). The sifting process is repeated up to k times, until a criterion for stopping is satisfied. The criterion guarantee that the IMF components retain enough physical meaning of both amplitude and frequency modulations, and it can be accomplished by limiting the size of the standard deviation computed from the two consecutive sifting results as

F 2 V N h (ω) − h (ω) = G 1(k −1) 1k W (6.5) SD B 2 , ω= G h 1(k −1) (ω) W 0 H X

178 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

where N is the number of the original signal samples. A typical value for SD can be set between 0.2 and 0.3. At the end of the sifting process, the relationship among the original data, IMFs, and the k-th residue can be expressed as

M ω = ω + ω (6.6) S( ) rk ( ) Bhk ( ), k =1 where M is the number of IMFs and rk(2) is the final residue of the decomposition. By imposing no assumption about the harmonic nature of the oscillations, the decomposition allows a compact representation, with fewer modes than Fourier or wavelet decomposition. Another feature of the empirical mode decomposition (EMD) is that it operates as a dyadic filter bank, with noise and high frequency components in the first few IMFs, and the lower frequency components in the lower modes [161]. The Hilbert-Huang transform has been proven to be a useful tool for non- stationary signals analysis, including the analysis of temporal [162] and spectral [163] interference patterns. Figure 6.5 shows the result of decomposition of a simulated noisy spectral interference signal for illustration purpose. The EMD decomposes the signal in six IMFs through the sifting process described above. The first component identified as IMF1 corresponds to fast oscillation, while the sixth IMF corresponds to low frequency. Discarding the first and last IMF has an effect of noise reduction and cantering (zero local mean condition) on the original signal. Therefore the signal is reconstructed from IMF2-IMF4 only. The lowest panel of Figure 6.5 shows the reconstructed signal (dotted line) superposed on the corresponding original signal (solid line). Good agreement is achieved. Although there is no criteria on selection of the number of IMFs necessary for recovering the phase of the embedded signal, the analysis of a large number of simulated noisy interference signals shows that, for our purpose, IMF2-IMF4 are sufficient for extraction of embedded signal’s phase. Computation of phase is performed using the HT combined with a standard phase unwrapping integration algorithm. Essentially, the computational requirements of EMD-based signal filtering and subsequent computation of phase are not excessive, making the algorithm amenable to real-time implementation.

179 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#



1

s 0.5 0 0.05 0

IMF1 -0.05 0.5 0

IMF2 -0.5

0.1 0

IMF3 -0.1

0.1 0

IMF4 -0.1

0.2 0

IMF5 -0.2 0.4 0.2 IMF6

1

f 0.5

s,s 0 -0.5 -40 -30 -20 -10 0 10 20 30 40 Relative frequency [GHz]

Figure 6.5 Empirical mode decomposition of a simulated interference pattern. From top to bottom: original signal, intrinsic mode functions (IMF1-IMF6), and the signal and EMD-processed signal obtained from superposition of the modes IMF2-IMF4.

6.2.4 Step 2: PMD Compensation by Pulse Shaping After the phase spectrum (i.e., the phase imposed by the propagation through the fibre at every frequency) is measured, the PMD can be compensated by restoring the phase. In the proposed method, a pulse shaper is used to impose an opposite phase at

180 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

each frequency. Once the phase introduced by the fibre is cancelled, the pulses will recover the temporal shape which they had before entering the fibre. Pulse shaping is a technique used to change the phase and amplitude of a broadband pulse [164-169]. The frequency components (Fourier components) of the pulse are spatially separated using dispersive elements, and an active component changes the amplitude and/or phase of each frequency component independently. Because of its operation and function, this active component is called a spatial light modulator (SLM). The SLM induces spatio-temporal distortions in the pulse, which are proportional to the magnitude of the shaping. After this, the frequency components are combined again to form a new pulse, with a phase and amplitude spectrum modified by the active component. Figure 6.6 illustrates the operation of a pulse shaper with gratings, lenses and SLM. Besides compensating the PMD impairment, the same pulse shaper can be also used to compensate for chromatic dispersion. f ff f

lens lens

grating grating spatial light modulator (SLM) input pulse shaped pulse

Figure 6.6 Pulse shaping operation.

In the pulse shaper, the dispersive component and the combining component are usually gratings or prisms. As for the active SLM component, various technologies can be used, including liquid crystal (LC) arrays, acousto-optical modulator (AOM), deformable mirror (DM), and liquid crystal on silicon (LCoS) processor. In the LC array-based SLM [164], a layer of nematic LC material is placed between two glass slides. The inner surfaces of the slides are coated with transparent indium tin oxide (ITO) conducting film. The ITO on one of the slides is patterned into a linear array of multiple independent electrodes to control individual wavelength pixels. By changing the applied voltages, the LC molecules axes are rotated, which in turn induces refractive index change. Individual phase modulation and pulse shaping is then

181 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

achieved. By having two LC modulators, the amplitude and phase can be modulated simultaneously [165]. A limitation of the LC array-based SLM is its restriction to discrete approximations to the desired spectrum. Also, since the pixellation creases some “dead spaces” between pixels.

The AOM-based SLM uses acousto-optic crystals, such as TeO2, as the active component [166]. A shaped RF pulse creates an acoustic wave that propagates through the crystal. Because the transit time of a femtosecond pulse through the crystal is far too short for significant acoustic propagation, the acoustic wave looks like a modulated diffraction grating, and the different colour components of the laser pulse are independently modulated. The AOM-based SLM has a much faster response time (a few microsecond, limited by the transmission time of the acoustic wave across the AOM) than what the LC-based SLM can achieve. However its efficiency is lower. Another technology is deformable mirrors [167, 168]. Here the spatially dispersed laser pulse is incident onto a mirror surface. At the back of the mirror there are multiple piezoelectric transducer pushers to deform the mirror surface. This will impose time delays for the reflected light at different wavelength components. The amount of delay is proportional to the angle that the mirror normal makes with the incoming beam. As a result, the phases of different wavelength components are modified. However, this method cannot modulate the amplitude of the beam. DM- based SLMs have larger pixels and do not have “dead spaces” as in LC array-based SLMs. The SLM can also be constructed using LCoS platform [169]. Similar to the DM method, the spatially dispersed beam is reflected by the SLM. However the LCoS allows the modulation of both amplitude and phase, similar to the LC array-based SLM.

6.3 Experimental Demonstration of Feedforward Real-Time All-Order PMD Measurement and Compensation Technique 6.3.1 First Order PMD Measurement Experiment The first experiment measures the linear PMD (i.e. DGD), because there is a simple relationship between the first order PMD and the spectral interference pattern as expressed in Equation (6.4). Another reason is that the available PMD emulator can only provide first order PMD effect.

182 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

The experiment setup is shown on Figure 6.7. An ILX Lightwave MPS-8033 precision fibre optic source is used as the optical source. It is a broadband ASE source at the telecom wavelength (1.55 μm). Its output light is passed through an Opto-Link OLCS polarization beam splitter, which acts as a polarizer to provide a linearly polarized probing beam. This probing beam is sent to the JDSU PE3 PMD emulator. Inside the PMD emulator, the optical signal is split into two orthogonal polarization components, a mechanical setup is used to apply delay to one of the arm. This generates a differential group delay between the two polarizations before combining them again, therefore variable linear (first order) PMD effect is introduced to the transmission signal. The output of the emulator is sent through another Opto-Link OLCS polarization beam combiner, which also acts as a polarizer, into an Ando AQ6317B OSA to record the spectral interference pattern. The captured spectra are then analysed by computer to count the number of fringe periods and thus calculate the PMD value. In this experiment, the PMD imposed by the PMD emulator is varied from 0 ps to 40 ps. This is about the range required for the WDM transmission system.

PMD ASE PBS PBC OSA Computer emulator

ASE: Amplified spontaneous emission PBS: Polarization beam splitter PBC: Polarization beam combiner OSA: Optical spectrum analyzer

Figure 6.7 First order PMD measurement experiment setup.

Figure 6.8 shows the spectral interference patterns measured for PMD delays from 0 to 40 ps in 10 ps increments. The x axis is wavelength in nm where the resolution bandwidth is 0.1 nm., and the y axis is power in linear scale. As expected, the period of the spectral interference pattern becomes smaller as the imposed PMD value increases. In other words, there are more fringe periods within the same spectrum window. The PMD for each spectral interference pattern is calculated using the method described previously. The extracted first order PMD value is plotted against the applied PMD value (Figure 6.9). These two values are very close. The errors between the extracted PMD value and the applied PMD value are 0, 4.7%, 3.3%, 0.6% and 0.8% respectively for 0, 10 ps, 20 ps, 30 ps, and 40 ps first order PMD. The slightly larger

183 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

error at the low DGD is due to the fact that that the fringe pattern period is larger and the computer cannot determine partial period accurately. The results of this experiment show that the proposed scheme can measure the linear PMD very accurately.

 0.45 0.45 0.45 0. 4 0.4 0. 4 0.35 0 ps 0.35 10 ps 0.35 20 ps 0. 3 0.3 0. 3 0.25 0.25 0.25 0. 2 0.2 0. 2 0.15 0.15 0.15 0. 1 0.1 0. 1

0.05 0.05 0.05 0 0 0 1537153815391540 154115421543 15371538 1539 15401541 1542 1543 1537 15381539 1540 1541 1542 1543

0. 4 0.4 0. 35 30 ps 0.35 40 ps 0. 3 0.3

0. 25 0.25

0. 2 0.2

0. 15 0.15

0. 1 0.1

0. 05 0.05

0 0 15371538 1539 1540 1541 1542 1543 1537 1538 1539 15401541 1542 1543

Figure 6.8 Spectral interference patterns with 0 to 40 ps PMD delays.

 45 40 35 30 25 20 15

Extracted DGD (ps) DGD Extracted 10 5 0 0 1020304050 Applied linear PMD(ps)

Figure 6.9 Measured DGD from spectral interference as function of the group delay imposed by the PMD emulator.

184 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

This method can be used to calculate higher linear PMD values. The only limiting factor is the resolution of the OSA. The Ando AQ6317B OSA has maximum resolution of 0.015 nm around the 1.55 μm wavelength range (even though the displayed resolution setting is 0.01 nm). In the recent years, OSA with higher resolutions have been commercially available, such as the AP2040 series OSA from Apex Technologies with wavelength resolution of 0.04 pm (corresponding to 5 MHz), and the BOSA series OSA from Aragon Photonics with resolution of 0.08 pm (corresponding to 10 MHz). These technology advancements enable even higher PMD measurement range using the proposed method. In actual field measurements, the input polarization varies in time. In order to obtain high contrast spectral interference patterns we have to consider also the case when the launched light has the same polarization with one of the chosen orthogonal axis decided by our polarizer. This could be resolved in a device by either using launching different polarization states, similar with the case of a spectral polarimeter [170, 171], and/or by using a fast switching waveplate [171] to rotate the polarization and perform the measurement again, with rotated input.

6.3.2 Higher Order PMD Measurement Demonstration Due to the lack of higher order PMD emulator, higher order PMD measurement is demonstrated through simulation. Here a 6 ps Gaussian pulse is used as the signal for the simulation. It propagates through a single channel in a single mode optical fibre. Two perpendicular polarizations acquire the differential phase spectra φ(ω) leading to PMD (dotted lines in the lower graphs of Figure 6.10). Two different scenarios are considered in this demonstration, one where second order dispersion is predominant (left), and one where the third order prevails (right). The spectral interference patterns shown as solid lines in the upper graphs of Figure 6.10 contain 15% amplitude noise, and assume initial Gaussian pulse shape for the input pulses. By applying the EMD-based processing algorithm previously described and depicted in Figure 6.5, we can obtain the filtered signals (dotted lines in upper figures), from which we retrieve the phase spectra shown as solid curves in the lower graphs of Figure 6.10. Note that the phase information can be accurately retrieved in the central part, but is less accurate in the wings, where the signal to noise

185 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

ratio is very poor. However, as shown in Figure 6.11, this has little impact on the recovery of the temporal profile of the initial pulse.

1 1

0.5 0.5

0 0 Normalized intensity Normalized intensity

-0.5 -40 -20 0 20 40 -40 -20 0 20 40 Relative frequency [GHz] Relative frequency [GHz]

20 60 True phase Retrieved phase 40 10

20 0

Phase [rad] 0 Phase [rad] -10

-20 -20

-40 -40 -20 0 20 40 -40 -20 0 20 40 Relative frequency [GHz] Relative frequency [GHz]

Figure 6.10 Upper panel: Second- (left) and third- (right) order PMD simulated noisy interference signals (solid lines) and EMD-based filtered signals (dotted lines); Lower panel: The true phase (dotted lines) used to generate the interference patterns, and the unwrapped phase estimated from the EMD-processed signals (solid lines). All signals are functions of the frequency shift from the centre wavelength of 1.54 μm.

Figure 6.11 shows the temporal profile of a Gaussian pulse before (red dashed lines) and after (black dashed lines) propagating through the medium where it experiences the phase shifts shown in Figure 6.10 left and right, respectively. It is obvious that the phase distortion makes the output pulses unrecognisable. The amount of PMD introduced by such a fibre would make impossible the transmission of information at a data rate of 25 Gb/s, corresponding to the bandwidth of the Gaussian pulses used in Figure 6.11. Once the phase spectra are recovered, they can be applied with opposite sign to the pulses before the detector by the use of a pulse shaper. This is demonstrated here by

186 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

simply adding a phase term exp[iφ(ω)] to the electric field, which is exactly what a pulse shaper will do. The simulation results show that the pulses are recovered (blue solid lines in Figure 6.11) with sufficient accuracy such that signals could be reliably transmitted.  2.0 2.0

1.5 1.5

1.0 1.0 Intensity

0.5 0.5

0 0 -100 0 100 -100 0 100

Time (ps) Time (ps)

Figure 6.11 Demonstration of the pulse distortion due to PMD and PMD compensation using the proposed method. Red dashed lines: initial Gaussian pulses; black dashed lines: distorted and unusable pulses after PMD due to the phase shifts imposed; blue solid lines: the recovered pulses after phase retrieval.

6.3.3 Pulse Shaping Experiment The next experiment demonstrates the pulse shaping capability for the proposed PMD compensator. An AOM-based pulse shaper is designed and constructed. Figure 6.12 shows the schematic of the AOM-based pulse shaper. The beam with the input pulse is projected onto a grating by a reflection mirror. The beam is dispersed to various spectrum components spatially. It is launched into the AOM through a lens. The AOM is modulated by RF pulses programmed by computer and generated by an arbitrary waveform generator. After passing through the AOM-based SLM, part of the beam is undiffracted and discarded. The remaining portions (the useful portions) are collected by a lens and sent to another grating, which combines them and sends the combined beam to the output. The output signal is the shaped pulse.

187 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

RF pulses Undiffracted beam

Grating Grating

Initial pulse Shaped AO pulse modulator Optical source

Figure 6.12 Schematic of an AOM-based pulse shaper.

In this experiment, the AOM modulator is an InP acousto-optic deflector made by Brimrose (Model IPD-200-100-1.55). It operates at 1.55 μm region with centre frequency of 200 MHz and bandwidth of 100 MHz. It has an active aperture of 1mm × 10 mm. The diffraction efficiency at the transducer is about 50% to 60%. Its Bragg angle is 30 mrad and the scanning angle is 60 mrad. It uses an SMA connector for the driving signal. Figure 6.13 is the drawing of the AOM device.

Figure 6.13 Mechanical drawing of the AOM device.

Similar to the previous experiment, the optical source is from the ILX Lightwave MPS-8033 broadband ASE source. Unlike most other experiments in this thesis where the optical signals propagate inside optical fibre, this experiment is performed at free space. The input broadband ASE in a fibre is launched into the free

188 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

space setting by a Thorlabs F810FC-1550 collimator with numerical aperture of 0.24 and focal length of 37.13 mm at 1550 nm. The output shaped signal is returned to the fibre by another F810FC-1550 fibre collimator. The gratings for dispersing the light into spatial components and for combining them are Thorlabs GR25-1210 ruled diffraction grating. It has 1200 grooves per millimetre, and a blaze angle of 36°52’. Its dispersion is 1.50 mrad/nm. Reflective surfaces in the optical path are Thorlabs PF10- 03-M01-10 protected gold mirrors, and Newport 10DC500ER.2 concave broadband metallic mirrors. The RF signal to modulate the dispersed pulse is generated by a LeCroy LW420 WaveStation arbitrary waveform generator. It has a -3dB bandwidth of 100 MHz and a maximum sample rate of 400 MSa/s. The output RF signal is amplified by a Mini- Circuit ZHL-1-2W 2~500 MHz high power RF amplifier and then sent to the Brimrose AOM’s SMA connector via a coaxial cable. Figure 6.14 is the photo of the AOM-based pulse shaper setup. The path where the light propagates is also drawn.

Input fibre Input grating Output fibre RF amplifier

Output grating Acousto-optic modulator

Figure 6.14 Experimental setup of the AOM-based pulse shaper. Yellow lines and arrows indicate the travelling path of light.

189 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

To demonstrate the pulse shaping capability, a simple program is written in LabVIEW, which can generate up to two dips onto the input pulse. The period and depth of these dips can be arbitrarily adjusted. The patterns generated by the program are loaded to the AWG, which then drives the AOM. Figure 6.15 shows the effect of pulse shaping. The top plot is the input spectrum. Six different output spectral patterns are generated under different patterns generated by the computer.

Input signal (a.u.) Intensity

1534 1535 1536 1537 1538 1539 1540 1541 Wavelength (nm) Intensity (a.u.) Intensity (a.u.) Intensity

1534 1535 1536 1537 1538 1539 1540 1541 1534 1535 1536 1537 1538 1539 1540 1541 Wavelength (nm) Wavelength (nm) Intensity (a.u.) Intensity (a.u.) Intensity

1534 1535 1536 1537 1538 1539 1540 1541 1534 1535 1536 1537 1538 1539 1540 1541 Wavelength (nm) Wavelength (nm) Intensity (a.u.) Intensity (a.u.) Intensity

1534 1535 1536 1537 1538 1539 1540 1541 1534 1535 1536 1537 1538 1539 1540 1541 Wavelength (nm) Wavelength (nm) Shaped signal

Figure 6.15 Examples of spectrum manipulation by pulse shaping.

190 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

As mentioned above, the response time of an AOM-based SLM is limited by the transmission time of the acoustic wave across the device. Therefore the constructed pulse shaper has a response time of about 4 microsecond under the acoustic velocity of 5.1×103 m/s. This limits the ability to perform real-time PMD compensation to at most 250 kHz. Furthermore, since the acoustic wave is a travelling wave and not a standing wave, the optical signal needs to have short pulses with pulse widths much lower than 4 microsecond. Therefore this AOM-based pulse shaper cannot be used to restore PMD- impaired signals in actual DWDM optical communication system. Other types of SLM, such as LC-based or LCoS-based, will be required. However, as a proof of concept, the pulse shaping function has been demonstrated. This shows that real time all-order PMD compensation can be achieved using the proposed scheme. Future research topics include constructing pulse shaper using other platforms that allows shorter response time and CW signal, and the demonstration of a complete PMD measurement and compensation system.

6.4 Conclusions In this chapter, I proposed and experimentally demonstrated a novel PMD compensation technique. It uses a feedforward loop that does not require iteration, which enables real-time PMD measurement and compensation. This PMD measurement method is simple yet effective, which reduces the computation complexity compared to other feedforward schemes. It is also effective to compensate for all orders of PMD simultaneously. The main concept is to restore the pulse by restoring the spectrum. The technique includes two steps: measuring the PMD via spectral interference, and compensating for the PMD using pulse shaping. The PMD measurement via spectral interference was described in details, including a procedure based on Hilbert-Huang transfer to retrieve higher order phase information. The performance of PMD measurement was demonstrated by experiment and simulation for first order PMD and higher order PMD respectively. Both delivered very accurate PMD measurement results. The PMD compensation capability was demonstrated using an AOM-based pulse shaper. Various manipulated spectrum patterns were obtained in the experiment. This shows that pulse shaper can effectively modify the signal spectrum to the target

191 )%, )%, <<< %%% (%  (%  (% ,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#,&%,  %+.,%(%) & #(%+)#

pattern. By imposing the appropriate pattern opposite to the PMD-induced spectral change, the signal spectrum can be restored, which also restore the pulse shape and recovered the original signal. This method can compensate PMD at all orders over a wide spectrum in real time, therefore it is suitable for application in the next generation WDM systems. This work has been published in two international conference papers [25, 153] and one journal paper [160]. One US patent application is pending [172].

192 )%, )%, === #!.+#+#!.+#+#!.+#+

)%, )%, ===    

7.1 Thesis Summary In this thesis I developed novel optical devices and subsystems for the next generation DWDM systems. Specifically I proposed and demonstrated five devices/subsystems to deal with five practical problems in the current DWDM systems. Firstly, to solve the problem of sub-optimal spectrum utilization in the DWDM network upgrade due to the limitation that the current optical interleaver can only provide symmetric 50%:50% interleaving ratio between the odd and even channels, I developed a novel tunable asymmetric interleaver whose interleaving ratio can be continuously tuned to any value. I proposed and experimentally demonstrated two methods to realise such interleaver. The first method uses cascading symmetric interleavers, and the second method uses programmable optical processor. Simulation results and experiment results in 10G/40G hybrid system and 40G/100G hybrid system clearly showed that the TAI improved the overall system performance by providing the optimum interleaving ratio (such as 30%:70% for the 10G NRZ-OOK/40G RZ-DQPSK system and 65%:35% for the 43G NRZ-DPSK/112G PDM-RZ-QPSK system). Secondly, I developed a colourless intra-channel optical equalizer to mitigate the inter-symbol interference caused by the passive optical components in the transmission link. By adding this periodic device with passband designed to restore a raised-cosine profile, the passband of the transmission channel is widened, which results in less ISI impairment and improved transmission performance. Experimental results showed that the OEQ widened the passband width by 20%, and improved the opening of the received signal’s eye by 36~40%. This led to a 1.3 order of magnitude improvement in BER performance. Since this device has a colourless periodic profile, a single unit can serve all channels in the DWDM system, therefore is a low cost solution. Thirdly, I developed a novel optical tunable filter that offers two degrees of tunability – both centre frequency and the passband width. It is realised by cascading two tunable edge filters. Using this filter as the key building block, I designed a low cost and expandable ROADM node architecture, which gives priority to the express

193 )%, )%, === #!.+#+#!.+#+#!.+#+

channels, reduced filtering effect, and allows easy customisation and in-service upgrade through a modular configuration. I built a two-ring, four-node network testbed to demonstrate these functions. The results verified that this proposed FBTF-based ROADM node can deliver all the required add/drop/cross-connect functions with little signal degradation. Fourthly, I developed a new transponder aggregator for the colourless and directionless multi-degree ROADM node. Unlike the existing solutions that require costly, large footprint and power consuming demultiplexer, WSS, or tunable filter array, this transponder aggregator does not require any wavelength selector. Instead it uses the local oscillator in each transponder’s coherent receiver to select the target channel. Through theoretical modelling and experimental verification, I demonstrated that this filterless transponder aggregator can achieve less than 0.5 dB OSNR penalty between receiving a single channel and receiving 96 DWDM channels, provided that the balanced receivers have reasonable common mode rejection ratio characteristic and there is sufficient power difference between the local oscillator and the per-channel signal. As part of this work, a method to measure and adjust the CMRR of a balanced photodetector is proposed and demonstrated. Finally, to mitigate the random time-varying PMD effect, I developed a simple yet effective technique that can measure and compensate all orders of the PMD effects in real time. With the main concept of restoring the pulse by restoring the spectrum, this technique includes two steps: measuring the PMD via spectral interference and phase retrieval, and compensating for the PMD using pulse shaping. Experiment showed that this technique can accurately measure the first order PMD. Simulation also demonstrated that higher order PMD can be effectively measured by processing the spectral interference signal and perform phase retrieval using Hilbert-Huang transform. To demonstrate the PMD compensation step, I constructed an AOM-based pulse shaper, which can manipulate the signal profile based on custom configuration. A PMD compensator subsystem can thus be constructed. In summary, these novel devices and subsystems offer new features and capabilities to mitigate respective impairments in high capacity DWDM system, improve signal quality, enhance reliability, reduce hardware size, lower power consumption, and / or reduce capital and operational expense. Therefore they are suitable to be applied in the next generation DWDM system.

194 )%, )%, === #!.+#+#!.+#+#!.+#+

7.2 Future Research For those who wish to carry on with further researches regarding the devices and subsystems proposed in this thesis, the following is a list of directions that I consider to be worthwhile pursuing: • Study the effect of passband slope for the TAI, while maintaining the same asymmetric interleaving ratio. • Use an optical processor (such as LCoS) to design OEQ with customized tunable equalization profile for individual DWDM channel, because these channels might arrive from different paths and have experienced different amounts of filter narrowing effect. • Study flexible-grid WDM network and investigate the application of FBTF in such application. • Develop hardware and DSP technique to allow single-ended receivers to be used in the filterless transponder aggregator. • Design pulse shaper using LC or LCoS platform to provide fast response time for DWDM network application. • Integrate the PMD measurement assembly with the pulse shaper to build a complete PMD compensator sub-system. • Develop optical device or subsystem to compensator for the nonlinear optical effects in the DWDM transmission system.

195   

[1] K. C. Kao and G. A. Hockham, “Dielectric-Fibre Surface Waveguides for Optical Frequencies”, Proc. Inst. Elect. Eng., Vol. 113, pp. 1151-1158, 1966 [2] K. Fukuchi, T. Kasamatsu, et al., “10.92 Tb/s Triple-Bband/Ultra-Dense WDM Optical Repeatered Transmission Experiment”, Technical Digest of OFC 2001, PD24, 2001 [3] X. Zhou, J. Yu, et al., “32Tb/s (320x114Gb/s) PDM-RZ-8QAM Transmission over 580km of SMF-28 Ultra-Low-Loss fiber”, Technical Digest of OFC- NFOEC 2009, PDPB4, 2009 [4] C. Schmidt-Langhorst, R. Ludwig, H. Hu and C. Schubert, “Single-Channel 1- Tb/s Transmission over 480 km DMF for Future Terabit Ethernet Systems”, Technical Digest of OFC-NFOEC 2009, OTuN5, 2009 [5] H. Masuda, E. Yamazaki, et al., “13.5-Tb/s (135 x 111-Gb/s/ch) No-Guard- Interval Coherent OFDM Transmission over 6,248 km using SNR Maximized Second-order DRA in the Extended L-band”, Technical Digest of OFC-NFOEC 2009, PDPB5, 2009 [6] Telecom Industry Association, “2008 Market Review and Forecast”, 2008 [7] Cisco, “Cisco Visual Networking Index – Forecast and Methodology, 2009- 2014”, Cisco white paper, June 2010 [8] T.J. Xia, G. Wellbrock, et al., “Transmission of 107-Gb/s DQPSK over Verizon 504-km Commercial LambdaXtreme Transport System”, Technical Digest of OFC-NFOEC 2008, NMC2, 2008 [9] M. Yagi, S. Satomi and S. Ryu, “Field Trial of 160-Gbit/s, Polarization-Division Multiplexed RZ-DQPSK Transmission System Using Automatic Polarization Control”, Technical Digest of OFC-NFOEC 2008, OThT7, 2008 [10] P. J. Winzer and R.-J. Essiambre, “Advanced Modulation Formats for High- Capacity Optical Transport Networks”, IEEE J. Lightwave Techol., Vol. 24, No. 12, pp. 4711-4728, 2006 [11] N. Kikuchi, K. Sekine and S. Sasaki, “Modulation/Demodulation Schemes for Optical Multi-Level Transmission”, Technical Digest of OFC-NFOEC 2007, OThL3, 2007

196 %-%,%!%+%-%,%!%+%-%,%!%+

[12] H. Kuwahara, “Technologies for 40 Gb/s and 100Gb/s Transmission”, Technical Digest of OFC-NFOEC 2008, NTuB1, 2008 [13] A. S. Siddiqui, S. G. Edirisinghe, J. J. Lepley, J. G. Ellison and S. D. Walker, “Dispersion-Tolerant Transmission Using a Duobinary Polarization-Shift Keying Transmission Scheme,” IEEE Photon. Technol. Lett., Vol. 14, No. 2, pp. 158– 160, 2002 [14] G. Raybon and P. J. Winzer, “100 Gb/s Challenges and Solutions”, Technical Digest of OFC-NFOEC 2008, OThG1, 2008 [15] J. Armstrong, “OFDM: from Copper and Wireless to Optical”, Technical Digest of OFC-NFOEC 2008, OMM1, 2008 [16] H. Takahashi, A. Al Amin, et al., “DWDM Transmission with 7.0-bit/s/Hz Spectral Efficiency Using 8x65.1-Gbit/s Coherent PDM-OFDM Signals”, Technical Digest of OFC-NFOEC 2009, PDPB7, 2009 [17] J. Yu, X. Zhou, et al., “17 Tb/s (161×114Gb/s) PolMux-RZ-8PSK Transmission over 662km of Ultra-Low Loss Fiber by Using Pure C-band EDFA Amplification and Digital Coherent Detection Technology”, Technical Digest of ECOC 2008, Th.3.E.2, 2008 [18] B. Vasic and I. B. Djordjevic, “Low-Density Parity-Check Codes for 40 Gb/s Transmission”, Technical Digest of OFC-NFOEC 2006, OThD2, 2006 [19] I. B. Djordjevic, M. Cvijetic, Lei Xu and Ting Wang, “Proposal for Beyond 100- Gb/s Optical Transmission Based on Bit-Interleaved LDPC-Coded Modulation”, IEEE Photon. Technol. Lett., Vol. 19, No. 12, pp 874 – 876, 2007 [20] J. Kahn and E. Ip, “Principles of Digital Coherent Receiver for Optical Communications”, Technical Digest of OFC-NFOEC 2009, OTuG5, 2009 [21] G.E. Kohnke, S.J. Spammer, et al., “Fiber Bragg Gratings for Dispersion Compensation”, Technical Digest of OFC 2002, ThAA3, 2002 [22] S. Doucet, R. Slavík and S. LaRochelle, “Tunable Dispersion and Dispersion Slope Compensator Using Novel Gires–Tournois Bragg Grating Coupled- Cavities”, IEEE Photon. Technol. Lett., Vol. 16, No. 11, pp. 2529-2531, 2004 [23] S. Lanne and E. Corbel, “Practical Considerations for Optical Polarization-Mode Dispersion Compensators”, IEEE J. Lightwave Techol., Vol. 22, No. 4, 1033- 1040, 2004

197 %-%,%!%+%-%,%!%+%-%,%!%+

[24] S. Kieckbusch, S. Ferber, et al., “Automatic PMD Compensator in a 160-Gb/s OTDM Transmission over Deployed Fiber Using RZ-DPSK Modulation Format”, IEEE J. Lightwave Techol., Vol. 23, No. 1, pp. 165-171, 2005 [25] A. Dogariu, P. N. Ji and T Wang, “All-Order PMD Compensation Using Spectral Interference and Pulse Shaping”, Proc. IEEE Sarnoff Symposium 2007, pp. 1-4, 2007 [26] P. N. Ji, L. Xu, et al., “Inter-Symbol Interference Comparison for Wavelength and Waveband Switching in All-Optical Optical Cross-Connect Nodes”, Proc. OSA Photonics in Switching 2006, O.14.4, pp. 187-189, 2006 [27] C. R. Doerr, S. Chandrasekhar, et al., “Simple Multichannel Optical Equalizer Mitigating Intersymbol Interference for 40-Gb/s Non return-To-Zero Signals”, J. Lightwave Technol., Vol. 22, No. 1, pp. 249-256, 2004 [28] G. Raybon, P. J. Winzer and C. R. Doerr, “10×107-Gbit/s Electronically Multiplexed and Optically Equalized NRZ Transmission over 400 km”, Technical Digest of OFC-NFOEC 2006, PDP32, 2006 [29] H. Sun, K.-T. Wu and K. Roberts, “Real-Time Measurements of a 40 Gb/s Coherent System”, Optics Express, Vol. 16, No. 2, pp. 873-879, 2008 [30] T. Pfau, R. Peveling, et al., “Towards Real-Time Implementation of Coherent Optical Communication”, Technical Digest of OFC-NFOEC 2009, OThJ4, 2009 [31] E. Ip and J. Kahn “Compensation of Dispersion and Nonlinear Impairment Using Digital Backpropagation”, IEEE J. Lightwave Technol., Vol. 26, No. 20, pp. 3416-3425, 2008 [32] S. Binetti, A. Bragheri, E. Iannone and F. Bentivoglio, “Mesh and Multi-Ring Optical Networks for Long-Haul Applications”, IEEE J. Lightwave Technol., Vol. 18, No. 12, pp. 1677-1684, 2000 [33] M. Herzog, M. Maier and A. Wolisz, “RINGOSTAR: An Evolutionary AWG- Based WDM Upgrade of Optical Ring Networks”, IEEE J. Lightwave Technol., Vol. 23, No. 4, pp. 1637-1651, 2005 [34] D. C. Kilper, A. Ferguson, B. O’Sullivan and S. K. Korotky, “Impact of Topology and Traffic on Physical Layer Monitoring in Transparent Networks”, Technical Digest of OFC-NFOEC 2009, OWI3, 2009

198 %-%,%!%+%-%,%!%+%-%,%!%+

[35] T. Wang, R. Izmailov, et al, “Hybrid Hierarchical Optical Cross-Connect Node and Network Applications”, in Optical Transmission, Switching, and Subsystems, C.F. Lam, C. Fan, et al. (eds), Proc. SPIE, Vol. 5281, pp. 112-121, 2004 [36] K. Sato, “The Future of Optical Networks”, Technical Digest of ECOC 2008, We.3.A.1, 2008 [37] P. N. Ji, M.-F. Huang, J. Yu and T. Wang, “Reconfigurable Waveband Cross- Connect and Its Application in 112 Gb/s WDM System”, Technical Digest of OFC-NFOEC 2009, OTuF2, 2009 [38] Y. Chen, D. Al-Salameh, et. al., “Metro Network with Reconfigurable Optical Add-Drop Module”, Proc. 2002 IEEE/LEOS Sum. Top. Meet., 25-26, 2002 [39] P. N. Ji, T. Wang and L. Zong, “ROADM for Metro DWDM Network Using Wavelength Selective Devices”, Proc. ATFO, pp. 139-150, 2004 [40] S. Tibuleac, “ROADM Network Design Issues”, Technical Digest of OFC- NFOEC 2009, NMD1, 2009 [41] P. Roorda and B. Collings, “Evolution to Colorless and Directionless ROADM Architectures”, Technical Digest of OFC-NFOEC 2008, NWE2, 2008 [42] O. Renais, E. Le Rouzic and G. Yven, “Migrating to a Next Gen WDM Network”, Technical Digest of 13th International Telecom. Network Strategy and Planning Symposium, Paper B4, 2008 [43] M. P. McGarry, M. Maier and M. Reisslein, “Ethernet PONs: a Survey of Dynamic Bandwidth Allocation (DBA) Algorithms”, IEEE Commun. Mag., Vol. 42, No. 8, pp. S8–15, 2004 [44] B. Skubic, J. Chen, et al., “A Comparison of Dynamic Bandwidth Allocation for EPON, GPON, and Next-Generation TDM PON”, IEEE Commun. Mag., Vol. 47, No. 3, pp. pp. S40-S48, 2009 [45] G. P. V. Thodime, V. M. Vokkarane and J. P. Jue, “Dynamic Congestion-Based Load Balanced Routing in Optical Burst-Switched Networks”, Technical Digest of IEEE Global Telecom. Conf. 2003. Vol. 5, pp. 2628 – 2632, 2003 [46] W. Wei, M. D. Feuer, et al., “A Novel LightLabel Orthogonal Frequency Division Multiplexing (LL-OFDM) System for Cross-layer Lightpath Monitoring and Setup”, Technical Digest of OFC-NFOEC 2009, PDPA7, 2009

199 %-%,%!%+%-%,%!%+%-%,%!%+

[47] A. Banerjee, J. Drake, et al., “Generalized Multiprotocol Label switching: An Overview of Routing and Management Enhancements,” IEEE Commun. Mag., Vol. 39, No. 1, pp. 144–150, 2001 [48] R. Parthiban, R. S. Tucker and C. Leckie, “Waveband Grooming and IP Aggregation in Optical Networks”, IEEE J. Lightwave Technol., Vol. 21, No. 11, pp. 2476–2488, 2003 [49] R. Martinez, C. Pinart, et al., “Challenges and Requirements for Introducing Impairment-Awareness into the Management and Control Planes of ASON/GMPLS WDM Networks,” IEEE Commun. Mag., Vol. 44, No. 12, pp. 76–85, 2006 [50] N. Sambo, N. Andriolli, et al., “GMPLS-Controlled Dynamic Translucent Optical Networks”, IEEE Network, Vol. 23, No. 3, pp. 34–40, 2009 [51] C. Mas, I. Tomkos and O. K. Tonguz, “Failure Location Algorithm for Transparent Optical Networks”, IEEE J. Selected Areas in Commun., Vol. 23, No. 8, pp. 1508-1519, 2005 [52] K. W. Lim, E. S. Son, K. H. Han and Y. C. Chung, “Fault Localization in WDM Passive Optical Network by Reusing Downstream Light Sources”, IEEE Photon. Technol. Lett., Vol. 17, No. 12, pp. 2691-2693, 2005 [53] S. Ramamurthy, L. Sahasrabuddhe and B. Mukherjee, “Survivable WDM Mesh Networks”, IEEE J. Lightwave Technol., Vol. 21, No. 4, pp. 870-883, 2003 [54] D. Griffith and S.K. Lee, “A 1+1 Protection Architecture for Optical Burst Switched Networks”, IEEE J. Selected Areas in Commun., Vol. 21, No. 9, pp. 1384-1398, 2003 [55] C.-H. Yeh, S.-C. Lee and S. Chi, “A Protection Method for Ring-Type TDM- PONs against Fiber Fault”, Technical Digest of OFC-NFOEC 2007, JThA76, 2007 [56] S. Ramasubramanian, “Supporting Multiple Protection Strategies in Optical Networks”, IEEE/ACM Transactions on Networking, Vol. 16, No. 6, pp. 1352– 1365, 2008 [57] B. Shine and J. Bautista, “Interleavers Make High-Channel-Count Systems Economical”, Lightwave, Vol. 17, No. 8, pp. 140-144, 2000

200 %-%,%!%+%-%,%!%+%-%,%!%+

[58] Q. Wang and S. He, “Optimal Design of a Flat-Top Interleaver Based on Cascaded M-Z Interferometers by Using a Genetic Algorithm”, Optics Commun., Vol. 224, pp. 229-236, 2003 [59] Q. Wang, Y. Zhang and Y. C. Soh, “All-Fiber 3×3 Interleaver Design with Flat- Top Passband”, IEEE Photon. Technol. Lett., Vol. 16, No. 1, pp. 168-170, 2004 [60] T. Zhang, K. Chen and Q. Sheng, “A Novel Interleaver Based on Dual-Pass Mach-Zehnder Interferometer”, Microwave and Optical Technol. Lett., Vol. 42, No. 3, pp. 253-255, 2004 [61] C-H. Huang, Y. Li, et al., “Low-Loss Flat-Top 50-GHz DWDM and Add/Drop Modules Using All-Fiber Fourier Filters”, Technical Digest of NFOEC 2000, pp. 311-316, 2000 [62] F. Liang, H. Zeng, et al., “A Bulk Optical Interleaver with Flat Passband by Using Cascaded Mach-Zehnder Interferometers, Optical Components and Transmission System”, in Optical Components and Transmission Systems, W. Hu, S. Sudo and P. Kaiser (eds), Proc. SPIE, Vol. 4906, pp. 442-452, 2002 [63] H. Li, R. Huang, “The Application and Technical Approaches of Interleaver”, in Passive Components and Transmission Systems, C. Fan and N. S. Kwong (eds), Proc. SPIE, Vol. 4581, pp. 79-87, 2001 [64] J. Shi, X. Chen, et al., “Design of Polarization-Based Interleaver Filter for DWDM Mux/Demux, Passive Components and Transmission Systems”, in Passive Components and Transmission Systems, C. Fan and N. S. Kwong (eds), Proc. SPIE, Vol. 4581, pp. 348-355, 2001 [65] J. Zhang, L. Liu and Y. Zhou, “A Tunable Interleaver Filter Based on Analog Birefringence Units”, Optics Commun., Vol. 227, pp. 283-294, 2003 [66] H. Arai, T. Chiba, et al., “Interleave Filter with Box-Like Spectral Response and Low Chromatic Dispersion”, Technical Digest of NFOEC 2000, Vol. 2, pp. 444- 451, 2000 [67] E. Sidick, P. Peters, et. al., “Highly Integrated and Scalable PLC Interleaver Modules”, in Optical Fiber and Planar Waveguide Technology II, S. Jian, S. Shen and K. Okamoto (eds), Proc. SPIE, Vol. 4904, pp. 80-87, 2002 [68] D.-W. Huang, T.-H. Chiu and Y Lai, “Arrayed Waveguide Grating DWDM Interleaver”, Technical Digest of OFC 2001, Vol. 3, pp. WDD80-1-3, 2001

201 %-%,%!%+%-%,%!%+%-%,%!%+

[69] L. Lin, F. Zujie, et. al., “Sampled FBG Based Optical Interleaver”, Technical Digest of CLEO, Vol. I, pp. I-404-405, 2001 [70] L. P. Chislain, R. Sommer, et. al., “Miniature Solid Etalon Interleaver”, Technical Digest of NFOEC 2001, pp. 1397-1403, 2001 [71] H. Chen, P. Gu, et. al., “Optical Interleaver Based on Multi-Cavity Fabry-Perot Thin Film Filter”, in Optical Components and Transmission System, W. Hu, S. Sudo and P. Kaiser (eds), Proc. SPIE, Vol. 4906, pp. 561-566, 2002 [72] C-H. Hsieh, R. Wang, et. al., “Flat-Top Interleavers Using Two Gires-Tournois Etalons as Phase-Dispersive Mirrors in a Michelson Interferometer”, IEEE Photon. Technol. Lett., Vol. 15, No. 2, pp. 242-244, 2003 [73] Y. Shao, Y. Jiang, et. al., “Research of a Novel Interleaver Basing on Modified Michelson Interferometer with Three-Mirror Fabry-Perot Interferometer and Gires-Tournois Resonator”, in Optical Components and Transmission System, Proceedings of SPIE, Vol. 4906, 2002, pp. 484-488 [74] B. Zhang, D. Huang, BGT “Interleaver Used in DWDM Optics Communications System”, in Optical Components and Transmission System, W. Hu, S. Sudo and P. Kaiser (eds), Proc. SPIE, Vol. 4096, pp. 493-497, 2002 [75] C. J. Kaalnd, Z. Jin, et al., “Novel Optical Wavelength Interleaver Based on Symmetrically Parallel-Coupled and Apodized Ring Resonator Arrays”, in Photorefractive Fiber and Crystal Devices: Materials, Ooptical Properties, and Applications IX, F. T. S. Yu, R. Guo and S. Yin (eds), Proc. SPEI, Vol. 5206, pp. 157-165, 2003 [76] B. B. Dingel, “Recent Development of Novel Optical Interleaver: Performance and Potential”, in Active and Passive Optical Components for WDM Communications III, A. K. Dutta, A. A. S. Awwal, et al. (eds), Proc. SPIE, Vol. 5246, pp. 570-581, 2003 [77] L. Ceuppens, T. Schmidt, et al., “Spectrally Efficient Modulation Enables Incremental Capacity Growth by Retrofitting 40Gb/s Channels into Existing 10Gb/s Networks”, Technical Digest of NFOEC 2003, pp. 516-525, 2003 [78] Michael H. Eiselt and Brian T. Teipen, “DPSK-3ASK Transmission Optimization by Adapting Modulation Levels”, in Optical Transmission, Switching, and Subsystems VI, K. Kitayama, P. C. Ghiggino, et al. (eds), Proc. SPIE, Vol. 7136, 71360H, 2008

202 %-%,%!%+%-%,%!%+%-%,%!%+

[79] S. X. F. Cao and X. Mao, “Dense Wavelength Division Multiplexer Utilizing an Asymmetric Pass Band Interferometer”, US Patent US6310690, published October 2001 [80] P. N. Ji, A. Dogariu, et al., “Optical Tunable Asymmetric Interleaver”, Technical Digest of OFC-NFOEC 2006, OTuM7, 2006 [81] P. N. Ji, T. Wang, et al., “Optical Tunable Asymmetric Interleaver and Upgrade for Dense Wavelength Division Multiplexed Networks”, US patent US2007/0116468, published May 2007 [82] K. M. Johnson, D. J. McKnight and I. Underwood, “Smart Spatial Light Modulators Using Liquid Crystals on Silicon”, IEEE J. Quant. Electron., Vol. 29, No. 2, pp. 699-714, 1993 [83] G. Baxter, S. Frisken, et al., “Highly Programmable Wavelength Selective Switch Based on Liquid Crystal on Silicon Switching Elements" Technical Digest of OFC-NFOEC 2006, OTuF2, 2006 [84] S. Frisken, D. Abakoumov, et al., “Advances in Wavelength Selective Switching and Processing Employing Liquid Crystal on Silicon”, Proc. 7th COIN, pp. 1-2, 2008 [85] M.A.F. Roelens, S. Frisken, et al., “Dispersion Trimming in a Reconfigurable Wavelength Selective Switch”, IEEE J. Lightwave Technol., Vol. 26, No. 1, pp. 73-78, 2008 [86] M. A. Roelens, J. A. Bolger, D. Williams and B. J. Eggleton, “Multi-Wavelength Synchronous Pulse Burst Generation with a Wavelength Selective Switch”, Optics Express, Vol. 16, No., 14, pp. 10152-10157, 2008 [87] J. D. Ralson, J. M. Kahn and K-P. Ho, “Spectrally Efficient and Impairment- Robust Modulation Techniques for 40 Gb/s Optical Transmission Systems”, Technical Digest of NFOEC 2002, pp. 317-324, 2002 [88] P. N. Ji, J. Yu, G.-D. Peng and T. Wang, “Passband Optimisation for Hybrid 40G/100G System Using Tunable Asymmetric Interleaver”, Technical Digest of ECOC 2010, Mo.1.A.3, 2010 [89] G. Jacobsen, Noise in Digital Optical Transmission Systems, Norwood, MA: Artech House, 1994

203 %-%,%!%+%-%,%!%+%-%,%!%+

[90] P. A. Humblet and M. Azizoglu, “On the Bit Error Rate of Lightwave Systems with Optical Amplifiers,” IEEE J. Lightwave Technol., Vol. 9, No. 11, pp. 1576– 1582, 1991 [91] J. Proakis, Digital Communications, 4th edition, 2001 [92] P. N. Ji, J. Yu, et al., “Waveband Deaggregator Cascading Effect in Multi- Granularity Optical Cross-Connect Nodes”, Proc. 19th LEOS, WW4, 2006 [93] P. M. A. Charrua and A.V. T. Cartaxo, “Optimized Filtering for AMI-RZ and DCS-RZ SSB Signals in 40-Gb/s/ch-Based UDWDM Systems”, IEEE Photon. Technol. Lett., Vol. 17, No. 1, pp. 223-225, 2005 [94] S. Bigo, “Multiterabit/s DWDM Terrestrial Transmission with Bandwidth- Limiting Optical Filtering”, IEEE J. Selected Topics in Quantum Electron., Vol. 10, No. 2, pp. 329-340, 2004 [95] X. Zhao and F.S. Choa, “Demonstration of a 10Gb/s Transmissions over a 1.5 km Long Multimode Fiber Using Equalization Techniques,” IEEE Photon. Technol. Lett., pp. 1187-1189, 2002 [96] G. S. Kanter, A. K. Samal, O. Coskun and A. Gandhi, “Electronic Equalization for Enabling Communications at OC-192 Rates Using OC-48 Components”, Opt. Express, Vol. 11, No. 17, pp. 2019-2029, 2003 [97] G. S. Kanter, A. K. Samal and A. Gandhi, “Electronic Dispersion Compensation for Extended Reach”, Technical Digest of OFC 2004, TuG1, 2004 [98] H. L. An, X. Z. Lin, E. Y. B. Pun and H. D. Liu, “Multi-Wavelength Operation of an Erbium-Doped Fiber Ring Laser Using a Novel Dual-Pass Mach-Zehnder Comb Filter”, Optics Communications, Vol. 169, No. 1-6, pp. 159-165, 1999 [99] A. Bellemare, M. Karasek, et al, “Room Temperature Multifrequency Erbium- Doped Fiber Laser Anchored on the ITU Frequency Grid”, IEEE J. Lightwave Technol., Vol.18, No. 6, pp. 825-831, 2000 [100] X. J. Gu, “Wavelength-Division Multiplexing Isolation Fiber Filter and Light Source Using Cascaded Long-Period Fiber Grating”, Optics Lett., Vol. 23, No. 7, pp. 509-510, 1998 [101] X. Fang and R. O. Claus, “Polarization-Independent All-Fiber Wavelength- Division Multiplexer Based on a Sagnac Interferometer”, Optics Lett., Vol. 20, No. 20, pp. 2146-2148, 1995 [102] A. Yariv, Optical Electronics in Modern Communications, 5th edition, 1997

204 %-%,%!%+%-%,%!%+%-%,%!%+

[103] T. Wang, L. Xu, P. N. Ji, et al., “Optical Equalization Filtering of DWDM Channels”, US patent application US2007/0206898, published September 2007 [104] P. N. Ji, L. Xu, et al., “Inter-Symbol Interference-Suppressed Colorless DPSK Demodulator”, US patent application US2008/0240736, published October 2008 [105] P. N. Ji, L. Xu, et al., “Intra-Channel Equalizing Optical Interleaver for DWDM Communication Systems”, US patent application US2009/0162066, published June 2009 [106] M. Mezhoudi, R. Feldman, et al., “The Value of Multiple Degree ROADMs on Metropolitan Network Economics”, Technical Digest of OFC-NFOEC 2006, NThA4, 2006 [107] B. Bacque and D. Oprea, “R-OADM Architecture – Now You Can Control the Light”, Tropic Networks white paper, 2003 [108] M. Vasilyev, I. Tomkos, et. al., “Transparent Ultra-Long-Haul DWDM Networks with ‘Broadcast-and-Select’ OADM/OXC Architecture”, IEEE J. Lightwave Technol., Vol. 21, No. 11, pp. 2661-2672, 2003 [109] J-K. Rhee, I. Tomkos and M-J Li, “A Broadcast-and-Select OADM Optical Network with Dedicated Optical-Channel Protection”, IEEE J. Lightwave Technol., Vol. 21, No. 1, pp. 25-31, 2003 [110] J. Gruber, J. Bosloy and G. Harney, “Optical Transport and Switching”, Technical Digest of NFOEC 2002, pp. 473-484, 2002 [111] P. N. Ji, Y. Aono and T. Wang, “Reconfigurable Optical Add/Drop Multiplexer Based on Bidirectional Wavelength Selective Switches”, Proc. Photonics in Switching 2010, PWB1, 2010 [112] J. Hsieh, “Tunable Filter Based Reconfigurable Optical Network”, Proc. 14th WOCC, p. 33, 2005. [113] Y. Hibino, “An Array of Photonic Filtering Advantages: Arrayed-Waveguide- Grating Multi/Demultiplexers for Photonic Networks”, IEEE Circuits and Devices Mag., Vol. 16, No. 6, pp. 21-27, 2000 [114] T. Wang, P. N. Ji, et al., “Flexible Band Tunable Filter”, US patent US7099529, issued August 2006 [115] H. J. R. Dutton, Understanding Optical Communications, IBM, 1998

205 %-%,%!%+%-%,%!%+%-%,%!%+

[116] J. M. H. Elmirghani and H. T. Mouftah, “Technologies and Architectures for Scalable Dynamic Dense DWM Networks”, IEEE Commun. Mag., Vol. 38, No. 2, pp. 58-66, 2000 [117] P. N. Ji, L. Zong, L. Xu and T Wang, “Passband Width Tunable Filter and Its Application in Low Cost ROADM Node”, Proc. 5th ICOCN/ 2nd ATFO, pp. 44- 48, 2006 [118] T. Wang, P. N. Ji, et al., “Flexible Band Tunable Add/Drop Multiplexer and Modular Optical Node Architecture”, US patent application US2005/0281558, published December 2005 [119] P. N. Ji, J. Yu, D. Qian and T. Wang, “Spectrally Efficient 3.4b/s/Hz Waveband Switching over Wavelength-Selective Switch-Based ROADM on 114 Gb/s PolMux-RZ-8PSK DWDM System”, in Optical Transmission, Switching and Subsystems VI, K. Kitayama, P. C. Ghiggino, et al. (eds), Proc. SPIE, Vol. 7136, 71361X, 2008 [120] L. Zong, P. N. Ji, et al., “Testbed for ROADM and WXC Based Metro WDM Networks”, in Network Architectures, Management, and Applications III, K.-W. Cheung, G.-K. Chang, G. Li and K. Sato (eds), Proc. SPIE, Vol. 6022, pp. 379- 385, 2005 [121] T. Wang, P. N. Ji, et al., “Flexible Waveband Aggregator and Deaggregator and Hierarchical Hybrid Optical Cross-Connect System”, US patent US7239772, issued July 2007 [122] P. N. Ji, and Y. Aono, “Colorless and Directionless Multi-Degree Reconfigurable Optical Add/Drop Multiplexers,” Proc.19th WOCC, OA1, 2010 [123] V. Kaman, R. J. Helkey and J. E. Bowers, “Multi-Degree ROADM’s with Agile Add-Drop Access", Proc. Photonics in Switching, pp. 31-32, 2007 [124] S. Thiagarajan, L. Blair and J. Berthold, “Direction-Independent Add/Drop Access for Multi-Degree ROADMs”, Technical Digest of OFC-NFOEC 2008, OThA7, 2008 [125] J. Chen, J. Jacob, et al., “En Route to Grouping-Constraint Free, Colorless Directionless ROADMs”, Technical Digest of OFC-NFOEC 2010, NThF1, 2010 [126] K. Papakos, “Leveraging Directionless & Colorless Updates to Existing ROADM-Based Networks”, Technical Digest of OFC-NFOEC 2009, Market Watch Panel IV, 2009

206 %-%,%!%+%-%,%!%+%-%,%!%+

[127] A. Sahara, Y. Tsukishima, et al., “Demonstration of Colorless and Directed/ Directionless ROADMs in Router Network”, Technical Digest of OFC-NFOEC 2009, NMD2, 2009 [128] G. Charlet, M. Salsi, et al., “72×100Gb/s Transmission over Transoceanic Distance, Using Large Effective Area Fiber, Hybrid Raman-Erbium Amplification and Coherent Detection”, Technical Digest of OFC-NFOEC 2009, PDPB6, 2009 [129] A. H. Gnauck, P. J. Winzer, et al., “10×112-Gb/s PDM 16-QAM Transmission over 630 km of Fiber with 6.2-b/s/Hz Spectral Efficiency”, Technical Digest of OFC-NFOEC 2009, PDPB8, 2009 [130] A. Sano, T. Kobayashi, et al., “240-Gb/s Polarization-Multiplexer 64-QAM Modulation and Blind Detection Using PLC-LN Hybrid Integrated Modulator and Digital Receiver”, Technical Digest of ECOC 2009, PD2.2, 2009 [131] M. Salsi, H. Mardoyan, et al., “115×110Gbit/s Coherent PDM-PQSK Transmission over 7200km”, Technical Digest of ECOC 2009, PD2.5, 2009 [132] Y. Ma, Q. Yang, et al., “1-Tb/s per Channel Coherent Optical OFDM Transmission with Subwavelength Bandwidth Access”, Technical Digest of OFC-NFOEC 2009, PDPC1, 2009 [133] D. Qian, N. Cvijetic, et al., “Single-Wavelength 108 Gb/s Upstream OFDMA- PON Transmission”, Technical Digest of ECOC 2009, PD3.3, 2009 [134] S. Chandrasekhar, X. Liu, et al., “Transmission of a 1.2-Tb/s 24-Carrier No- Guard-Internal Coherent OFDM Superchannel over 7200-km of Ultra-Large- Area Fiber”, Technical Digest of ECOC 2009, PD2.6, 2009 [135] G. Charlet, M. Salsi, et al., “Transmission of 81 Channels at 40Gbit/s over a Transpacific-Distance Erbium-Only Link, Using PDM-BPSK Modulation, Coherent Detection, and a New Large Effective Area Fibre”, Technical Digest of ECOC 2008, Th.3.E.3, 2008 [136] A. Caballero, D. Zibar and I. Tafur Monroy, “Digital Coherent Detection of Multi-Gigabit 16-QAM Signals at 40 GHz Carrier Frequency Using Photonic Downconversion”, Technical Digest of ECOC 2009, PD3.4, 2009 [137] R. Noé, “Phase Noise-Tolerant Synchronous QPSK/BPSK Baseband-Type Intradyne Receiver Concept with Feedforward Carrier Recovery,” IEEE J. Lightwave Technol., Vol. 23, No. 2, pp. 802−808, 2005

207 %-%,%!%+%-%,%!%+%-%,%!%+

[138] L. Kazovsky, “Multichannel Coherent Optical Communications Systems,” IEEE J. Lightwave Technol., Vol. 5, No. 8, pp. 1095-1102, 1987 [139] L. E. Nelson, S. L. Woodward and P. D. Magill, “Real-Time Detection of a 40 Gbps Intradyne Channel in Presence of Multiple Received WDM Channels,” Technical Digest of OFC-NFOEC 2010, OMJ1, 2010 [140] G. Agrawal, Fiber-Optic Communication Systems, pp. 114−116, John Wiley & Sons, New York, 2002 [141] P. N. Ji, Y. Aono, et al., “Transponder Aggregator Without Wavelength Selector for Colorless and Directionless Multi-Degree ROADM Node”, US patent application 12-900220, published October 2009 [142] P. N. Ji, Y. Aono, et al., “Real-Time Receiver Optimization in Filter-Less Coherent Receiving System”, US provisional patent application 61/332390, filed April 2010 [143] C. D. Poole and R. E. Ranger, “Phenomenological Approach to Polarisation Dispersion in Long Single-Mode Fibres”, Electronics Lett., Vol. 22, No. 19, pp. 1029-1030, 1986 [144] R. Khosravani, S. A. Havstad, et al., “Polarization-Mode Dispersion Compensation in WDM Systems”, IEEE Photon. Technol. Lett., Vol. 13, No. 12, pp. 1370-1372, 2001 [145] S. Lanne and E. Corbel, “Practical Considerations for Optical Polarization-Mode Dispersion Compensators”, IEEE J. Lightwave Technol., Vol., 22, No. 4, pp. 1033-1040, 2004 [146] S. Kieckbusch, S. Ferber, et.al., “Automatic PMD Compensator in a 160-Gb/s OTDM Transmission over Deployed Fiber Using RZ-DPSK Modulation Format”, IEEE J. Lightwave Technol., Vol. 23, No. 1, pp. 165-171, 2005 [147] P. B. Phua, H. A. Haus and E. P. Ippen, “All-Frequency PMD Compensator in Feedforward Scheme”, IEEE J. Lightwave Technol,. Vol. 22, No. 5, pp. 1280- 1289, 2004 [148] J. P. Gordon and H. Kogelnik, “PMD Fundamentals: Polarization Mode Dispersion in Optical Fibers”, Proc. Natl. Acad. Sci. USA, Vol. 97, No. 9, pp. 4541-4550, 2000 [149] M. Mujat and A. Dogariu, “Polarimetric and Spectral Changes in Random Electromagnetic Fields”, Optics Lett., Vol. 28, No. 22, pp. 2153-2155, 2003

208 %-%,%!%+%-%,%!%+%-%,%!%+

[150] C. D. Poole and D. L. Favin, “Polarization-Mode Dispersion Measurements Based on Transmission Spectra Through a Polarizer”, IEEE J. Lightwave Technol,. Vol. 12, No. 6, pp. 917-929, 1994 [151] E. Wolf, “Unified Theory of Coherence and Polarization of Random Electromagnetic Beams”, Phys. Lett. A, Vol. 312, No. 5-6, pp. 263-267, 2003 [152] P. S. Westbrook, L. E. Nelson, et al., “Spectral Polarimetry with a Differential Group Delay Bias”, Optics Lett., Vol. 29, No. 22, pp. 2593-2595, 2004 [153] A. Dogariu, P. N. Ji, and T. Wang, “Real-Time All-order PMD Compensation Using Spectral Interference”, Proc. IEEE Sarnoff Symposium 2008, PP. 1-4, 2008 [154] M. Takeda, H. Ina and S. Kobayashi, “Fourier-Transform Method of Fringe- Pattern Analysis for Computer-Based Topography and Interferometry”, J. Opt. Soc. Am., Vol. 72, No. 1, pp. 156-160, 1982 [155] L. R. Watkins, S. M. Tan and T. H. Barnes, “Determination of Interferometer Phase Distributions by Use of Wavelets”, Optics Lett., Vol. 24, No. 13, pp. 905- 907, 1999 [156] A. Dogariu, A. Nahata, et al., “Optical Pulse Propagation Through Metallic Nano-Apertures”, Appl. Phys. B, Vol. 74, Suppl., pp. s69-s73, 2002 [157] E. Bedrosian, “A Product Theorem for Hilbert Transform”, Proc. IEEE, Vol. 51, No. 5, pp. 868-869, 1963 [158] A. H. Nuttall and E. Bedrosian, “On the Quadrature Approximation to the Hilbert Transform of Modulated Signals”, Proc. IEEE, Vol. 54, No. 10, pp. 1458-1459, 1966 [159] N. E. Huang, Z. Shen, et al., “The Empirical Mode Decomposition and the Hilbert Spectrum for Nonlinear and Non-Stationary Time Series Analysis”, Proc. R. Soc. London, Ser. A, Vol. 454, No. 1971, pp. 903-995, 1998 [160] A. Dogariu, P. N. Ji, L. Cimponeriu and T. Wang, “PMD Compensation via Real- Time Phase Retrieval from Spectral Interference”, Optics Commun., Vol. 282, No. 18, pp. 3706-3711, 2009 [161] P. Flandrin, G. Rilling and P. Goncalv`es, “Empirical Mode Decomposition as a Filter Bank”, IEEE Sig. Process. Lett, Vol. 11, No. 2, pp. 112-114, 2004

209 %-%,%!%+%-%,%!%+%-%,%!%+

[162] F. A. M. Rodriguez, A. Federico and G. H. Kaufmann, “Hilbert Transform Analysis of a Time Series of Speckle Interferograms with a Temporal Carrier”, Appl. Opt., Vol. 47, No. 9, pp. 1310-1316, 2008 [163] S. Equis and P. Jacquot, “The Empirical Mode Decomposition: a Must-Have Tool in Speckle Interferometry?”, Opt. Express, Vol. 17, No. 2, pp. 611-623, 2009 [164] A. M. Weiner, D. E. Leaird, et. al., “Programmable Femtosecond Pulse Shaping by Use of a Multielement Liquid-Crystal Phase Modulator”, Optics Lett., Vol. 15, No. 6, pp. 326-328, 1990 [165] M. M. Wefers, K. A. Nelson, “Programmable Phase and Amplitude Femtosecond Pulse Shaping”, Optics Letters, Vol. 18, No. 23, pp. 2032-2034, 1993 [166] C. W. Hillegas, J. X. Tull, et. al., “Femtosecond Laser Pulse Shaping by Use of Microsecond Radio-Frequency Pulses”, Optics Lett., Vol. 19, No. 10, pp. 737- 739, 1994 [167] K. F. Kwong, D. Yankelevich, et al., “400-Hz Mechanical Scanning Optical Delay Line”, Optics Lett., Vol. 18, No. 7, pp. 558-560, 1993 [168] A. Efimov and D. H. Reitze, “Spectral Adaptive Optics: Phase Compensation for Ultrashort Chirped Pulse Amplifier Systems”, in Generation, Amplification, and Measurement of Ultrashort Laser Pulses III, W. E. White and D. H. Reitze (eds), Proc. SPIE, Vol. 2701, pp. 190-197, 1996 [169] A. M. Clarke, D. G. Williams, et al., “Parabolic Pulse Shaping for Enhanced Continuum Generation Using an LCoS-based Wavelength Selective Switch”, Proc. 14th OECC, WD6, 2009 [170] P. S. Westbrook, L. E. Nelson, S. Wielandry and J. Fini, “Spectral Polarimetry with a Differential Group Delay Bias”, Optics Lett., Vol. 29, No. 22, pp. 2593- 2595, 2004 [171] S. X. Wang and A. M. Weiner, “Fast Wavelength-Parallel Poalrimeter for Broadband Optical Networks”, Optics Lett., Vol. 29, No. 9, pp. 923-925, 2004 [172] A. Dogariu, P. N. Ji, et al., “All Order Polarization Mode Dispersion Compensation with Spectral Interference Based Pulse Shaping”, US patent application US2008/0002972, published January 2008

210