Your SlideShare is downloading. ×
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Homodyne OLT-ONU Design for Access Optical Networks
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

Homodyne OLT-ONU Design for Access Optical Networks

2,559

Published on

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
2,559
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
100
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. Ph. D. Thesis Optical Communications Group Department of Signal Theory and Communications Universitat Politècnica de Catalunya Homodyne OLT-ONU design for access optical networks Author Josep Mª Fàbrega Advisor Josep Prat Thesis presented in fulfillment of the doctorate program of the signal theory and communications department March 2010
  • 2. The work described in this thesis was performed in the Signal Theory and Communications department of the Universitat Politècnica de Catalunya / BarcelonaTech. Josep Mª Fàbrega Homodyne OLT-ONU design for access optical networks Subject headings: Optical communications, fibers and telecomm Copyright © 2010 by Josep Mª Fàbrega All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means without the prior written consent of the author. Printed in Barcelona, Spain ISBN: 978-84-693-3168-2 Reg: 10/53978
  • 3. ”The most exciting phrase to hear in science, the one that heralds the most discoveries, is not Eureka! (I found it!) but ’That’s funny...’” Isaac Asimov
  • 4. ` UNIVERSITAT POLITECNICA DE CATALUNYA (UPC) Abstract Optical Communications Group (GCO) Signal Theory and Communications Department (TSC) Doctor of Philosophy by Josep M. F`brega a Nowadays, when talking about access networks, advanced multimedia applications are changing customer demands, requiring much higher speed connection. Thus, other al- ternatives to deployed Time Division Multiplex Passive Optical Networks (TDM-PONs) are appearing to increase available bandwidth. Wavelength Division Multiplex provides virtual point-to-point connections, so multiplies the effective bandwidth that the fiber can offer. A significant step forward is Ultra-Dense WDM (UD-WDM), where wavelengths are separated by just a few GHz, increasing the number of channels that can be accommo- dated on a single fiber. Following this line, if narrow channel-spacing could be achieved, a new philosophy of Wavelength-To-The-User (λTTU) can be envisaged, multiplying the number of connections as well as maintaining high data rates. One of the enabling technologies for such challenge can be coherent transmission and reception systems. First of all because they allow the use of improved modulation formats (like Phase Shift Keying - PSK), extending the reach of the networks. Secondly, as they use electrical filtering for channel selection, narrow channel spacing can be achieved while maintaining high speed connection. The most promising technology for achieving these performances is homodyne reception. Several novel transceiver architectures, based in homodyne reception, are proposed and experimentally evaluated in this work. The most robust and simple of the considered architectures has been fully developed and prototyped in order to be used in a net- work test-bed. For that prototype, transmission experiments demonstrate a sensitivity of −38.7 dBm sensitivity at 1 Gb/s, while featuring a power budget of 47 dB. Furthermore, different PON architectures are proposed and specifically designed for the proposed transceivers. With the experimental prototype previously developed, network deployment is obtained, capable to serve up to 1280 users at maximum distance of 27 km and featuring a maintained data rate of 1 Gb/s per user.
  • 5. Acknowledgements First of all I want to express my gratitude to my advisor Prof. Josep Prat for having given to me the opportunity to join the optical communications research group and develop my Ph.D. within it. His guidance and friendship have set the cornerstone of the work presented in this thesis. These investigations would not have been possible without the full support of the optical communications group at UPC. My special thanks to Jos´ L´zaro, Bernhard Schrenk, e a Carlos Bock, Joan Gen´ and Jaume Comellas for their advice and fruitful discussions, e also demonstrating their sincere friendship. A warm hug to thank all the colleagues for making an enjoyable atmosphere everyday during these years. In this aspect I would like to emphasize the support of the remaining members of the Access and Transmission team: Eduardo T. L´pez, Mireia Omella, Victor Polo and specially Francesc Bonada, o for his unvaluable help in the network administration. Also I want to acknowledge the support of those that not belong to GCO: The entire SI-TSC team and our colleagues from i2CAT, with who we shared the same space for many years. Special thanks to Lutz Molle and Ronald Freund, for their valuable support and friend- liness, particularly during my stay at HHI. Thanks to Ahmad ElMardini, Rich Baca and Ricardo Saad, from Tellabs Inc., for their help during the test period of the SCALING contract. Also I would like to mention Marco Forzati and ACREO for bringing us the opportunity of collaboration with them and Syntune. I am very thankful to all master thesis students I supervised and co-supervised. The herewith presented work wouldn’t been possible without their contributions. In chrono- ıs u n ` logical order: Llu´ Vilabr´, Joan Miquel Pi˜ol, Miquel Angel Mestre and Marc Vilalta (almost finishing). On the personal level, I would like to thank all my family for their support, specially the most important person in my live, Vanessa Ortega, for her encouragement and endurance. For financial assistance I am indebted to several public projects and private contracts: COTS contract (Nortel Networks), SCALING contract (Tellabs Inc.), EU-FP7 BONE and SARDANA projects; Spanish MICINN projects TEC2008-01887 (TEYDE), RA4D and RAFOH; EU-FP6/7 E-Photon(+) and EuroFOS networks of excellence, and the MEC PTA-2003-02-00874 grant. vii
  • 6. Contents Abstract v Acknowledgements vii List of Figures xv List of Tables xxi Abbreviations xxiii Symbols xxv 1 Introduction 1 1.1 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.2 Complementary work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.3 Thesis overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2 State of the art 9 2.1 Modulation formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.2 Homodyne systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 2.3 PSK receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.3.1 Homodyne receiver performances . . . . . . . . . . . . . . . . . . 12 2.3.1.1 SNR and BER for BPSK signals . . . . . . . . . . . . . 14 2.3.1.2 Phase errors in homodyne detection of BPSK signals . . 16 2.3.1.3 SNR and BER for DPSK signals . . . . . . . . . . . . . 19 2.3.1.4 Phase errors in homodyne detection of DPSK signals . . 21 2.3.2 oPLL based systems . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.3.2.1 Additive noise impact in a generic OPLL . . . . . . . . . 23 2.3.2.2 Phase noise impact in a generic OPLL . . . . . . . . . . 25 2.3.2.3 Loop delay impact in a generic OPLL . . . . . . . . . . 26 2.3.2.4 Costas loop . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.3.2.5 Decision-Driven OPLL (DD-OPLL) . . . . . . . . . . . . 30 2.3.2.6 Balanced OPLL . . . . . . . . . . . . . . . . . . . . . . . 33 ix
  • 7. Contents x 2.3.2.7 Subcarrier modulated OPLL (SCM-OPLL) . . . . . . . 36 2.3.3 Phase and polarization diversity systems . . . . . . . . . . . . . . 38 2.3.3.1 Multiple differential detection . . . . . . . . . . . . . . . 38 2.3.3.2 Wiener filter phase estimation . . . . . . . . . . . . . . . 41 2.3.3.3 M-power law phase estimation with regenerative frequency dividers . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 2.3.3.4 Viterbi-Viterbi phase estimation . . . . . . . . . . . . . 45 2.4 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 3 Lock-In amplifier OPLL architecture 47 3.1 System model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 3.1.1 Loop analysis and linearization . . . . . . . . . . . . . . . . . . . 48 3.1.2 Noise, dithering and loop delay impacts . . . . . . . . . . . . . . . 52 3.1.3 Acquisition parameters . . . . . . . . . . . . . . . . . . . . . . . . 53 3.1.3.1 Hold in range . . . . . . . . . . . . . . . . . . . . . . . . 53 3.1.3.2 Pull in range . . . . . . . . . . . . . . . . . . . . . . . . 54 3.1.4 Data crosstalk and cycle slipping effects . . . . . . . . . . . . . . 54 3.2 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.2.1 Phase noise simulations . . . . . . . . . . . . . . . . . . . . . . . . 55 3.2.2 Time response simulations . . . . . . . . . . . . . . . . . . . . . . 57 3.2.3 Amplitude of the dithering signal . . . . . . . . . . . . . . . . . . 60 3.2.4 Comparison with other loops . . . . . . . . . . . . . . . . . . . . . 60 3.3 Experiments and discussion . . . . . . . . . . . . . . . . . . . . . . . . . 64 3.4 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 4 Advances in phase and polarization diversity architectures 69 4.1 Full phase diversity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 4.1.1 Karhunen-Lo`ve series expansion phase estimation . . . e . . . . . . 70 4.1.1.1 Receiver scheme . . . . . . . . . . . . . . . . . . . . . . 70 4.1.1.2 Phase estimation algorithm . . . . . . . . . . . . . . . . 71 4.1.1.3 Algorithm performances and discussion . . . . . . . . . . 72 4.2 Time switched phase / polarization diversity . . . . . . . . . . . . . . . . 74 4.2.1 Phase diversity combined with differential detection . . . . . . . . 74 4.2.1.1 Expected system performances . . . . . . . . . . . . . . 76 4.2.1.2 Simplified scheme and phase noise analysis . . . . . . . . 77 4.2.1.3 Frequency drift . . . . . . . . . . . . . . . . . . . . . . . 82 4.2.1.4 Channel spacing . . . . . . . . . . . . . . . . . . . . . . 84 4.2.2 Fuzzy data estimation . . . . . . . . . . . . . . . . . . . . . . . . 90 4.2.2.1 Receiver scheme . . . . . . . . . . . . . . . . . . . . . . 91 4.2.2.2 Data estimation . . . . . . . . . . . . . . . . . . . . . . 91 4.2.2.3 System performances . . . . . . . . . . . . . . . . . . . . 94 4.2.3 Direct drive time switching . . . . . . . . . . . . . . . . . . . . . . 95 4.2.3.1 Receiver scheme . . . . . . . . . . . . . . . . . . . . . . 95 4.2.3.2 Phase noise analysis . . . . . . . . . . . . . . . . . . . . 98
  • 8. Contents xi 4.2.3.3 Frequency drift analysis . . . . . . . . . . . . . . . . . . 101 4.2.3.4 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . 102 4.2.3.5 Experiments . . . . . . . . . . . . . . . . . . . . . . . . 102 4.2.4 Searching for a polarization diversity . . . . . . . . . . . . . . . . 104 4.3 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 5 ONU and OLT architectures 111 5.1 Summary of techniques and issues to take into account . . . . . . . . . . 111 5.1.1 Phase noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 5.1.2 Polarization mismatch . . . . . . . . . . . . . . . . . . . . . . . . 112 5.1.3 Modulation techniques and Rayleigh backscattering . . . . . . . . 113 5.2 ONU and transceiver architectures . . . . . . . . . . . . . . . . . . . . . 114 5.2.1 Transceivers based in a full phase diversity scheme . . . . . . . . 114 5.2.1.1 Transceiver with 90 degree hybrid and digital processing 114 5.2.1.2 Transceiver with 90 degree hybrid and analog processing 115 5.2.1.3 Transceiver including 90 degree hybrid and PBS, with dig- ital processing . . . . . . . . . . . . . . . . . . . . . . . 115 5.2.1.4 Transceiver including 90 degree hybrid and PBS, with analog processing . . . . . . . . . . . . . . . . . . . . . . 116 5.2.2 Transceivers based in time-switching phase diversity . . . . . . . . 117 5.2.2.1 Transceiver including phase switch with digital processing and standard balanced detector . . . . . . . . . . . . . . 117 5.2.2.2 Transceiver including phase switch with analog processing and standard balanced detector . . . . . . . . . . . . . . 117 5.2.2.3 Transceiver including direct laser switching with digital processing and standard balanced detector . . . . . . . . 117 5.2.2.4 Transceiver including direct laser switching with analog processing and standard balanced detector . . . . . . . . 118 5.2.3 Transceiver based in Optical Phase-Locked Loop . . . . . . . . . . 119 5.2.3.1 Transceiver with OPLL and analog processing . . . . . . 119 5.2.4 Transceiver comparison . . . . . . . . . . . . . . . . . . . . . . . . 120 5.3 OLT architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 5.4 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 6 Network topologies 125 6.1 Pure coupler splitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 6.2 Subband WDM tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 6.3 Advanced: WDM ring-tree SARDANA network . . . . . . . . . . . . . . 127 6.4 Case studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 6.4.1 Subband WDM tree PON . . . . . . . . . . . . . . . . . . . . . . 128 6.4.2 Ring-tree ultra-dense WDM PON . . . . . . . . . . . . . . . . . . 130 6.5 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 7 Conclusions and future work 137 7.1 General conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
  • 9. Contents xii 7.2 Future lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 7.2.1 Compact coherent transceiver . . . . . . . . . . . . . . . . . . . . 139 7.2.2 Full bidirectionality over a single fiber . . . . . . . . . . . . . . . 139 7.2.3 Spectrum management . . . . . . . . . . . . . . . . . . . . . . . . 140 A Passive optical network solution using a subcarrier multiplex 141 A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 A.2 Receiver scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 A.3 Experiments and discussion . . . . . . . . . . . . . . . . . . . . . . . . . 143 A.3.1 Downstream characterization . . . . . . . . . . . . . . . . . . . . 143 A.3.2 Full-duplex measurements . . . . . . . . . . . . . . . . . . . . . . 145 A.4 Network measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 A.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 B Automatic wavelength control design 151 B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 B.2 Loop design and performances . . . . . . . . . . . . . . . . . . . . . . . . 152 B.3 Practical implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 B.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 C Static and dynamic wavelength characterization of tunable lasers 159 C.1 Experiments and discussion . . . . . . . . . . . . . . . . . . . . . . . . . 159 C.1.1 Static characterization: wavelength map . . . . . . . . . . . . . . 160 C.1.1.1 Static characterization setup . . . . . . . . . . . . . . . . 160 C.1.1.2 Static characterization results . . . . . . . . . . . . . . . 160 C.1.2 Dynamic characterization . . . . . . . . . . . . . . . . . . . . . . 162 C.1.2.1 Dynamic characterization setup . . . . . . . . . . . . . . 163 C.1.2.2 Dynamic characterization results . . . . . . . . . . . . . 163 C.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 D Phase noise digital modeling 169 E Lock-In OPLL prototype scheme and printed circuit board 173 F Research publications 177 F.1 Patents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 F.2 Book contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 F.3 Journal publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 F.4 Conference publications . . . . . . . . . . . . . . . . . . . . . . . . . . . 178 F.5 Submitted publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 F.5.1 Book contributions . . . . . . . . . . . . . . . . . . . . . . . . . . 180 F.5.2 Journal publications . . . . . . . . . . . . . . . . . . . . . . . . . 180 F.5.3 Conference publications . . . . . . . . . . . . . . . . . . . . . . . 180
  • 10. Contents xiii Bibliography 181
  • 11. List of Figures 1.1 Nielsen’s law prediction of bandwidth and data obtained until 2006 (square points). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2 FTTH access roadmap. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2.1 Coherent receiver scheme, using balanced photo-detection. . . . . . . . . 10 2.2 Optical spectrum of a wavelength to the user environment. λLO is the nominal wavelength of the local oscillator, for a homodyne case. . . . . . 11 2.3 Comparison between homodyne and heterodyne electrical spectra. . . . . 11 2.4 Generic homodyne receiver. . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.5 Constellation representation of a BPSK signal in the I and Q plane. . . . 14 2.6 Bit error probabilities for BPSK and DPSK, as a function of SNR. . . . . 16 2.7 BPSK error probability for different phase error standard deviations. . . 17 2.8 BER-floor as a function of φe standard deviation. . . . . . . . . . . . . . 18 2.9 Generic homodyne receiver including a differential decoder. . . . . . . . . 19 2.10 Optical Phase Locked Loop simplified scheme . . . . . . . . . . . . . . . 23 2.11 Iso-curves of √ variance of additive noise (left) and phase noise (right), the all for ξ = 1/ 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 2.12 PLL parameters optimization for 1 ns loop delay and 1 MHz linewidth. . 27 2.13 Iso-curves of the variance of additive noise and phase noise, all for ξ = 2. (a-b) are for a null loop delay, whereas (c-d) are for a 1 ns loop delay. . . 28 2.14 Costas PLL scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 2.15 Decision driven PLL scheme. . . . . . . . . . . . . . . . . . . . . . . . . . 31 2.16 Balanced PLL scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.17 Balanced PLL phasor scheme. . . . . . . . . . . . . . . . . . . . . . . . . 34 2.18 Noise variance for the balanced PLL scheme. . . . . . . . . . . . . . . . . 36 2.19 General scheme for a subcarrier decision driven optical phase-locked loop. 36 2.20 Scheme of a phase diversity front end. . . . . . . . . . . . . . . . . . . . . 38 2.21 Schematic of phase and polarization diverse receiver. . . . . . . . . . . . 39 2.22 Scheme of a DPSK detection, in a phase and polarization diversity homo- dyne receiver. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.23 LMS error for a Wiener filter with a lag of 10 symbols. . . . . . . . . . . 43 2.24 Scheme of a phase estimator for polarization multiplexed QPSK signals based in regenerative frequency dividers. . . . . . . . . . . . . . . . . . . 44 3.1 Voltage after balanced detector (V3 (t)) as a function of the phase error (φS (t) − φLO (t)). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 xv
  • 12. List of Figures xvi 3.2 Lock-In amplified oPLL schematic. . . . . . . . . . . . . . . . . . . . . . 49 3.3 Spectral distribution of the terms 3.14, 3.15, and 3.16. . . . . . . . . . . . 50 3.4 Phase noise evolution and phase signal introduced by the loop. Inset (b) is a zoom between 200 ns and 300 ns. . . . . . . . . . . . . . . . . . . . . 56 3.5 Loop natural frequency versus damping factor relationship for optimal con- figurations (transient response and phase noise) with 10 ns main loop delay. 57 3.6 BER-floor for optimal configurations as a function of the laser linewidth evaluated at several main loop delays. . . . . . . . . . . . . . . . . . . . . 58 3.7 OPLL time response for a phase step of 1 rad. Inset figure is a zoom between 500 ns and 550 ns. . . . . . . . . . . . . . . . . . . . . . . . . . 59 3.8 Setting time of the optimal configurations for several loop main delays. . 59 3.9 Rise time of the optimal configurations for several loop main delays. . . . 60 3.10 Maximum overshoot of the optimal configurations for several loop delays. 61 3.11 Phase dithering effect for large loop delays. . . . . . . . . . . . . . . . . . 61 3.12 Phase error deviation evaluated at a loop delay of 10 ns. . . . . . . . . . 62 3.13 Pull in margins of the simulated architectures. . . . . . . . . . . . . . . . 63 3.14 Hold in margins of the simulated architectures. . . . . . . . . . . . . . . . 63 3.15 Experimental Setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 3.16 Electrical power spectrum after photodetection. . . . . . . . . . . . . . . 65 3.17 Electrical power spectrum after photodetection. . . . . . . . . . . . . . . 66 4.1 Scheme for a standard intradyne receiver. . . . . . . . . . . . . . . . . . . 70 4.2 Phase error deviation as a function of time interval squared per spectral width product (T 2 ∆ν). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.3 Block diagram of the phase estimation algorithm. . . . . . . . . . . . . . 73 4.4 Phase error deviation as a function of the spectral width per bitrate ratio. 73 4.5 Time-switched diversity differential homodyne receiver scheme. . . . . . . 75 4.6 Example of the time diversity operation, from scheme shown in figure 4.5. Blue line is Vouti , green line is Voutq and red line is Vout after filtering. . . 76 4.7 I, Q, and I+Q outputs Eye-diagrams, at 50 MHz total laser linewidth. . 77 4.8 Statistical normalized eye opening (20Log) for the I/Q receiver (both first and second approach) and a lock-in oPLL. . . . . . . . . . . . . . . . . . 77 4.9 Statistical normalized eye-opening (20log) for the I/Q receiver (both first and second approach) as a function of the laser frequency drift. . . . . . 78 4.10 Receiver scheme for phase noise analysis. . . . . . . . . . . . . . . . . . . 78 4.11 BER-floor of several cases: theoretical (dashed line), theoretical but includ- ing the penalty due to phase switching (dotted line), numerical simulation (continuous line) and measurements (square points). . . . . . . . . . . . . 80 4.12 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 4.13 Sensitivity results and output eye-diagram . . . . . . . . . . . . . . . . . 82 4.14 Modeled BER as a function of the laser frequency drift per bitrate ratio. 83 4.15 Measured BER as a function of the laser frequency drift. . . . . . . . . . 84 4.16 Time-Switched Phase-Diversity DPSK receiver for channel spacing study. 85 4.17 g1 (t) and g2 (t) pulse shapes and autocorrelation of g2 (t), R2 (τ ) . . . . . . 86
  • 13. List of Figures xvii 4.18 Spectrum after photodetection: Ideal homodyne reception (a) and using time-switched phase-diversity (b) . . . . . . . . . . . . . . . . . . . . . . 87 4.19 Complex representation of signal samples including interference. . . . . . 88 4.20 Sensitivity penalty due to channel crosstalk. Square points are experimen- tal data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 4.21 Experimental setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 4.22 Receiver scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4.23 IQ plane data plotting without differential decoding (left), and after dif- ferential decoding (right) for a signal corrupted by a phase noise due to 100 kHz of total laser linewidth . . . . . . . . . . . . . . . . . . . . . . . 92 4.24 I and Q components membership functions . . . . . . . . . . . . . . . . . 93 4.25 Data estimation scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 4.26 BER-floor as a function of the laser linewidth at 1 Gb/s . . . . . . . . . 95 4.27 Generic receiver module . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 4.28 BER floor versus the linewidth per bitrate ratio . . . . . . . . . . . . . . 97 4.29 Differential BPSK receiver√ scheme . . . . . . . . . . . . . . . . . . . . . . 97 4.30 Bessel coefficients for γ = 2 . . . . . . . . . . . . . . . . . . . . . . . . 99 4.31 Comparison between decision on Id (t) (using delay-and-add, DAD) and Im (t) (NDAD). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 4.32 Maximum tolerated linewidth per bit rate ratio at BER 10−3 as a function of the gain factor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 4.33 Receiver sensitivity for several configurations. . . . . . . . . . . . . . . . 103 4.34 Experimental setup for the direct drive time-switching. . . . . . . . . . . 104 4.35 SNR factor penalty at 10−3 BER vs gain factor γ. . . . . . . . . . . . . . 105 4.36 SNR factor penalty at 10−3 BER vs frequency drift. . . . . . . . . . . . . 105 4.37 I, Q, H, V time distribution of each bit . . . . . . . . . . . . . . . . . . . 106 4.38 Intradyne differential receiver with polarization and phase diversity. . . . 106 4.39 Alternative implementation for achieving time-switched phase and polar- ization diversities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 5.1 Transceiver with 90◦ hybrid and digital processing. . . . . . . . . . . . . 114 5.2 Transceiver with 90◦ hybrid and analog processing. . . . . . . . . . . . . 115 5.3 Digital configuration scheme using 90◦ hybrid combined with PBS. . . . . 116 5.4 Analog configuration scheme using 90◦ hybrid combined with PBS. . . . 116 5.5 Digital configuration scheme using phase switch. . . . . . . . . . . . . . . 117 5.6 Analog configuration scheme using phase switch. . . . . . . . . . . . . . . 118 5.7 Digital configuration scheme using standard balanced detector. . . . . . . 118 5.8 Analog configuration scheme using standard balanced detector. . . . . . . 119 5.9 Analogue configuration scheme for the oPLL transceiver prototype. . . . 119 5.10 OLT scheme with double fiber and including the birefringent polarization switch. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 5.11 OLT scheme with double fiber and including the FRM based polarization switch. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 6.1 Pure coupler splitting network scheme. . . . . . . . . . . . . . . . . . . . 126
  • 14. List of Figures xviii 6.2 Network scheme and routing profile. . . . . . . . . . . . . . . . . . . . . . 126 6.3 SARDANA network architecture. . . . . . . . . . . . . . . . . . . . . . . 127 6.4 OLT and CPE transmission modules. . . . . . . . . . . . . . . . . . . . . 129 6.5 Up-and Down-stream transmission results. . . . . . . . . . . . . . . . . . 130 6.6 Sensitivity penalty as a function of channel spacing. . . . . . . . . . . . . 130 6.7 Network topology and wavelength plan. . . . . . . . . . . . . . . . . . . . 131 6.8 Central office scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 6.9 Experimental network testbed . . . . . . . . . . . . . . . . . . . . . . . . 133 6.10 Sensitivity results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 A.1 Half-duplex experimental setup. . . . . . . . . . . . . . . . . . . . . . . . 143 A.2 Low pass equivalent of the mixer’s response for a 5 GHz carrier. . . . . . 144 A.3 Sensitivity results for setup described on figure A.1 . . . . . . . . . . . . 144 A.4 Downstream power penalty at BER 10−10 due to extinction ratio. Square points are experiments, whereas continuous line is derived from Eqs. 1 and 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 A.5 Experimental setup for single fibre full-duplex measurements. . . . . . . . 146 A.6 Electrical power spectrums after photo-detection at the receiver side: (a) before electrical filtering at the ONU, (b) after electrical filtering at the ONU; (c) before electrical filtering at the OLT, and (d) after electrical filtering at the OLT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 A.7 Sensitivity results for the proposed OLT and ONU architectures. . . . . . 147 A.8 Scenario 1 schematic. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 A.9 Downstream sensitivity curves for the three different network scenarios. . 148 A.10 Upstream sensitivity curves for the three network scenarios. . . . . . . . 149 A.11 Schematic of scenario 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 A.12 Scheme for scenario 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 B.1 Scheme of the proposed analog frequency estimation loop. . . . . . . . . 152 B.2 Optical SSB-modulation VCO. . . . . . . . . . . . . . . . . . . . . . . . . 152 B.3 Frequency discriminator output vs. frequency difference between LO and received signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 B.4 Loop delay impact on loop setting time. . . . . . . . . . . . . . . . . . . 154 B.5 Error signal variance vs. laser linewidth. . . . . . . . . . . . . . . . . . . 155 B.6 Schematic to be implemented. . . . . . . . . . . . . . . . . . . . . . . . . 155 B.7 Max hold function for the output spectrum of the optical VCO. . . . . . 156 B.8 Experimental setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 C.1 Experimental setup for stability regions characterization. . . . . . . . . . 160 C.2 (a) Wavelength map: Plot of the wavelength (colour scale) in function of reflector currents. (b) Logic stable regions map in function of reflector currents. The phase current for a) and b) is Iph = 2.4 mA. . . . . . . . . 161 C.3 (a) Plot of the wavelength in function of the phase current. Reflector cur- rents are biased at Iref 1 = 22.8 mA and Iref 2 = 8.6 mA. (b, c) Wavelength region map as a function of both reflector currents for a phase current of 1.8 and 2.2 mA, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . 162
  • 15. List of Figures xix C.4 Plots of the wavelength as a function of the gain current for different re- flector currents (Iph = 2.4 mA): (a) Ir ef 1 = 10.8 mA, Iref 2 = 29 mA; (b) Iref 1 = 12.4 mA, Iref 2 = 8.9 mA; (c) Iref 1 = 10.2 mA, Iref 2 = 11.9 mA. . 162 C.5 Experimental setup for transient response characterization. . . . . . . . . 163 C.6 (a) Stable regions map for Iph = 2.4 mA. The black points denote working points used to measure the transition between two modes. The white lines denote such transitions, and the number is used as experiment identifier. (b) Voltage versus time plot of the signals driving reflector sections for experiment 4 (see table C.1). . . . . . . . . . . . . . . . . . . . . . . . . . 164 C.7 (Id.a) WPT plot: Plot of the wavelength versus time, and power (gray scale) versus both wavelength and time for experiment ’Id’ (see table C.1 and/or figure C.6 (a)). (Id.b) SMSR versus time plot for experiment ’Id’ (see table C.1 and/or figure C.6 (a)). . . . . . . . . . . . . . . . . . . . . 165 C.8 (a) WPT plot: Plot of the wavelength versus time, and power (logarithmic colour scale) versus both wavelength and time for experiment 4. (b) Main mode and secondary mode power versus time (in logarithmic scale). . . . 166 C.9 (a) WPT plot zoom of experiment 5: Plot of the wavelength versus time, and power (logarithmic colour scale) versus both wavelength and time. (b) Wavelength versus time of the main and secondary modes of depicted in (a). (c) Zoom of (b) during the transition between inter-mode (1539.8 nm) and mode 2 (1545.2 nm). . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 D.1 Phase of phase noise spectrum. . . . . . . . . . . . . . . . . . . . . . . . 172 E.1 Printed circuit board outline of the Lock-IN OPLL prototype. . . . . . . 173
  • 16. List of Tables 2.1 Common modulation formats and their SNR differences. . . . . . . . . . 9 2.2 Comparison between BER values, the standard deviation of the phase error process for 1 dB penalty at such BER, and the BER-floor for that standard deviation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 2.3 Comparison of phase estimation methods. . . . . . . . . . . . . . . . . . 44 3.1 Phase error standard deviation for the optimal configurations as a function of linewidth and delay. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 3.2 Convergence values for setting and rise times, at several loop delays. . . . 60 3.3 Table summarizing results at 10 ns delay. . . . . . . . . . . . . . . . . . . 64 3.4 Measured values of the local oscillator linewidth. . . . . . . . . . . . . . . 64 4.1 Fuzzy logic estimator rules base. . . . . . . . . . . . . . . . . . . . . . . . 93 5.1 Phase noise cancellation techniques summary table. The linewidth toler- ance is for a 10−3 BER-floor, whereas the penalty is respect to an ideal system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 5.2 Polarization handling methods summary table. . . . . . . . . . . . . . . . 113 5.3 Transceiver architectures summary table. The linewidth tolerance is for 1 dB penalty at 10−10 BER, whereas the penalty is respect to an ideal system.120 6.1 Power budget summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 B.1 Comparison between possible optical VCO approaches. . . . . . . . . . . 156 C.1 The acronyms read in ”kind of transition” column, have a brief explana- tion of the working points location: InM (Inside the same Mode); CM (Consecutive Modes in the same super-mode); NCM (Non-Consecutive modes in the same super-mode); CS (Consecutive Super-modes); NCS (Non-Consecutive Super-modes); Iph (change in phase current). . . . . . 167 xxi
  • 17. Abbreviations ADC Analog to digital converter AFC Automatic frequency control ASK Amplitude shift keying AWC Automatic wavelength control AWG Arrayed waveguide grating BER Bit error ratio BPF Band pass filter BPSK Binary phase shift keying CO Central office CPE Customer premises equipment DAC Digital to analog converter DD Direct detection DFB Distributed feedback DPSK Differential phase shift keying ECL External cavity laser ER Extinction ratio FEC Forward error correction FRM Faraday rotator mirror FTTH Fiber to the home FWHM Full width half maximum GCSR Grating-coupled sampled reflector GPON Gigabit-capable passive optical network I In-phase IM Intensity modulation LD Laser diode LMS Least Mean Square LPF Low pass filter MAP Maximum a posteriori MG-Y modulated-grating Y-branch (laser) MZI Mach-Zehnder interferometer xxiii
  • 18. Abbreviations xxiv MZM Mach-Zehnder modulator NRZ Non return to zero OLT Optical line termination ONU Optical network unit OPLL Optical phase-locked loop OSA Optical spectrum analyzer OSNR Optical signal to noise ratio OSRR Optical signal to Rayleigh backscattering ratio PBS Polarization beam splitter PI Proportional integral PLC Planar lightwave circuit PLL Phase-locked loop PM Phase modulator PON Passive optical network PPG Pulse pattern generator PRBS Pseudo-random bit sequence PSD Power spectrum density PSK Phase shift keying Q Quadrature QPSK Quadrature phase shift keying RB Rayleigh backscattering RMS Root mean square RN Remote node RZ Return to zero SCM Sub-carrier modulation SIR Signal to interference ratio SNR Signal to noise ratio SOP State of polarization SSB Single side band TDM Time division multiplexing UD Ultra-Dense VCO Voltage controled oscillator VOA Variable optical attenuator WDM Wavelength division multiplexing
  • 19. Symbols ∆ν Laser linewidth ∆f Signal bandwidth F Electronic receiver equivalent noise factor Fa Excess noise factor Id Photodiode dark current k Boltzmann constant m Modulation index M Multiplication factor of the APD PS Optical received power from transmitter PLO Optical power from local q Electron charge Photodiode responsivity Rb Bit rate RL Load resistor Rx Receiver part T Room temperature Tb Bit time Tx Transmitter part xxv
  • 20. To my family. . . xxvii
  • 21. Chapter 1 Introduction More than 40 years have passed since Charles K. Kao publicly demonstrated the pos- sibility of transmitting information through optical fibers [1]. During this time, optical networks have evolved from being an entelechy to a reality that sustains and makes possi- ble the information society in which we live. In recognition, Kao received the 2009 Nobel prize in physics for the groundbreaking achievements concerning the transmission of light in fibers for optical communication. Actually, the concept of optical access networks is very wide and includes many ap- proaches. One of the most popular is the so-called Passive Optical Network (PON) [2], due to its flexibility and low requirements. Typically a PON has a point to multipoint topology, establishing connection between a remote network terminal (Optical Line Ter- mination, OLT) and the customer premises, where an Optical Network Unit (ONU) is placed. When looking at the tendencies of optical access networks, one can realize that user bit rate demand is expected to be increasing in the near future, mostly due to triple-play services and advanced multimedia applications. Precisely, in 1998 Jakob Nielsen predicted that average bandwidth per user gets incremented in a 50 % per year. Until now it has been accomplished and, in case this law is followed, in 2020 each user will demand to get at home an average bandwidth of 1 Gb/s. This makes completely obsolete the technologies commercially available nowadays, and current Fiber To The Home (FTTH) techniques will may get obsolete in long term, being replaced by emerging FTTH technologies. When looking at the several techniques available to upgrade existing access networks, a roadmap can be drawn, shown in figure 1.2. Under the time point of view, now is the deployment of FTTH, point to point (PtP) or GPON/EPON standards. Neverthe- less, PON standardization bodies are pushing technology towards higher FTTH capacity systems, mostly by increasing the aggregate bit rate. Precisely, the IEEE has recently completed and launched the 10G-EPON P802.3av and the FSAN has the NGPON1 rec- ommendation for 10 Gb/s (also named as XGPON) well advanced. In these systems, the 1
  • 22. Chapter 1. Introduction 2 Figure 1.1: Nielsen’s law prediction of bandwidth and data obtained until 2006 (square points). guarantied effective bandwidth per user will be about 150-300 Mb/s, as the bit rate is shared among e.g. 32 users. These first next generation PONs only encompass a line rate increase (down/up), not yet deployed, and not much is defined in about using WDM technology, which is left for a longer term generation of PONs (like NGPON2), mainly due to the fact that there are several technical hurdles in WDM technologies for PONs, as the ONU colourlessness, the wavelength stability and the cost. So, by increasing the bit rate to 10 Gb/s, the ONU at the CPE is expected to operate at a very high bit rate in the opto-electronics transceivers just to use a small fraction of it. If that is considered in fast electronics (e.g. in CMOS ASICs) the power consumption is almost proportional to the clock speed, one can infer that there is a huge power inefficiency corresponding to the user bandwidth inefficiency; leading to a substantial global power waste. To reverse this tendency, it is obvious that some new philosophy has to be investigated with the corresponding technology challenges. The answer proposed is to try to exploit the pure WDM dimension while minimizing the electronics speed, and maintaining the global numbers unchanged: • Number of served users per PON (in the order of magnitude of 1000). • Guarantied bandwidth per user. If today’s goal is to serve 100 Mb/s, a next step in longer term it can be up to 1 Gb/s; for example current personal computers are nowadays including a 10/100/1000 MEthernet interface, thus 1GEthernet can be considered a very practical goal.
  • 23. Chapter 1. Introduction 3 • Total fiber bandwidth (40 nm ≈ 5 THz in C-band); although by leaving save guard- bands, and normal modulation formats, only about 1 Tb/s is used in normal prac- tise. Figure 1.2: FTTH access roadmap. A very ultra dense WDM network, with a few GHz channel spacing (below 5 GHz), would be ideal for the numbers presented. With this very narrow channel spacing, many optical carriers could be accommodated on a single fiber and a large number of users could be connected to the network, each of them having an exclusive wavelength. Nevertheless, the major challenge for such networks are the huge technical requirements listed above. The main enabling technology for the proposed network philosophy, capable to reach the presented goals, is coherent transmission. It received great attention at the late 80s and beginning of the 90s, and after a certain period of latency, it has been resurrected. It presents many advantages with respect to the conventional direct detection systems like its excellent wavelength selectivity, low sensitivity and tunability performances [3]. However, it was mainly focused towards long-haul WDM applications, but not seriously considered to be used in access PONs. As these networks have multiple low capacity channels, a major concern in direct-detection (DD) based systems, is the use of optical filters in order to delimitate these channels mainly because of its low selectivity at the GHz spacing scale. Thus, for a very narrow spaced channels, a coherent receiver using electrical filtering is a promising way to solve the problem. Heterodyne optical receivers can be a first approach [4, 5], but due to its inherent image frequency interference, a best solution is homodyne reception. In such homodyne systems, the reception part has a local laser that oscillates at the same wavelength as the received signal. In a second stage both signals are optically mixed and photo-detected. Afterwards, signal processing (analog or digital) is applied to the electrical signal in order to recover transmitted data. The improvements are clear, with respect to other options:
  • 24. Chapter 1. Introduction 4 • Allows the use of advanced modulation formats (like Phase Shift Keying - PSK, or OFDM), while extending the reach of the networks. • Uses electrical filtering for channel selection, achieving narrow channel spacing while maintaining high speed connection. • Concurrent detection of light signal’s amplitude, phase and polarization recovering more detailed information to be conveyed and extracted, thereby increasing toler- ance to network impairments (such as chromatic dispersion) and improving system performance. • Linear transformation of a received optical signal to an electrical signal that can then be analyzed using modern DSP technology. • Local laser can be tuneable, allowing colourless operation, and it can be reused as an optical source for data transmission. • An increase of receiver sensitivity by 15 to 20 dB compared to incoherent systems. So, homodyne systems match perfectly the proposed network requirements, though some issues like transceivers’ cost have to be addressed. Summarizing, with a coherent transceiver at both sides of the access link, the capabilities can be extended to: • High density, enabling the connection to a high number of users (more than 1000 users per output fiber), meaning narrow channel spacing. • High transmission speed, guaranteeing a minimum bandwidth of 1 Gb/s per user. • External network totally passive, with no insertion of any type of equipment that could include an electrical supply at the external plant (optical distribution net- work). • High power budget, for maintaining a standard central office output power, a low sensitivity receiver has to be implemented, reaching less than −30 dBm. • Highest ONU bandwidth efficiency, with lowest electronics requirements (1 GHz BW), serving every user with the 1G Ethernet LAN standard. • High optical spectral efficiency, by minimizing the wavelength channel spacing below 5 GHz only. • Low power consumption ONUs, reducing it in about one order of magnitude. • Transparency and Independence among channels, in terms of coding, protocol and bit rate, thus avoiding the complex synchronization and ranging of current PONs.
  • 25. Chapter 1. Introduction 5 1.1 Objectives The objective of this thesis is to evaluate and propose advanced OLT and ONU archi- tectures based on coherent systems for access network deployments. The idea is not to restrict to the receiver architecture itself, but also evaluate the uplink and downlink per- formances of the network in order to find the most effective solution. Specifically the objectives of the thesis are the following: • Identify current coherent systems architectures. Perform an study of the state of the art analyzing the main coherent technologies that are currently being investigated. • Propose advanced architectures that overcome the limitations of the existing ones and fit with the specifications of passive optical networks. • Evaluate some of the advanced techniques by means of simulations and experiments: – Optical Phase-Locked Loops: Costas, Decision-driven, Balanced subcarrier and Lock-In amplified loops. – Phase diversity receivers with zero intermediate frequency: Phase estimation algorithms and differential receivers. • Implement a fully working transceiver prototype of the most reliable and cost- effective architecture. • Research the published work on advanced access network architectures and propose network reuse scenarios to achieve the desired ultra dense WDM operability. • Experimentally demonstrate the performances of the transceiver prototype in the more promising network schemes. 1.2 Complementary work As a complementary work to the accomplishment of the present thesis, other studies have been carried out: IM-DD transmission systems using subcarrier multiplexing, design and study of an automatic frequency control for coherent systems, and tunable laser transient characterization. These studies are understood to help obtaining a more comprehensive view of the concepts developed in the thesis, even if they are rather outside its scope. For the SubCarrier Multiplexed (SCM) system, the objective is to explore an alternative implementation for future PON deployments. Precisely, it is a bi-directional full-duplex 2.5 Gb/s / 1.25 Gb/s in a SCM single fiber PON. The downstream signal is DPSK coded and up-converted by using a 5 GHz subcarrier, while the upstream data is transmitted in
  • 26. Chapter 1. Introduction 6 burst-mode NRZ. A theoretical model for SCM downstream is proposed and experimen- tally validated. Furthermore, three different deployment scenarios are evaluated: Large coverage area and low density of users; area with medium density of users; and improved access network, covering as much users as possible. For the last case, the power budget could be increased up to 29 dB, matching clearly the typical values of GPON deploy- ments, and serving up to 1280 users. A more detailed report of the system and the tests performed can be found in appendix A. Regarding the automatic frequency control (AFC), details can be found in appendix B. There it is shown how a simulation model was developed for a Cross Product AFC [6]. Parallel to that, a first prototype design was started and several key components (e.g. optical VCO) were identified and characterized, for building the full prototype. Finally, for assuring that everything was the right way, some proof-of-concept experiments were performed in an 8PSK-RZ 30 Gb/s transmission system. Last but not the least, through the high-resolution wavelength-power-time measurement, the dynamic behaviour of a tunable laser (a modulated-grating Y-branch, MG-Y) while switching between modes has been also characterized. A complete report on these mea- surements can be found in appendix C. The optical spectrum at every instant and its evolution along the tuning transient was obtained. With this, it was easy to identify, not only the wavelength temporal drift, but also the transitory mode hopping or interferences over other wavelength channels. 1.3 Thesis overview All the presented objectives and concepts will be explored and analyzed in the present document, which has been organized in 7 chapters. In chapter 2, the most important coherent technologies, that shape the actual scene, will be introduced. After a brief analysis of the coherent detection of BPSK and DPSK modulation formats, optical phase locked loops will be introduced and their influence on phase modulated signals detection will be evaluated. Next, the phase and polarization diversity concepts will be explained as well as the main techniques used in these receivers. Chapter 3 will put forward a new optical phase locked loop (OPLL), based in the lock- in amplification concept. There, the influence of noise will be analyzed, jointly with its associated penalties for a coherent receiver using phase modulated signals. Also, comparison will be performed between this new OPLL and the schemes presented in the state of the art. Chapter 4 will deal with some advances proposed towards an improved and lower cost phase/polarization diversity receiver. There new digital phase/data estimation methods
  • 27. Chapter 1. Introduction 7 will be described, and a step forward will be taken by proposing a novel coherent receiver type searching time-switched phase and polarization diversities. Chapter 5 describes a set of possible OLT and ONU designs. Special emphasis is put on the possible transceiver architectures, aiming to use the same design at both sides, OLT and ONU. Chapter 6 will give an overview of standard and advanced topologies for FTTx, driven by the concepts presented in this first chapter and taking into account the transceivers discussed in chapter 5. Afterwards, two case studies are presented demonstrating exper- imentally the two more promising network architectures. Finally, the conclusions chapter will summarize the work and present future research lines to continue developing this topic.
  • 28. Chapter 2 State of the art 2.1 Modulation formats The modulation format to be used in a network is strongly linked with the fact of how it will be generated at the transmitter side, and the type of reception. As an example, a table can be found, where SNR increments are depicted when switching from one modulation format to another [7]. This is shown in table 2.1. In that table, the modulation format that has better SNR performances is homodyne phase shift keying (PSK). Heterodyne Homodyne IM-DD ASK FSK PSK ASK PSK IM-DD - 10/25 dB 13/28 dB 16/31 dB 13/28 dB 19/34 dB ASK Het. −10/−25 dB - 3 dB 6 dB 3 dB 9 dB FSK Het. −13/−28 dB −3 dB - 3 dB 0 dB 6 dB PSK Het. −16/−31 dB −6 dB −3 dB - 3 dB 3 dB ASK Hom. −13/−28 dB −3 dB 0 dB −3 dB - 6 dB PSK Hom. −19/−34 dB −9 dB −6 dB −3 dB −6 dB - Table 2.1: Common modulation formats and their SNR differences. In the access networks that are being deployed today, the modulation format used is IM/DD due to its simplicity. However, its low SNR performances are a major inconvenient when regarding an extended reach access network. That is the reason why it would be preferable to use a more robust format, like PSK, and a coherent detection scheme. According to table 2.1, a minimum SNR increment of 19 dB is expected when migrating from IM/DD to a PSK with homodyne detection. Of course, it is not a fixed increment, as it also depends on the photodetector type. E.g. if a PIN diode is used, the receiver performances in IM-DD are going to be worse than when using an avalanche photodiode. 9
  • 29. Chapter 2. State of the art 10 2.2 Homodyne systems Nowadays, optical fibre communications are, in a certain sense, as primitive as radio communications when crystal (galena) radio receivers were used. The reason is that there is no need to recover phase information of the optical carrier. Among all, coherent optical transmission systems were investigated at the late 80s, but abandoned due to electronics limitations and the irruption of the EDFA at the beginning of the 90s. Almost 20 years after, technology is more advanced, allowing a full development of coherent systems. Coherent systems present many advantages with respect to the conventional direct de- tection systems because of its excellent wavelength selectivity and low sensitivity. First, in a WDM environment, when using a coherent receiver, channel selection is done after photo-detection, i.e. is done by an electrical filter (instead of an optical filter); thus, se- lectivity is defined by this filter performances. Regarding sensitivity, coherent reception allows to use PSK and other advanced modulation formats. This fact, combined with the use of a local oscillator, is the reason why they can improve sensitivity in 19 dB up to 34 dB, when compared to an Intensity-Modulation Direct-Detection (IM-DD) system [7]. Figure 2.1: Coherent receiver scheme, using balanced photo-detection. The main difference between DD and coherent systems, is that the received signal is mixed with a local laser in an optical coupler. Afterwards, the resulting combination is photo-detected. This is shown in figure 2.1. Current after photo-detection Ip (t) has all information carried by the received optical field. In this chapter, a review of the synchronous detection technology is presented. Depending on the use of an intermediate frequency stage, coherent systems can be homodyne or heterodyne. In a heterodyne system, incoming signal is downconverted into an intermediate frequency (usually higher than bit rate). Afterwards, in a second stage, signal is mixed with an electrical oscillator, now downconverting into a baseband signal. As signals are electrically synchronized inside intermediate frequency module, it is an interesting implementation of a synchronous receiver. Namely, it avoids the need of very narrow lasers. However, the problems are: • This Intermediate Frequency (IF) is very high, limiting the electronics functionality.
  • 30. Chapter 2. State of the art 11 • The electrical spectrum is doubled, thus introducing a 3 dB penalty. This is shown in figure 2.3. • An additional filter should be placed in order to avoid image frequency in a multi- channel environment. Figure 2.2: Optical spectrum of a wavelength to the user environment. λLO is the nominal wavelength of the local oscillator, for a homodyne case. Figure 2.3: Comparison between homodyne and heterodyne electrical spectra. A further simplification, at least at a first glance, is the use of homodyne systems. In such systems intermediate frequency is zero. This avoids image frequency problems and the 3 dB penalty. But it needs to directly synchronize local laser and received signals, entailing some handicaps: • Laser phase noise impact on overall receiver performances. • Penalty due to synchronization loop delay. Optical homodyne systems were presented at the 80s, when one of the main investigation fields was coherent systems. In order to properly synchronize local laser and received signals, early systems used an optical Phase-Locked Loop (OPLL) module. But the optical path between local laser and optical mixer (i.e. optical hybrid + photo-detection stages) introduces a non-negligible loop delay, resulting in a significant penalty. Thus, in order to avoid it, extremely low linewidth lasers had to be used.
  • 31. Chapter 2. State of the art 12 Another approach towards homodyne reception came later, with the concept of zero- IF/intradyne diversity receivers. The main goal of these type of receivers is to replace the feedback loop (OPLL) by a feedforward processing. So, phase locking is done inside this feedforward processing. 2.3 PSK receivers As shown in the introduction, the main core of a coherent system is the receiver. This subsystem, properly combined with a robust modulation format, improves the optical link as commented. This section is organized as follows: First, homodyne receivers are introduced and ba- sic results are summarized. Next OPLLs are introduced and the existing approaches developed are explained. Finally, optical diversity techniques are discussed. 2.3.1 Homodyne receiver performances In this subsection the basic results of an ideal homodyne receiver will be surveyed. First using Binary PSK modulation and afterwards using differential encoded PSK. Also the phase errors influence (mainly due to laser phase noise) will be theoretically evaluated for both cases. These modulation formats have been chosen because of their simplicity, robustness and high performances, as seen in table 2.1. A generic homodyne receiver can be shown in figure 2.4, for a balanced structure. Figure 2.4: Generic homodyne receiver. From that scheme, the following set of equations can be written [8]: eS (t) = PS exp j ω0 t + φS (t) (2.1) eLO (t) = PLO exp j ω0 t + φLO (t) (2.2)
  • 32. Chapter 2. State of the art 13 where eS (t) and eLO (t) are the optical field expressions for the received and local oscil- lator signals respectively; φS (t) and φLO (t) are the received and local oscillator phases respectively; and ω0 t is the nominal wavelength (assuming no mismatch). Also the complex amplitudes of both signals can be defined as: ES (t) = PS ejφS (t) (2.3) jφLO (t) ELO (t) = PLO e (2.4) By agreement the optical coupler is assumed to have the following transfer matrix: 1 1 1 S=√ (2.5) 2 1 −1 As the optical combining device is a standard coupler and ideally there is no wavelength mismatch, the resulting currents I1 (t), I2 (t) at the output of each photodetector can be expressed as: 2 1 I1 (t) = ES (t) + ELO (t) (2.6) 2 = (PS + PLO ) + PS PLO cos φS (t) − φLO (t) (2.7) 2 2 1 I2 (t) = − ES (t) + ELO (t) (2.8) 2 = (PS + PLO ) − PS PLO cos φS (t) − φLO (t) (2.9) 2 being the responsivity of the photodiode. Then, the resulting current after the balanced receiver Ip (t) can be written as: Ip (t) = I1 (t) − I2 (t) (2.10) = 2 PS PLO cos φS (t) − φLO (t) (2.11) The signal amplitude at regeneration is highly dependant on the phase mismatch φS (t) − φLO (t) that must be minimized. The most used module to do so is the OPLL. The fluctuation phase error mainly comes from the lasers phase noise.
  • 33. Chapter 2. State of the art 14 2.3.1.1 SNR and BER for BPSK signals One of the most important advantages of homodyne PSK systems is the increase in receiver sensitivity. For BPSK, the bits are coded into two symbols: 0 and 180. Thus, In-phase and Quadrature components of the coded signal are going to be as shown in figure 2.5. Please note that for the receiver proposed, the decision is made along the real (In-phase) axis. Figure 2.5: Constellation representation of a BPSK signal in the I and Q plane. When making a first analysis, the photodetected signal after balanced detection Ip (t) is going to be low-pass filtered by a matched filter [9] and, next, it enters at the decision and sampling stage. Thus, the bit decision is made upon Ip (t) once filtered. By now, it can be only assumed that the receiver current fluctuates because of photodetector’s shot noise (in case a PIN diode is used) and thermal noise. The variance of those current fluctuations is obtained by adding the two contributions [10]: σ 2 = σS + σT 2 2 (2.12) 2 σS = 2q (PS + PLO ) + ID BE (2.13) 2 4kB T σT = FN BE (2.14) RL where ID is the dark current of the photodiode (almost negligible), q is the electron charge, BE is the most limiting electrical bandwidth, kB is the Boltzmann’s constant, T is the temperature in K, FN is the noise figure of the electrical stage, and RL is the impedance of the electrical part.
  • 34. Chapter 2. State of the art 15 From this model, the SNR can be calculated when φS (t)−φLO (t) = 0 dividing the average signal power by the average noise power: I 2 SNR = (2.15) σ2 2 4 PS PLO = (2.16) 4kB T 2q (PS + PLO ) + ID BE + RL FN BE Assuming the symbols are equiprobable, the bit error probability Pe can be calculated as: 1 Pe = [P (0◦ |180◦ ) + P (180◦ |0◦ )] (2.17) 2 where P (0◦ |180◦ ) is the probability of deciding 0◦ when 180◦ is received, and P (180◦ |0◦ ) is the probability of deciding 180◦ when 0◦ is received. As shown in figure 2.5, the only change between 0◦ and 180◦ is the sign along the real axis, whereas the modulus remains constant. Thus, the optimum decision threshold is going to be 0 [10]. Simplifying the development and assuming Gaussian statistics, the conditioned probabilities can be written as [9]: 0 1 SNR P (0◦ |180◦ ) = √ exp − dI (2.18) σ 2π −∞ 2 ∞ ◦ ◦ 1 SNR P (180 |0 ) = √ exp dI (2.19) σ 2π 0 2 and they can be expressed in terms of the complementary error function (erfc): 1 SN R P (0◦ |180◦ ) = P (180◦ |0◦ ) = erfc (2.20) 2 2 So, the bit error probability Pe can be calculated as [10]: 1 SNR Pe = erfc (2.21) 2 2 Figure 2.6 shows how the error probability varies with the SNR. Usually, the receiver sensitivity corresponds to the average optical power for which SNR = 15.6 dB, being Pe = 10−9 . Another SNR useful value is 9.8, that corresponds to a Pe = 10−3 because if Forward Error Correction (FEC) codes are used, errors can be corrected after data decision, and this 10−3 can be turned on to 10−9 or lower [11].
  • 35. Chapter 2. State of the art 16 Figure 2.6: Bit error probabilities for BPSK and DPSK, as a function of SNR. 2.3.1.2 Phase errors in homodyne detection of BPSK signals In this subsubsection the phase noise influences on the BPSK ideal receiver are going to be evaluated. It is assumed that there is a phase tracking and/or estimation/cancellation in order to keep the phase errors sufficiently small. To start such analysis, the received photocurrent has to be redefined as: Ip (t) = 2 PS PLO cos φd (t) + φe (t) (2.22) φS (t) = φd (t) + φN S (t) (2.23) φLO (t) = φN LO (t) (2.24) φe (t) = φN S (t) − φN LO (t) (2.25) where φN S (t) and φN LO (t) are the noise contributions to the phases of the received and local oscillator signals respectively, and φd (t) is a signal containing data ideal pulses (0◦ -180◦ ). Obviously, the phase error term (φe (t)) is modeled as a random variable. For the BPSK case, its statistical properties depend on the phase tracking method. Previously, the error probability has been found in terms of SNR. The expression used assumes a perfect phase match, but usually there is a certain amount of phase error. As there is a phase tracking, it can be assumed that φe (t) varies at a speed much lower than data. i.e. it remains constant during the symbol interval [12]. In this case, the conditional error probability in terms of the sampled phase error φe is: 1 SNR Pe (φe ) = erfc cos(φe ) (2.26) 2 2
  • 36. Chapter 2. State of the art 17 whereas the average error probability is written as: π 1 SNR Pe = p(φe )erfc cos(φe ) dφe (2.27) 2 −π 2 being p(φe ) the probability density function of the phase error. The statistic of the phase is usually approximated by a Gaussian distribution with zero mean. In this case the average error probability becomes: π φ2e 1 − 2σ 2 SNR Pe = e φe erfc cos(φe ) dφe (2.28) 2 2 2πσφe −π 2 In figure 2.7 the error probability is plotted versus the SNR for several phase error stan- dard deviation values. As can be seen, the fact of having a phase error deviation different from zero gives an error floor, i.e., the error probability limit is a finite value. A useful example can be that the standard deviation of phase error must be less than 10◦ in order to maintain less than 0.5 dB power penalty at 10−9 BER. Figure 2.7: BPSK error probability for different phase error standard deviations. In the limit case of infinite SNR, equation 2.28 gives the floor value of the probability of error. It only depends on the variance of the phase error, and gives the limit value. After
  • 37. Chapter 2. State of the art 18 some algebra, such floor is found to be [13]: φ2e +∞ − φ2e 1 − 2σ 2 2 2σ 2 Pe = e φe dφe = e φe dφe (2.29) 2 2 π 2πσφe cosφe <0 2πσφe 2 Evaluating this integral, the BER-floor value can be easyly plotted and see how the BER is limited by phase tracking errors. This is depicted in figure 2.8, showing that a BER of 10−9 cannot be achieved when σφe is higher than 14.9◦ . Figure 2.8: BER-floor as a function of φe standard deviation. BER Standard deviation for 1 dB penalty BER-floor equivalent 10−9 11◦ 2.31 · 10−16 10−3 19◦ 2.04 · 10−6 4.86 · 10−6 14.9◦ 10−9 2.54 · 10−1 28◦ 10−3 Table 2.2: Comparison between BER values, the standard deviation of the phase error process for 1 dB penalty at such BER, and the BER-floor for that standard deviation. This BER-floor will be useful for evaluating the architectures to be discussed during the present thesis. Thus, it is appropriate to represent in table 2.2 a set of values that will be used later. The idea is to have the floor values (easy to find) and search the equivalent BER, for 1 dB penalty. For example, a BER of 10−9 has the 1 dB penalty point at a phase error standard deviation of 11◦ , which corresponds to a BER-floor of 2.31 · 10−16 .
  • 38. Chapter 2. State of the art 19 2.3.1.3 SNR and BER for DPSK signals In the case of differentially encoded PSK signals, the coherent detector becomes slightly different, as shown in figure 2.9. In some books it is referred as differentially coherent detector [12]. Special emphasis must be put on the multiplier used, as it should be a four quadrant multiplier. Also, the local oscillator does not have to be tracking the received signal phase, since this kind of detection is more robust against phase mismatch. Figure 2.9: Generic homodyne receiver including a differential decoder. As now the receiver front-end is the same as used in the previous subsection, the SNR expression is the same of equation 2.16. Nevertheless in this case the inputs to the multiplier during the kth bit interval are: Ip (t) + n(t) = [I + Ini (t)] cos(φd (t) − φe (t)) − Inq (t) sin(φe (t)) (2.30) Ip (t−Tb )+n(t−Tb ) = [I +Ini (t)] cos(φd (t−Tb )−φe (t−Tb ))−Inq (t) sin(φe (t−Tb )) (2.31) The low-pass filter then removes the high-frequency terms from the product, leaving at the input of the decision circuit the decision variable amplitude Y . In case the phase error difference between consecutive symbols is negligible (φe (t) ≈ φe (t − Tb )), Y can be written as: 1 1 Y = (I + Ini )(I + Ini ) + Inq Inq = (α2 − β 2 ) (2.32) 2 2 where all four noise components are independent identically distributed Gaussian random variables with zero mean and variance 2σ 2 . α2 and β 2 variables can be expressed as: α2 = (I + αi )2 + αq 2 (2.33) 2 β = βi2 + 2 βq (2.34)
  • 39. Chapter 2. State of the art 20 with 1 αi = (Ini + Ini ) (2.35) 2 1 αq = (Inq + Inq ) (2.36) 2 1 βi = (Ini − Ini ) (2.37) 2 1 βq = (Inq − Inq ) (2.38) 2 Note that αi , αq , βi , βq are zero-mean Gaussian random variables with variance σ 2 . There- fore, α has a Rician probability density function, whereas β has a Rayleigh probability density function [14]. The average probability of error is found to be when Y < 0, in the case that the consec- utive symbols (ak , ak−1 ) are equal: Pe = P (Y < 0|ak = ak−1 ) = P (α2 < β 2 ) = P (β > α) (2.39) So, it can be calculated in a more direct form as: ∞ ∞ Pe = pα (α)pβ (β)dαdβ (2.40) 0 α where pα (α) and pβ (β) are the probability density functions of α and β respectively. Calculating the inner integral (β), the probability of error becomes: ∞ α 2α2 + I 2 2Iα Pe = exp I0 dα (2.41) 0 σ2 σ2 σ2 √ √ So, making a change of variables by letting λ = 2α and ν = I/ 2: ∞ 1 −I 2 λ λ2 + ν 2 λν Pe = exp exp I0 dλ (2.42) 2 2σ 2 0 σ2 2σ 2 σ2 Now, the integrand is exactly the same function as the Rician probability function, with a total area equal to unity. Hence, the final result becomes: 1 −I 2 1 −SNR Pe = exp = exp (2.43) 2 2σ 2 2 2 Just for comparing both signaling cases, figure 2.6 shows the two error probabilities (BPSK and DPSK) as a function of SNR. Even for a Gaussian noise assumption, they
  • 40. Chapter 2. State of the art 21 exhibit different statistics when calculating Pe . Nevertheless, note that at 10−9 the dif- ference between them is of only 0.5 dB. 2.3.1.4 Phase errors in homodyne detection of DPSK signals Just following what has been shown for the BPSK case, the expression reported in equa- tion 2.22 can also be used. For the DPSK case, φe statistical properties depend on the phase noise source. If it is only coming from the lasers’ phase noise, it can be assumed that φe (t) varies at a speed much lower than data. i.e. it remains constant during the symbol interval. Please, remember that phase noise is always of the order of MHz, while data is supposed to be of the order of Gb/s (3 orders of magnitude difference). In this case, the conditional error probability in terms of phase error is: 1 −SNR Pe (θ) = exp cos2 (θ) (2.44) 2 2 where θ = φe (t0 ) − φe (t0 − Tb ), being t0 the optimum sampling time. So, now the average error probability can be written as: π 1 −SNR Pe = p(θ) exp cos2 (θ) dθ (2.45) 2 −π 2 Regarding θ statistics, the laser phase noise is modeled as a Wiener process [10]: t φe (t) = φP N (τ )dτ (2.46) 0 where φP N (t) is a white Gaussian process with variance 2π∆ν, where ∆ν is the total laser spectral width (also known as Full Width Half Maximum - FWHM). Thus, assuming Tb t0 : θ = φe (t0 ) − φe (t0 − Tb ) (2.47) t0 t0 −Tb = φP N (τ )dτ − φP N (τ )dτ (2.48) 0 0 Tb = φP N (τ )dτ (2.49) 0 2 This means that θ is also a Gaussian process [14] with zero mean and variance σθ = 2π∆νTb .
  • 41. Chapter 2. State of the art 22 Continuing the mathematical development, the average error probability becomes: π θ2 1 − 2σ 2 −SNR Pe = e θ exp cos2 (θ) dθ (2.50) 2 2 2πσθ −π 2 Please note that this expression is almost the same that has been found in the previous subsection (equation 2.28). Thus, figure 2.7 is also valid for the DPSK case, except that now the phase error standard deviation is known. The 0.5 dB penalty point found before (10◦ phase error standard deviation), now means that Rb = 1/Tb should be higher than π∆ν/50. Similarly to what was shown before, in the limit case of infinite SNR, equation 2.50 gives the floor value of the probability of error. Thus, equation 2.29 also gives the BER-floor values for DPSK case. 2.3.2 oPLL based systems A phase locked loop is a feedback system in which the feedback signal is used to lock the output frequency and phase of the input signal. Phase locked loops in electrical domain have been one of the most frequently used com- munications circuits. Several applications like filtering, frequency synthesis, motor speed control, signal detection, etc. are common users of such device. While electrical PLL (used in heterodyne systems) is a well known device, optical version (used in homodyne systems) offers several technological problems which have delayed its development to the general market. Next figure 2.10 shows the basic components for a simplified OPLL when no noise influence is considered. The three basic elements are the phase comparator, the electrical filter and the VCO module. In our case, the phase comparator is comprised by the optical coupler and the photodetection front end, while the VCO module is a tunable laser. After filtering DC terms and high frequency terms at the output of phase comparator, the signal remaining is: V (t) = GP C PS sin(φe (t)) (2.51) √ where phase error is defined as φe (t) = φS (t) − φLO (t), and GP C = RL PLO This leads to the well-known PLL characteristic equation: +∞ dφe (t) dφS (t) dφLO (t) dφS (t) = − = − AG sin φe (τ ) f (t − τ )dτ (2.52) dt dt dt dt −∞
  • 42. Chapter 2. State of the art 23 Figure 2.10: Optical Phase Locked Loop simplified scheme √ where A = PS ; G = GV CO GP C , f (t) is the loop filter transfer function, and GV CO is the VCO gain in [rad/sV]. Although the PLL is not linear because the phase detector is non-linear, it can be accu- rately modelled as a linear device when the phase difference between the phase-detector input signals is small. For the linear analysis, it is assumed that the phase detector output is a voltage which is a linear function of the difference in phase between its inputs. This offers an easy way to study its behaviour by means of Laplace transformation, being the OPLL transfer function: ΦLO (S) AGF (S) H(S) = = (2.53) ΦS (S) S + AGF (S) A Proportional-Integral (PI) filter is usually used to act as a PLL regulator. Then F (S) is: 1 + τ2 S F (S) = (2.54) τ1 S and the OPLL transfer function becomes: 2 2ξωn S + ωn H(S) = (2.55) S 2 + 2ξS + ωn2 Being ωn = AG/τ1 the natural frequency of the PLL and ξ = ωn τ2 /2 the loop damping coefficient. 2.3.2.1 Additive noise impact in a generic OPLL Phase Locked Loop’s target is to match input signal phase. However this objective can be limited by several parameters which affect the receiver performance. Additive noise
  • 43. Chapter 2. State of the art 24 can interfere in the phase locked loop behaviour. In fact this noise produces an additional phase error that reduces the system’s functionality. In order to show a simplified model, a unique additive noise source (VN (t)) has been considered, added after the phase comparator module, and coming from the input shot noise plus electronic noise. Thus, the resulting characteristic equation for this case, in the Laplace domain, is found to be [15]: SΦS (S) GV CO VN (S)F (S) Φe (S) = − (2.56) S + AGF (S) S + AGF (S) The noise transfer function can be defined as follows: ΦLO (S) Φe (S) GV CO F (S) H(S) AS (S) = =− =− =− (2.57) VN (S) ΦS (S)=0 VN (S) ΦS (S)=0 S + AGF (S) AGP C The phase error is given by the contribution of the signal and the contribution of shot noise. In steady state the characteristic equation is linear and the superposition principle can be applied. Then, Φe (S) can be decomposed into two contributions (signal and noise): Φe (S) = ΦeS (S) + ΦeN (S) (2.58) where H(S) GV CO F (S) ΦeN (S) = VN (S)AS (S) = −VN (S) = −VN (S) (2.59) AGP C S + AGF (S) The interesting parameter is the phase error variance. Assuming φeN (t) is a white Gaus- sian process, and that VN (t) has as a power spectrum density SAN (ω): 2 σAN = φ2 (t) eN (2.60) +∞ dω = SAN (ω)|AS (ω)|2 (2.61) −∞ 2π +∞ 1 dω = 2 G2 SAN (ω)|H(ω)|2 (2.62) A PC −∞ 2π 1 BN = (2.63) SNR Be +∞ where the noise equivalent bandwidth of the PLL has been defined as BN = −∞ |H(ω)|2 dω , 2π and Be is the electrical input bandwidth (typically 0.7 · Rb ).
  • 44. Chapter 2. State of the art 25 For the case of a first order loop filter BN can be calculated as [15] ωn 1 BN = ξ+ 2 (2.64) 2 4ξ 2.3.2.2 Phase noise impact in a generic OPLL It is well known that phase noise is the major limitation on Phase Locked Loops. This parameter is especially harmful in optical PLL where two lasers (transmitter and local oscillator) are important sources of this noise. In fact the treatment could be done equivalently considering all phase noise located at the local oscillator side. The linewidth imposed is the sum of the one imposed separately for each contribution. In many cases, the impossibility to find a cost effective laser with low phase noise has limited the development of optical systems using such device. The phase error contribution can be separated by superposition due to the linearity of the system. So, it can be described as follows: Φe (S) = ΦeD (S) + ΦeN (S) (2.65) ΦeN (S) = ΦN (S)AΦ (S) (2.66) ΦN (S) = (2.67) S + AGF (S) = ΦN (S) 1 − H(S) (2.68) +∞ 2 dω σP N = SP N (ω)|AΦ |2 (2.69) −∞ 2π +∞ dω = SP N (ω)|1 − H(ω)|2 (2.70) −∞ 2π = 2π∆νBP N (2.71) where SP N = 2π∆ν/ω 2 is the phase noise baseband equivalent power spectrum density 2 +∞ 1−H(ω) dω (PSD) with lorentzian shape, and BP N = −∞ ω 2π For the case of a first order loop filter, shown in equation 2.55, BP N can be easily inte- grated leading to: 1 BP N = (2.72) 2ξωn
  • 45. Chapter 2. State of the art 26 Note that the phase error variance is inversely proportional to the loop bandwidth (high- pass), so the performance will be better as this parameter grows. While the additive noise PLL bandwidth BN is low-pass, the PLL bandwidth for phase noise is high-pass; meaning that a trade off between them must be selected. If the PLL is optimized for cancelling the phase noise, then its tolerance against additive noise will be worse, and viceversa. √ This is shown in figure 2.11. The loop has been evaluated for ξ = 1/ 2, variable natural frequency (from 1 kHz to 100 MHz), and a bitrate of 1 Gb/s. Figure 2.11: Iso-curves of the variance of additive noise (left) and phase noise (right), √ all for ξ = 1/ 2. 2.3.2.3 Loop delay impact in a generic OPLL The analysis performed in the previous subsections, has assumed both negligible loop propagation delay and absolute stability for all closed loop systems. If the assumption of zero effective time delay is not made then it becomes evident that, for systems of wider bandwidth, absolute stability may not be guaranteed. From the phase locked loop linear model a few modifications should be introduced to characterize the loop propagation delay, and the new OPLL transfer function can be found: AGF (S)e−τ S H(S) = (2.73) S + AGF (S)e−τ S Except the new transfer function expression, the phase error variance equation remains the same as in the negligible delay case. Using the expressions, the standard deviation
  • 46. Chapter 2. State of the art 27 Figure 2.12: PLL parameters optimization for 1 ns loop delay and 1 MHz linewidth. of phase noise can be evaluated as a function of the loop delay, natural frequency and linewidth. Figure 2.12 shows, for a certain loop delay, that the loop parameters ξ and ωm have to be optimized, thus changing the optimum natural frequency and the maximum linewidth allowed, as shown in equation 2.71. Set ξ to a certain value, increasing ωn produces higher bandwidths, however, over an optimum value, closed loop response is highly distorted and PLL requirements become harder. This is clearly shown in [16], where, after a deep analysis, it is demonstrated that the loop linewidth requirements get relaxed when loop is overdamped (in fact, when ξ > 2); and stability is achieved for 2ξωn < 1.6. In order to find a closed form for the loop delay impact on the optical PLL noise,Norimatsu [17] used a Pad approximation to e−τ S . This is an easy way to obtain the analytic solution for equivalent noise bandwidths. A second order Pad approximation gives only 0.4% difference respect to numerical calculations from equation 2.73. In this case, 12 + 6τ S + (τ S)2 e−τ S = (2.74) 12 − 6τ S + (τ S)2 thus, phase and additive noises can be calculated as 2 1 σAN τ = · x+1 144 + 36xy 2 + 72xy + 144x − x2 y 4 − 48x2 y 2 + 72x2 y − x3 y 4 + 6x3 y 3 − 12x3 y 2 2 · σAN 144 − 144y − 12xy 3 + 36xy 2 − 72xy − x2 y 4 + 6x2 y 3 − 12x2 y 2 (2.75)
  • 47. Chapter 2. State of the art 28 2 144 − 36x2 + 72xy − x2 y 4 + 6x2 y 3 − 12x2 y 2 σP N τ = σ2 (2.76) 144 − 144y − 12xy 3 + 36xy 2 − 72xy − x2 y 4 + 6x2 y 3 − 12x2 y 2 P N √ √ where ξ = x/2 and ωn τ = y x. Even they are long expressions, they permit us to calculate directly the influence of the noises in the presence of a long loop delay. Using these expressions, the loop delay impact on a generic PLL could be evaluated. Such evaluation was for ξ = 2, variable natural frequency (from 1 kHz to 100 MHz), and 2 1 ns loop delay (equivalent to 20 cm of optical fiber). For the σAN the bitrate assumed was 1 Gb/s. Results are shown in figure 2.13. From these results, without a loop delay, 2 2 σP N and σAN have a behaviour similar to what is depicted in figure 2.11. Nevertheless, when the loop delay impact is taken into account, an inestability point can be seen near ωn = 81 MHz. Figure 2.13: Iso-curves of the variance of additive noise and phase noise, all for ξ = 2. (a-b) are for a null loop delay, whereas (c-d) are for a 1 ns loop delay. 2.3.2.4 Costas loop Several phase locked loops have been developed for microwave applications. Some of them have been modified and converted to the optical domain as homodyne receivers to lock the phase of the incoming signal. Costas, Decision driven and Balanced phase locked loop are the most successful configurations.
  • 48. Chapter 2. State of the art 29 The first specific implementation is based on Costas-Loop design. The receiver extracts both In-phase an Quadrature signals and mixes them in order to eliminate the informa- tion. This kind of PLL is implemented using a 90◦ hybrid, capable to extract the In-phase and Quadrature components after optical mixing. These components are then multiplied, leading to a an error signal proportional to sin(2φe (t)), which is driven to the loop filter. Figure 2.14: Costas PLL scheme. The behaviour of the hybrid can be described by the transfer matrix S:    √ √  E1 1−k √k √  E2  ¯ ¯   1  1−k  √ − k  ES Eo =   = S · EI = √  √ ·  (2.77)  E3  2  √k j 1−k  ELO √ E4 k −j 1 − k where k is the fraction of input power that arrives at each detector (ideally 1/2). Thus, the equations related to this loop are those corresponding to the voltages at each point of the scheme: ∗ V1 (t) = RL E1 (t)E1 (t) 1 = RL (1 − k)PS + kPL O + 2d k(1 − k)PS PLO cos φe (t) (2.78) 2 ∗ V2 (t) = RL E2 (t)E2 (t) 1 = RL (1 − k)PS + kPL O − 2d k(1 − k)PS PLO cos φe (t) (2.79) 2
  • 49. Chapter 2. State of the art 30 ∗ V3 (t) = RL E3 (t)E3 (t) 1 = RL (1 − k)PS + kPL O − 2d k(1 − k)PS PLO sin φe (t) (2.80) 2 ∗ V4 (t) = RL E4 (t)E4 (t) 1 = RL (1 − k)PS + kPL O − 2d k(1 − k)PS PLO sin φe (t) (2.81) 2 VI (t) = V1 (t) − V2 (t) = 2 RL d k(1 − k)PS PLO cos φe (t) (2.82) VQ (t) = V3 (t) − V4 (t) = 2 RL d k(1 − k)PS PLO sin φe (t) (2.83) Vm (t) = VI (t) · VQ (t) 2 2 = 2 RL k(1 − k)PS PLO sin 2φe (t) (2.84) Hence, assuming the loop is in tracking mode (φe (t) 1), and that all the power between photodetection paths is equally distributed (k = 1/2) 2 2 Vm (t) = RL PS PLO φe (t) (2.85) So, the loop can be linearized, leading to equation 2.53, but with some changes: A should 2 be substituted by PS , and GP C by RL PS PLO . It should be noted that in the Costas loop, phase ranging is performed with more accuracy than other loops. Since there is no data detection inside the loop, there is no data to phase error crosstalk. In order to optimize data detection, power distribution between I-Q paths has to be optimized. For instance, in [18], a Costas loop was optimized with 22 ns delay and 80 kHz of total laser linewidth, by using a 4:1 splitting ratio for local laser signal, and a 7:1 ratio for received signal. In this case, for data rate of 5 Gb/s they obtained a phase error variance of 2.9. 2.3.2.5 Decision-Driven OPLL (DD-OPLL) In this case, the first phase comparator is the well-known Decision-Driven OPLL (DD- OPLL). This is better than Costas-Loop in terms of linewidth requirements [17, 19], and its general schematic is depicted in figure 2.15. The main difference with the Costas loop is that here the influence of the data (BPSK coded, coming from I branch) on the
  • 50. Chapter 2. State of the art 31 phase error signal (coming from Q branch) is removed by performing the data decision and afterwards multiplying with such error signal. Lets see how it can be described Figure 2.15: Decision driven PLL scheme. mathematically, in order to better understand its principles. First lets describe the main optical and electrical signals. From the transmitter side, the optical field (eS (t)) and its complex amplitude (ES (t)) can be defined: eS (t) = PS exp(j(ωt + φS (t))) (2.86) ES (t) = PS exp(jφS (t)) (2.87) where the phase is defined as: φS (t) = φD (t) + φN S (t) (2.88) 0 d=1 φD (t) = (2.89) π d = −1 Here φN S (t) is the noise process associated to the transmitter side, and φD (t) is the data signal. From the local oscillator side, the optical signals are defined in a similar way: eLO (t) = PLO exp(j(ωt + φLO (t))) (2.90) ELO (t) = PLO exp(jφLO (t)) (2.91) φLO (t) = φC (t) + φN LO (t) (2.92)
  • 51. Chapter 2. State of the art 32 being φC (t) the phase introduced by the local loop, and φN LO (t) the noise process of the local laser. Thus, after optical combination and further photodetection, the electrical signals after each balanced detector can be written: V1 (t) = GP C PS d sin φN (t) − φC (t) + n1 (t) (2.93) V2 (t) = GP C PS d cos φN (t) − φC (t) + n2 (t) (2.94) where the overall phase noise process is φN (t) = φN S (t) + φN LO (t), the shot noise (pho- √ todetection and receiver parts) is characterized by n1,2 (t), and GP C = RL PLO . As said before, it can be seen in figure 2.15 that the Decision-Driven implementation must separate perfectly the data from the OPLL operation. In this sense a perfect data recovery is assumed, which is a good approximation for the usual BER target values (e.g. 10−9 ). Please note that FEC data encoding is regarded as an additional resource, so at this stage the target BERs are higher. m(t) = dV1 (t − Tb ) = dGP C PS d sin φN (t) − φC (t) + dn1 (t) (2.95) m(t) = GP C PS sin φN (t) − φC (t) + n1 (t) (2.96) Data has no influence on the shot noise impact, therefore it can be omitted. As in generic PLL here it can be assumed a linear operation: |φN (t − Tb ) − φC (t − Tb )| 1 (2.97) m(t) = GP C PS φN (t) − φC (t) + n1 (t) (2.98) Thus the same phase-detection function than the generic PLL is obtained. Therefore the analysis is completely equivalent and results obtained in the previous subsections are applicable to this loop. One difference is the additional bit delay, that may have some effect when it is not properly optimized, and for very high PLL bandwidths. Also, another remarkable difference is the fact that it requires a 90◦ optical hybrid that is not a simple element. Even though some feasible approaches have been proposed [20], it is still a complex device. Please note that binary data detection is performed over the In-phase (I) branch, while Quadrature (Q) branch is kept for phase tracking. So, power distribution between both photodetection paths should be carefully designed for each particular case, in order to optimize both, data detection and tracking performances.
  • 52. Chapter 2. State of the art 33 2.3.2.6 Balanced OPLL The third PLL design corresponds on Balanced PLL. This receiver does not try to elimi- nate the bits influence on the phase ranging; but just considers it as an extra noise which produces certain penalty. Usually high frequency bits do not represent an important impact to low bandwidth processes [21]. Figure 2.16: Balanced PLL scheme. From figure 2.16, the following set of equations can be written, describing the main optical and electrical signals: eS (t) = PS exp(j(ωt + φS (t))) (2.99) ES (t) = PS exp(jφS (t)) (2.100) φS (t) = φD (t) + φN S (t) (2.101) eLO (t) = PLO exp(j(ωt + φLO (t))) (2.102) ELO (t) = PLO exp(jφLO (t)) (2.103) φLO (t) = φC (t) + φN LO (t) (2.104) π φD (t) = + ϕd(t) (2.105) 2 φe (t) = φN S (t) − φN LO (t) − φC (t) (2.106) where ϕ corresponds to the phase deviation between the two different bits d(t) = 1 and d(t) = −1; being 2ϕ the peak to peak phase deviation. For a better understanding, this phase is depicted in the phasor scheme of figure 2.17.
  • 53. Chapter 2. State of the art 34 Figure 2.17: Balanced PLL phasor scheme. The 180◦ hybrid (usually implemented by an optical 50:50 coupler) behaviour can be described by next matrix: 1 1 1 ES Eo = √ · (2.107) 2 1 −1 ELO So, the equations for the photodetected signals are: 1 V1 (t) = RL PS + PLO + 2 PS PLO cos φS (t) − φLO (t) (2.108) 2 1 V2 (t) = RL PS + PLO − 2 PS PLO cos φS (t) − φLO (t) (2.109) 2 Next, V3 (t) can be calculated, and properly rewritten as: V3 (t) = 2 RL PS PLO cos φS (t) − φLO (t) π = 2 RL PS PLO cos + ϕd(t) + φN S (t) − φC (t) + φN LO (t) 2 = 2 RL PS PLO sin ϕd(t) + φN S (t) − φC (t) − φN LO (t) (2.110)
  • 54. Chapter 2. State of the art 35 If V3 (t) is expanded by using trigonometric identities: V3 (t) = 2 RL PS PLO sin(ϕ)d(t) cos φN S (t) − φC (t) − φN LO (t) + + 2 RL PS PLO cos(ϕ)d(t) sin φN S (t) − φC (t) − φN LO (t) (2.111) where the phase error term can be easily identified as φe (t) = φN S (t) − φN LO (t) − φC (t). Then, assuming it is small enough, the phase error ranging signal can be obtained. this phase error signal will be driving the loop filter: V3 (t) ≈ 2 RL PS PLO sin(ϕ)d(t) + RL PS PLO cos(ϕ)d(t)φe (t) (2.112) being the first term the most useful for data, and the second term the one that contains φe (t) and, thus, the phase error. This kind of PLL is influenced by phase noise, shot noise and by crosstalk between data- detection and phase ranging signals. However, this effects can be controled by carefully adjusting ϕ. It should be noted that ϕ can not be 0◦ nor 90◦ . Precisely, [22] demonstrates that for very low linewidth (1 kHz) and 10 Gb/s data rate, the optimum value is ϕ = 85◦ . Thus, in this case loop behavior in front of noise (additive and phase noise) is affected by data crosstalk, leading to [22]: 2 2π∆ν ωn (4ξ 2 + 1) q ωn (4ξ 2 + 1) σN = + 2Tb tan2 (ϕ) + (2.113) 2ξωn 8ξ 2 PS cos2 (ϕ) 8ξ 2 As an example, the noise variance σN was evaluated for several laser linewidth values (up to 100 MHz) and different ϕ values, from 80 up to 100. The values of the other parameters used for such calculations where standard ones: 1 Gb/s bitrate (intended for access networks), photodetector responsivity of 0.8 A/W, PS of 0 dBm, loop damping √ factor ξ = 1/ 2, and loop natural frequency of 1 MHz. Results are shown in figure 2.18. There it is shown that as the laser linewidth increases the noise power also increases, as expected. As said before, the ϕ value is also important: when it is approaching 90◦ , the second term of equation 2.112 gets close to 0◦ and the phase error can not be ranged. Thus, the phase error variance has a high peak at that point. In order to avoid this, one should take ϕ of 80◦ − 85◦ degrees (or 95◦ − 100◦ ). Then, even the symbols received are not orthogonal at all, the PLL is capable to track and compensate the phase error. In summary, although balanced OPLL theoretically has worse behaviour than other kind of phase locked loops, it must be taken into account due to its simplicity.
  • 55. Chapter 2. State of the art 36 Figure 2.18: Noise variance for the balanced PLL scheme. 2.3.2.7 Subcarrier modulated OPLL (SCM-OPLL) In this loop, a subcarrier is used in order to track phase of the received signal [23]. Such subcarrier is achieved by modulating the output of the local laser using a Mach-Zehnder Modulator (MZM) driven by an electrical VCO. There exist several approaches, but in the present work the analyzed approach is the one based on the Decision-Driven loop, as it one of the most robust [23]. A scheme is shown in 2.19. Figure 2.19: General scheme for a subcarrier decision driven optical phase-locked loop.
  • 56. Chapter 2. State of the art 37 In that scheme, the same analysis we did in the previous loops can be done. So, lets start writing down the equations which describe the optical signals: eS (t) = PS exp(j(ωS t + φS (t))) (2.114) φS (t) = φD (t) + φN S (t) (2.115) eLO (t) = PLO exp(j(ωLO t + φLO (t))) (2.116) π π φLO (t) = φC (t) + φN LO (t) + + cos(ωc t) (2.117) 2 2 ωc = ωs − ωLO (2.118) where, in this case, the MZM is properly biased at 90◦ point and the electrical VCO drives the MZM with a sinusoidal signal that introduces a 180◦ phase shift with frequency ωc . Thus, the signal used for phase ranging becomes: m(t) = RL PLO PS · π π · sin ωS t + φD (t) + φN S (t) − ωLO t − φC (t) − φN LO (t) − − cos(ωc t) 2 2 = RL PLO PS · · sin ωS t + φN S (t) − ωLO t − φC (t) − φN LO (t) cos(ωc t) 1 = RL PLO PS · 2 · sin φN S (t) − φC (t) − φN LO (t) + sin 2ωc tφN S (t) − φC (t) − φN LO (t) (2.119) Assuming ωc big enough, this signal can be low-pass filtered. Also, the phase error can be identified as φe (t) = φN S (t) − φN LO (t) − φC (t), and rewrite the expression as: 1 e(t) = RL PLO PS sin φe (t) (2.120) 2 With this signal at the input of the loop filter, the loop can be linearized, also leading to equation 2.53, but taking into account that now: 1 GP C = RL PLO (2.121) 2 Please note that this loop has an electrical VCO. Thus, the main advantage of this loop is that its the performances are expected to be improved in terms of hig-speed phase tracking and acquisition. Nevertheless, it needs several optical components (MZM, 90◦ hybrid, quadruple photodetector array), meaning that its cost may be quite high. Also, it should be noted that in this case the local laser is not running free at all: a wavelength control loop should be placed in order to keep it close to the received wavelength. In
  • 57. Chapter 2. State of the art 38 chapter 3 a comparison with other loops will be shown, when evaluating a novel PLL approach proposed in this thesis. 2.3.3 Phase and polarization diversity systems Instead of synchronizing the local laser with the optical incoming carrier by using a feedback loop, the phase diversity systems do it by means of a feedforward processing module. Thus a free-running laser is placed at the LO side. Figure 2.20: Scheme of a phase diversity front end. In the post-processing module the phase of the received signal is estimated and the optical impairments are compensated by means of electronic and/or digital signal processing. In order to allow this stage to work properly, In-Phase and Quadrature components of the received signal and local oscillator combination have to be present. That is the reason why a 90◦ hybrid is the most common way to obtain both optical signal components [20]. This architecture is based on an intradyne system, using a local oscillator which is nominally at the same frequency as the incoming signal. The 90◦ hybrid separates the incoming optical signal into I and Q components, corresponding to the real and imaginary parts of a complex signal. An scheme of such a receiver front-end is shown in figure 2.20. Along a different line, polarization diversity is used and obtained in a similar way, in order to achieve polarization insensitivity. As commented before, If polarization diversity is also desired, then the receiver front-end becomes much complex. It is shown in figure 2.21. Once seen the front-end architectures lets see how they are used to detect PSK or DPSK signals. Also, some digital signal processing techniques for estimating phase will be shown, although they are used for detecting multilevel PSK or DPSK signals. 2.3.3.1 Multiple differential detection The idea of this receiver is to perform an analog processing (differential detection) of the signals provided at the output of the optical front-end. It was first analyzed and
  • 58. Chapter 2. State of the art 39 Figure 2.21: Schematic of phase and polarization diverse receiver. Figure 2.22: Scheme of a DPSK detection, in a phase and polarization diversity homodyne receiver. experimentally demonstrated in [24]. The main advantage of this detection scheme is its simplicity, and the fact that it can be implemented with simple analog electrical components. Figure 2.22, shows the detection scheme for that type of receiver. At the output of each branch of the diversity front-end, a differential detector (delay and multiply) is placed. Next, all signals coming from this detectors are added.
  • 59. Chapter 2. State of the art 40 Given eS (t) the received signal optical field, and eLO (t) the local oscillator optical field, for this receiver: eS (t) = PS (cos(ϕ)ˆ + sin(ϕ)e−jθ y ) exp jω0 t + φS (t) x ˆ (2.122) eLO (t) = PLO (ˆ + y ) exp jω0 t + φLO (t) x ˆ (2.123) φe (t) = φS (t) − φLO (t) (2.124) where x and y are the polarization orthogonal components (H,V); and ϕ and θ are the ˆ ˆ phases related to the polarization. Since polarization fluctuations are very slow (∼ 1 Hz), these variables are assumed to be constant during the bit sequence. Please note that while eS (t) has a random polarization, eLO (t) has a +45◦ linear polarization. So, the photodetected voltages are: VHI (t) = Cd (t) cos(ϕ) cos φe (t) (2.125) VV I (t) = Cd (t) sin(ϕ) cos φe (t) + θ (2.126) VHQ (t) = Cd (t) cos(ϕ) sin φe (t) (2.127) VV Q (t) = Cd (t) sin(ϕ) sin φe (t) + θ (2.128) √ being φe (t) = φS (t) − φLO (t) the phase error; C = 2 RL PS PLO the maximum voltage, d(t) = d (t) · d (t − Tb ) the detected data. Then, the output signals at each branch after differential demodulation can be calculated: C2 VHIo (t) = d(t) cos2 (ϕ) cos φe (t) cos φe (t − Tb ) (2.129) 2 C2 = d(t) cos2 (ϕ) cos φe (t) − φe (t − Tb ) + cos φe (t) + φe (t − Tb ) 2 C2 VV Io (t) = d(t) sin2 (ϕ) cos φe (t) + θ cos φe (t − Tb ) + θ (2.130) 2 C2 = d(t) sin2 (ϕ) cos φe (t) − φe (t − Tb )) + cos φe (t) + φe (t − Tb ) + 2θ 2 C2 VHQo (t) = d(t) cos2 (ϕ) sin φe (t) sin φe (t − Tb ) (2.131) 2 C2 = d(t) cos2 (ϕ) cos φe (t) − φe (t − Tb ) − cos φe (t) + φe (t − Tb ) 2
  • 60. Chapter 2. State of the art 41 C2 VV Qo (t) = d(t) sin2 (ϕ) sin φe (t) + θ sin φe (t − Tb ) + θ (2.132) 2 C2 = d(t) sin2 (ϕ) cos φe (t) − φe (t) − cos φe (t) + φe (t − Tb ) + 2θ 2 thus, given Vo (t) = VHIo (t) + VV Io (t) + VHQo (t) + VV Qo (t), and after some algebra: C2 Vo (t) = d(t)cos φe (t) − φe (t − Tb ) (2.133) 2 Now, identifying terms from the values calculated, it is easy to see that: 4 PS PLO SNR = 4kB T (2.134) 2q (PS + PLO ) + ID BE + RL FN BE And using equation 2.50 the probability of error for this receiver can be calculated, with a noisy reference source. The system works pretty well [24], although the cost is relatively high due to the amount of components that have to be used. An alternative version using digital signal processing was proposed for higher order mod- ulation formats [25]. However, it gets out of the scope of this thesis, since poly-phase signals are mainly intended for core networks and long-haul transmission systems. 2.3.3.2 Wiener filter phase estimation In this case polarization issues are disregarded in order to focus on the phase noise mitigation, and the diversity front-end used is the one depicted in figure 2.20; but with an analog to digital converter (ADC) at both outputs, Re and Im. After that, a phasor r(n) describing received signal can be constructed: r(n) = VI (t0 ) + jVQ (t0 ) (2.135) being t0 the optimum sampling time. Along a different line, once digitized, laser phase noise can be regarded as an auto- regressive (AR) process seeded by white noise. This is developed in detail and demon- strated in appendix D. Taking into account this fact and other noises interacting as white noise; good phase estimation can be done using well-known linear prediction techniques by employing a Wiener filter design. This was done in [26], with very good results.
  • 61. Chapter 2. State of the art 42 A Lorentzian-shaped laser tone contains phase noise, such that, once digitized: φ(n) = φ(n − 1) + w(n) (2.136) 2 where w(n) is a zero-mean Gaussian noise sequence, having a variance of σw = 2π∆νTs , with linewidth ∆ν, and fs = 1/Ts being the sampling rate. Then, the digitized signal can be reconstructed from the I and Q components, and written as: r(n) = VI (t0 ) + jVQ (t0 ) = d(n) exp jφ(n) + p(n) (2.137) where d(n) is the data, and p(n) is the complex additive noise (Gaussian, with variance 2 σp ). In order to remove the data modulation r(n) is raised to the M th power: s(n) = rM (n) = exp jM φ(n) + M p (n) exp jM φ(n) + o p2 (2.138) where p (n) is a different Gaussian noise with the same variance as p(n). Since that the phase noise is small, a small angle approximation applies: θ(n) = arg s(n) ≈ M φ(n) + Im{p (n)} (2.139) ˆ Then, the problem of finding the best estimate of φ(n) (φ(n)) can be stated as Wiener filter problem: ˆ 1 ˆ φ(n) = θ(n) (2.140) M The Wiener filter can be easily calculated, and, in terms of Z-transform it is [26]: ˆ θ(z) = H(z)θ(z) (2.141) D 1−α = αD + (1 − α) αD−k z −k θ(z) (2.142) 1 − αz −1 k=1 where D is the lag number of symbols, and M 2 σw + 2σp − M σw 2 2 M 2 σw + 4σp 2 2 α= 2 2σp For the particular case of the Wiener filter, the error of the estimator can be easily calculated and found to be: ˆ e(z) = θ(z) − θ(z) = θ(z) 1 − H(z) (2.143)
  • 62. Chapter 2. State of the art 43 Thus, the transfer function of the error can be written as: e(z) He (z) = = 1 − H(z) (2.144) θ(z) and the LMS error of this estimator can be calculated [14]: π 2 1 E{e (n)} = |He (ejω )|2 θ(ejω )dω (2.145) 2π −π Given this measure of the estimator’s error, a numerical analysis of a Wiener filter, with a lag of 10 symbols, has been performed. Results are shown in figure 2.23. There it is shown how the estimator performances are bounded by the phase noise and the SNR. The SNR parameter refers to the electrical SNR after photodetection. So, for low linewidths and low SNR, the estimator error increases. Also, when the linewidth is high enough, the estimator error is also high for high SNR values. So, from the point of view of the estimator LMS error, the optimal working zones should be defined for high linewidth (> 1 MHz) and non-optimal SNR (< 15 dB). Figure 2.23: LMS error for a Wiener filter with a lag of 10 symbols. Given this estimator, a comparison with other methods was carried out by Taylor in [26, 27]. The results obtained for the BPSK case (1.5 Gb/s), and a total laser linewidth of 48 MHz, are summarized in table 2.3. Please note, that for a Gaussian approximation,
  • 63. Chapter 2. State of the art 44 a probability of error of 10−9 is achieved for a Q factor of 15.56 dB. Thus, under this hypothesis, the BER was found for each case and is shown in table 2.3. There the Wiener filter estimation (with a lag D=10 symbols) is only ∼0.1 dB away from the Maximum a posteriori (MAP) estimate; which, a priori, is the best possible estimate. Estimator Q factor (dB) BER Wiener filter D=10 8.61 3.5 · 10−3 Wiener filter D=0 6.25 2 · 10−2 PLL 5.88 2.45 · 10−2 MAP 8.66 3.4 · 10−3 Table 2.3: Comparison of phase estimation methods, as seen in [27]. BER is estimated from Q factor and assuming Gaussian statistics. 2.3.3.3 M-power law phase estimation with regenerative frequency dividers This technique, is another digital signal processing algorithm for phase estimation and it was proposed in [28]. Here, once I and Q components are photodetected, they are digitized by high speed ADC. After that, a phasor describing received signal can be constructed for each orthogonal polarization state detected (XV , XH ). Figure 2.24: Scheme of a phase estimator for polarization multiplexed QPSK signals based in regenerative frequency dividers. After the ADCs, the resulting digitized signal rH,V is raised to the M th power (M is the number of different symbols that can be transmitted) and filtered, for cancelling the mod- ulation. In the example shown in figure 2.24, QPSK modulation format is assumed, for which M = 4. Next, a low-pass filter is placed in order to eliminate the broadband noise and high frequency harmonics generated when raising to the M th power. Afterwards,
  • 64. Chapter 2. State of the art 45 the output signal of the filter, drives a stack of regenerative frequency dividers, to divide the signal frequency by M . In the QPSK case, a couple of them are enough. Finally, the resulting signal describes a phasor with an estimation of the received signal phase. The main difference between the present approach and the Wiener filter, is the use of additional regenerative frequency dividers, that make the algorithm and, thus, the receiver more complex. In case a polarization division multiplex is desired, the phasors correspondent to orthog- onal states rH , rV are each sent to a carrier recovery and demodulation unit, as shown in figure 2.24. There, the phase-aligned frequency-multiplied carrier signals of both branches are added before being passed through the low-pass filter and the regenerative frequency dividers. Finally, after phase cancellation, demodulation is performed. An important point is that the components of detected data can be correlated with the recovered data symbols, for compensating polarization fluctuations [28]. In summary, this is an extremely complex frequency estimator. Performances expected in [28] were achieved in part by its experimental work reported in [29], obtaining a 0.5% linewidth per bitrate ratio phase noise tolerance at BER-floor of 10−3 . 2.3.3.4 Viterbi-Viterbi phase estimation This technique is also another digital method for carrier phase estimation, which was first demonstrated for the optical domain in [30]. Perhaps it is the most extensively used technique for coherent detection of poly-phase optical signals and M-QAM data. In this case, the received complex samples r(n) are first raised to the M th power to remove the M-ary phase modulation. To more accurately estimate the phase error out of the shot-noise, the raised samples of a block of length N are averaged, and a common phase error estimate for the block is then obtained by calculating the argument of the complex sum vector: N 1 θ(n) = arg rM (k) (2.146) M k=1 To combat the arising M-ary phase ambiguity, differential encoding has to be employed, and the last symbol of a block has to be corrected after segment changes [31]. A summary of the performances of this estimator was presented in [31], giving a 10−4 linewidth per bitrate ratio phase noise tolerance for 1 dB penalty at BER of 10−4 and QPSK modulation.
  • 65. Chapter 2. State of the art 46 2.4 Chapter summary This chapter has provided a comprehensive review of optical coherent receivers, and their evolution all through the years. Therein, special focus has been put on simple optical phase modulation (BPSK and DPSK) and the influence of phase noise in the reference optical source. This was used to set up a background level to develop the present thesis. First, the BPSK and DPSK modulation formats have been introduced, and after they have been compared to the other formats available, while analyzing detection with noisy reference signals and showing the need to compensate the phase noise inherent to lasers. Next, the optical phase-locked loop concept has been introduced as a first solution, some well-known results for the generic expression have been given for different situations: Loop linearization plus additive noise, phase noise and loop delay impacts. Later, the main optical PLLs have been analyzed: Decision driven, Costas, subcarrier modulated, and balanced. The first three architectures are the more complex as they use a 90◦ optical hybrid (plus an amplitude modulator, for the case of the subcarrier modulated). But they also present the best linewidth tolerances and are very robust against possible data crosstalk. Nevertheless, their linewidth tolerances are in the order of the hundreds of kHz. Regarding the balanced OPLL, its simplicity should be high- lighted, even it has demonstrated very poor performances and being very sensitive to data crosstalk. Finally, the phase diversity architecture has been seen. A brief analysis has been provided and the most popular architectures and algorithms have been analyzed: 1. Analog processing for differential detection 2. Wiener filter phase estimation 3. M-power law phase estimation with regenerative frequency dividers 4. Viterbi-Viterbi phase estimation Even though the first one is based in a simple concept, nowadays the other ones, especially the ones based in Wiener filter and Viterbi-Viterbi algorithms, are becoming more popular due to the improvement of digital signal processors. It should be noted that with such algorithms one is allowed to use lasers on the order of tens of MHz, for a data rate of 1 Gb/s, representing a big improvement respect the OPLLs.
  • 66. Chapter 3 Lock-In amplifier OPLL architecture Phase locked loops in electrical domain have been one of the most frequently used com- munications circuits. Several applications like filtering, frequency synthesis, motor speed control, signal detection, etc. are common users of such device. While the electrical PLL (used in heterodyne systems) is a well known device, the optical version (used in homodyne systems) poses several technological problems which have delayed its development to the general market. As pointed in chapter 2, its three basic elements are the Phase Comparator, the Electrical Filter and the VCO module (tunable Laser). In the present chapter a new OPLL architecture is introduced, aiming to enhance the phase noise noise tolerance of these devices, while targeting a reduced cost [32, 33]. The idea is to use simple optics, as in the Balanced OPLL, but improving its phase noise tolerance [33]. In a homodyne receiver, the phase noise has a Lorentzian spectrum that masks completely the detected data and the phase error signal (also a baseband signal). Then, an option is to play with the electrical spectra in order to find a frequency band where the phase noise spectrum does not mask the phase error signal, improving the PLL performances. The Lock-In amplifier is a tool widely used in physics that performs this frequency shift by introducing a dithering at a certain point of the system. Thus, in a Lock-In amplified OPLL, the main idea (and novelty) is to sinusoidally dither the phase of the local laser by a small amount (e.g. 50 mrad), enough for achieving a certain phase modulation. This is done at a frequency above the loop bandwidth and below the bit rate. As the phase response of a coherent balanced receiver is a cosine (an even function, shown in figure 3.1), the dithering makes possible the measurement of the phase error by searching the derivative and removing the inherent positive/negative ambiguity. Also, the dithering leads to an amplitude modulated phase error signal after the balanced receiver, which is afterwards filtered and demodulated in a proper way. 47
  • 67. Chapter 3. Lock-In amplifier OPLL architecture 48 Figure 3.1: Voltage after balanced detector (V3 (t)) as a function of the phase error (φS (t) − φLO (t)). 3.1 System model For a better understanding of this optical PLL, lets proceed with the loop analysis. There the signals at each point of the loop will be described mathematically, and their interaction detailed. 3.1.1 Loop analysis and linearization The phase-locked loop model scheme is depicted in figure 3.2. It is a homodyne balanced receiver with a Proportional-Integral (PI) loop filter. Dithering is introduced after PI filter, and just next to photodetection stage the amplitude modulated error signal is properly filtered and demodulated. In the scheme presented, the following set of expressions is verified, as a basis for starting the analysis: eS (t) = PS exp(j(ωt + φS (t))) (3.1) ES (t) = PS exp(jφS (t)) (3.2) φS (t) = φD (t) + φN S (t) (3.3) eLO (t) = PLO exp(j(ωt + φLO (t))) (3.4) ELO (t) = PLO exp(jφLO (t)) (3.5) AKV CO φLO (t) = φC (t) + φN LO (t) + sin(ωc t) (3.6) ωc AKV CO φLO (t) = φT LO (t) + sin(ωc t) (3.7) ωc
  • 68. Chapter 3. Lock-In amplifier OPLL architecture 49 where ES is the transmitted optical field, ELO is the local oscillator optical field, PS is the transmitter optical power, PLO is the local oscillator optical power, φD (t) is the phase coded data, φN S (t) is the transmitter laser phase noise, φN LO (t) is the local laser phase noise, φC (t) is the phase introduced by the control loop filter, A is the electrical oscillator amplitude, KV CO is the local laser constant, and ωc is the dithering frequency. Figure 3.2: Lock-In amplified oPLL schematic. For the loop analysis, data φD (t) influence is assumed to be negligible. Afterwards, once the loop is analyzed and proven to be feasible, data effects on the loop performances will be evaluated. In order to start, the equations related to the balanced receiver can be found: 1 V1 (t) = RL PS + PLO − 2 PS PLO cos(φS (t) − φLO (t)) (3.8) 2 1 V2 (t) = RL PS + PLO + 2 PS PLO cos(φS (t) − φLO (t)) (3.9) 2 So the output signal is: V3 (t) = V2 (t) − V1 (t) (3.10) AKV CO = 2 RL PS PLO cos φS − φT LO (t) + sin(ωc t) (3.11) ωc = 2 RL PS PLO · η (3.12) AKV CO being η = cos φS − φT LO (t) + ωc sin(ωc t) . This voltage V3 (t) is the only way to measure the phase error. So, lets pay some special attention on it. First, the cosine part η can be expanded using the trigonometric identities
  • 69. Chapter 3. Lock-In amplifier OPLL architecture 50 related to the sum of arguments: AKV CO η = cos φS (t) − φT LO (t) cos sin(ωc t) − ωc AKV CO − sin φS (t) − φT LO (t) sin sin(ωc t) (3.13) ωc The term AKωcCO is the dithering amplitude, and it is a parameter to be designed and V optimized. Thus, it can be small enough not to interfere with the phase error, while keeping the desired amplitude modulation. Also, in tracking mode, the error signal itself is expected to be low. So, the dithering amplitude can be as low as to assume AKωcCO V 1. In this case, sine and cosine terms containing the dithering amplitude can be expanded in Taylor-McLaurin series and truncated after the first and second term respectively. Hence, η can be rewritten as: η = A2 KV CO 2 cos φS (t) − φT LO (t) 1− 2 + (3.14) 4ωc A2 KV CO 2 + 2 cos φS (t) − φT LO (t) cos(2ωc t) + (3.15) 4ωc AKV CO + sin φS (t) − φT LO (t) sin(ωc t) (3.16) ωc From that, the term 3.14 is a baseband signal containing the phase error information, but maintaining the ambiguity inherent to the cosine (even function). The same cos φS (t) − φLO (t) is also present in the term 3.15, now modulated with a carrier centred at 2ωc . Thus, the only useful term is the last one 3.16, containing a sin φS (t)−φLO (t) modulated with a carrier running at ωc . The distribution of these terms over the electrical spectrum is shown in figure 3.3. 6 3.14 3.16 3.15 - ωc 2ωc ω Figure 3.3: Spectral distribution of the terms 3.14, 3.15, and 3.16.
  • 70. Chapter 3. Lock-In amplifier OPLL architecture 51 As φS (t) − φLO (t) is mainly due to the phase noise, its variations can be assumed to be slower than ωc . i.e. the carrier frequency ωc is higher than the bandwidth of φS (t)−φLO (t). Additionally, the band-pass filter hf 1 (t) is designed to filter the term centered at ωc , eliminating the other components. So, V4 (t) can be written as: V4 (t) = 2 RL PS PLO · η ∗ hf 1 (t) (3.17) After the filter, the only surviving term will be the one running at ωc , and V4 (t) can be then rewritten as: V4 (t) = 2 RL PS PLO · γ (3.18) where AKV CO γ = η ∗ hf 1 (t) = sin φS (t) − φT LO (t) sin(ωc t) (3.19) ωc The next step is to demodulate the amplitude modulated phase error term present in V4 (t). That is the reason why V4 (t) is multiplied with a pure tone running at ωc , V5 (t). After multiplication, the component at two times ωc is eliminated by hf 2 (t), a low pass filter with square ideal response, giving V6 (t): V6 (t) = [V4 (t) · V5 (t)] ∗ hf 2 (t) (3.20) = [2 RL PS PLO · γ · A sin(ωc t)] ∗ hf 2 (t) (3.21) √ RL PS PLO A2 KV CO = sin φS (t) − φLO (t) (3.22) ωc and V5 (t) is defined as A sin(ωc t). Next, V6 (t) is introduced to the loop filter f (t), a standard proportional-integral control. Its output is added to the output signal of the 90◦ phase shift (driven by V5 (t)), giving the voltage V7 (t) that drives the local laser: V7 (t) = 2 RL PS PLO · γ ∗ hf 2 (t) ∗ f (t) + A cos(ωc t) (3.23) Ideally, the local laser gives a frequency proportional to V7 (t), being φLO (t): φLO (t) = KV CO V7 (t)dt + φN LO (t) (3.24) AKV CO = 2KV CO RL PS PLO · γ ∗ hf 2 (t) ∗ f (t)dt + sin(ωc t) + φN LO (t) ωc Then, identifying terms, the loop can be closed and its its behavior can be modeled with the following differential equation: dφC (t) = K sin(φS (t) − φLO (t)) ∗ f (t) (3.25) dt
  • 71. Chapter 3. Lock-In amplifier OPLL architecture 52 √ A2 RL PS PLO KV CO 2 where K = ωc Thus, while the local laser is controlled in low-pass response, the deviation measure signal is in band-pass. An important feature of this technique, is that an increase of the loop SNR in front of the additive noise is reported by controling the bandwidths of the Lock-In amplifier filters hf 1 (t), hf 2 (t) [34]. Assuming loop is in tracking mode (φS (t) − φLO (t) 1), this loop can be linearized by approximating the sine by its argument, like the typical PLLs as those analyzed by several known books as [15]. Then, the loop differential equation can be transformed in the Laplace domain, and rewritten as it was shown in section 2.3.2: ΦLO (S) KF (S) H(S) = = (3.26) ΦS (S) S + KF (S) Hence the expressions deduced for generic linearizable loops are applicable in the Lock-In amplified loop. The most important ones, when regarding the transceiver performances, are commented in the following subsections. 3.1.2 Noise, dithering and loop delay impacts Regarding additive and phase noises, it should be noted that the presented PLL expres- sions have been linearized, having the same output expressions as when working with a generic PLL. Then, the same analysis made in sections 2.3.2.1 and 2.3.2.2 can be carried out, obtaining the same outcome. Just as a reminder, the final results from equations 2.63 and 2.71 are revisited: 2 ωn 1 σAN = ξ+ 2 (3.27) 2SNRBe 4ξ 2 2π∆ν σP N = (3.28) 2ξωn Lets now analyze the dithering effects. In fact, a small phase perturbation is introduced in the phase term of V3 (t) and cannot be recovered at all. This perturbation is a sinusoid with an argument that varies rapidly compared to the loop natural frequency. So, from the statistical point of view, this argument represents a residual phase that can be regarded as a random phase process uniformly distributed from −π to +π. Please note that this process is independent from the additive and phase noises. Also, the variance of a sinusoid of a random phase with uniform distribution is the half of the squared amplitude [14]. Thus, the phase error variance due to the dithering can be written as: 2 A2 KV CO 2 σd = 2 (3.29) 2ωc
  • 72. Chapter 3. Lock-In amplifier OPLL architecture 53 As an example, for a small dithering amplitude of 50 mrad a phase error deviation of 2◦ is obtained, that has almost no impact on the BER of the proposed receiver. Hence, as all the noises plus dithering can be regarded as independent processes, the total standard deviation of the phase error can be written as: 2 2 2 σ = σAN + σP N + σd 2 ωn 1 2π∆ν A2 KV CO = ξ+ 2 + + 2 (3.30) 2SNRBe 4ξ 2ξωn 2ωc When loop delay is not negligible, performances in tracking mode are the same as for a general PLL scheme. So, loop stability will be achieved for damping factor ξ higher than 2, and a 2ξωn τ product under 1.6 [16]. Also, the theoretical limit for 1 dB penalty at BER 10−9 will be of 0.74 for the 2ξωn τ product. This will be shown, when talking about the loop simulations performed. 3.1.3 Acquisition parameters 3.1.3.1 Hold in range Presuming that the PLL is locked, the hold in range is defined as the maximum frequency shift of the received signal with respect to the nominal frequency of the local laser, for which the PLL is still tracking the received signal. Since the PLL could be linearized, and the active filter in the loop is of first order, the hold in range theoretical value is demonstrated to be [35]: ∆ωH = ±KF (0) (3.31) So, in principle, the hold in range can be made arbitrarily large by increasing the DC loop gain. Thus, a PLL with a theoretically perfect integrator (F (0) = ∞), should have an infinite hold in range. In practice, the gain is always limited because of saturation of elements such as amplifiers in the loop and laser tuning range. In the presented case a proportional-integral control is implemented, searching for a maximum of this limit. Please note that the assumptions for the hold in range calculation are that the PLL is locked, i.e. the loop can be linearized.
  • 73. Chapter 3. Lock-In amplifier OPLL architecture 54 3.1.3.2 Pull in range It is the maximum frequency difference between the received signal and the local laser, for which the PLL, initially not locked, is capable to lock and start tracking the phase of the received signal. In a first order loop, the pull in range equals the hold in range [35, 36]. Nevertheless, in the calculation of the pull in range presumes the loop can be unlocked, having to use the full loop equation. Thus, for a second order loop (similar to our case) the solution has to be computed, aided by simulation tools. Precisely, for a generic second order tracking loop with imperfect integrator, it has been found [36] that the pull in margin can be approximated by: 8 2 ∆ωP = 2ξK − ωn (3.32) π And the locking time can be expressed as: (∆ωp )2 Tp ≈ 3 (3.33) 2ξωn Thus, the larger the open loop gain is, the larger the pull in range obtained. Also, another important point is the loop delay influence on the PLL. As the hold in range is determined when the loop is locked, loop delay will not cause any effect to the hold in range. However, the pull in range will be dramatically limited by the loop delay [16]. Nevertheless, the PLL can be optimized to have a pull in range of ∆ωp = 2/τ in case the control filter accomplishes [16]: τ1 4ξωn τ = 2ξωn τ J1 (3.34) τ2 π where J1 is the Bessel function of first kind and order 1, and τ1 , τ2 are the control filter parameters. Since for the calculation of the pull in range the unlocked loop is considered, it is obvious that for our case this will not be exact, as phase ranging and locking will be driven by a Lock-In amplifier. Nevertheless, such expressions will be used as a theoretical limit. 3.1.4 Data crosstalk and cycle slipping effects In this loop, cycle slipping influence arises mainly when trying to overcome the effects of data over the phase ranging signal. In order to avoid these data to phase-lock crosstalk, a full wave rectifier has to be placed between the balanced receiver and the band-pass filter of figure 3.2. But this produces a non desired effect: π-periodic slip cycles are observed, instead of the typical 360◦ cycle slips. This is due to the fact that the full wave rectifier doubles the argument of the ranging singal.
  • 74. Chapter 3. Lock-In amplifier OPLL architecture 55 In order to avoid the effects of this undesired slip-cycling onto the data signal, differential coding will be used. Thus, only one erroneous bit will result every time the data signal is reversed. For the loop delays and laser linewidths considered, the time average between cycle slips is in the order of years [37]. Time average between π-periodic slip cycles is expected to be in the same order. 3.2 Simulations The Lock-In amplified oPLL performances were evaluated by means of computer simula- tions, aiming to analyze: • Phase noise cancellation • Step response performances • Dithering amplitude optimization • Comparison with other loops The first three studies were carried out using Matlab/Simulink, whereas the last one was performed using VPI systems software tools. By now, only the generic system parameters are given, as the specific information for each case can be found in the next subsections. The system was designed to operate at a dithering frequency of 700 MHz assuming a maximum phase error bandwidth of 200 MHz, avoiding the phase noise masking effects described in the introduction of the present chapter. Also the amplitude used was around 50 mrad (2.86◦ ), introducing an additional error phase deviation of 2◦ . So, in order to obtain 11◦ of total error phase deviation (< 1 dB penalty at BER of 10−9 ), a maximum of 10.82◦ from noise influence was allowed. Filters used (bandpass and lowpass) were based on Butterworth approximations because of their narrow transition band and flat modulus band-pass response. However, they introduce a variable delay (2.5 ns as total average) in addition to the loop main delay. A Bessel filter was also studied, because of its flat delay. Nevertheless, as they have very long transition bands, they were discarded. The filter f (t) was a PI control, leading to a second order Phase-Locked Loop. So, design parameters to optimize were damping factor, natural frequency, and loop delay. 3.2.1 Phase noise simulations This subsection presents the simulation of several configurations (damping factor and natural frequency) in order to determine the Lock-In loop limitations for minimizing the
  • 75. Chapter 3. Lock-In amplifier OPLL architecture 56 phase noise. Also these simulations determined the optimal parameters for the loop, only considering laser phase noise. Several sweeps of damping factor and natural frequency were made for different loop main delays (1 ns, 5 ns and 10 ns) and laser linewidths (from 100 kHz to 10 MHz). Details on how the phase noise was digitally modeled in Matlab environment can be found on appendix D. A sample of these simulations is shown in figure 3.4. There it can be seen the time evolution of the phase noise (1 MHz linewidth) and the phase signal introduced by the OPLL. Figure 3.4(b) corresponds to a zoom of figure 3.4(a) between 200 ns and 300 ns, for showing the dithering introduced by the loop. Figure 3.4: Phase noise evolution and phase signal introduced by the loop. Inset (b) is a zoom between 200 ns and 300 ns. The results obtained practically have the same behavior as a decision-driven architecture, which is shown in [16, 17]. E.g. 11.39◦ phase error deviation is obtained for 1 MHz total laser linewidth and a 3.5 ns loop delay optimum PLL; which fits perfectly with the results in [17] (see figure 5). At a fixed linewidth and loop delay, phase noise cancellation is in the same grade at the several optimal configurations, as it is shown in figure 3.5. So, a priori, there are multiple optimal configurations with damping factors over 2. Since all optimal configurations for a given loop (same laser linewidth and delay) have similar phase error RMS values, they have been reproduced in table 3.1 and the BER-floor for each one was plotted as shown in figure 3.6. Regarding table 3.1, it should be noted that for 1 MHz linewidth and 1 ns loop delay a standard deviation of 11.74◦ was obtained, very close to the limit for operating with less than 1 dB power penalty at a BER of 10−9 (see sections 2.3.1.2 and 2.3.1.3). For a BER of 10−3 (FEC limit, phase error deviation < 19◦ ), the 1 dB penalty points are near 2 MHz with 1 ns delay and 1 MHz with 5 ns. BER-floor was obtained by applying the resulting error phase deviation to the expression deduced by [13] for PSK signals (i.e. infinite SNR is assumed), as in sections 2.3.1.2
  • 76. Chapter 3. Lock-In amplifier OPLL architecture 57 Figure 3.5: Loop natural frequency versus damping factor relationship for optimal configurations (transient response and phase noise) with 10 ns main loop delay. Linewidth 1 ns delay 5 ns delay 10 ns delay 1 MHz 11.74◦ 15.05◦ 18.23◦ 2 MHz 17.93◦ 23.35◦ 28.24◦ 3 MHz 23.98◦ 28.34◦ 35.08◦ 4 MHz 28.29◦ 34.69◦ 40.35◦ 5 MHz 32.23◦ 36.42◦ 53.82◦ Table 3.1: Phase error standard deviation for the optimal configurations as a function of linewidth and delay. and 2.3.1.3). Note that 10−9 floor can be achieved with 1 MHz laser linewidth and a loop main delay near 4.5 ns. This means that the loop is at 1 dB penalty for a BER of 4.86 · 10−6 , that with FEC codes can be tolerated. Searching for the limit of FEC codes, the loop can be at 1 dB penalty of a 10−3 BER, with its equivalent BER-floor of 2.04 · 10−6 . For this case 2 MHz can be tolerated with 1 ns loop delay. Conversely, for a large loop delay of 10 ns, a 1 MHz linewidth is allowed. 3.2.2 Time response simulations When a PLL is optimized for a better cancellation of the phase noise, its loop bandwidth is maximized, as seen in section 2.3.2.2. Thus the optimum system for phase noise cancellation is very similar to the optimal one in step response: it will have minimum rise and setting times, while keeping a tolerable overshoot.
  • 77. Chapter 3. Lock-In amplifier OPLL architecture 58 Figure 3.6: BER-floor for optimal configurations as a function of the laser linewidth evaluated at several main loop delays. However, there may be some differences. In order to quantify these differences, several computer simulations were made for evaluating the step response without noise. The step introduced in all the cases was a phase step of 1 rad. Here also several sweeps of damping factor and natural frequency were performed for different loop main delays; as in the previous subsection. Parameters measured were 10%-90% rise and setting times, and maximum overshoot. A sample of these simulations is shown in figure 3.7. There it can be seen the time response of the loop for the proposed phase step of 1 rad. The inset of the figure corresponds to a zoom between 500 ns and 550 ns, showing the dithering introduced by the proposed OPLL. After the proposed sweeps, results confirm the behavior expected: when maximizing the loop bandwidth, phase noise is minimized successfully (minimum phase error deviation) and the step response has an optimum setting time. The behavior of the optimal settings is the same demonstrated in [16] for a decision-driven oPLL limited by its loop delay. This is shown in figure 3.5 for a 10 ns main loop delay. When an optimal configuration is achieved with high damping factor, the setting time and the rise time are very close (see figures 3.8 and 3.9). So, the loop has better time response. Also, for damping factors over 3 the setting time and the rise time converge to a certain value. This is shown in table 3.2. These loops are optimized implementations for phase noise cancellation by using a 2.86◦ (50 mrad) amplitude of dithering signal.
  • 78. Chapter 3. Lock-In amplifier OPLL architecture 59 Figure 3.7: OPLL time response for a phase step of 1 rad. Inset figure is a zoom between 500 ns and 550 ns. Figure 3.8: Setting time of the optimal configurations for several loop main delays. For configurations with a small damping factor and minimum setting time, the maximum overshoot becomes larger. As it is a very narrow overshoot, it does not affect too much the loop frequency response and phase noise can be successfully cancelled. As shown in figure 3.10, the maximum overshoot converges to a minimum value for damping factors over 3 and optimal loop configurations (minimum setting time). So, in maximum overshoot terms, it is desirable to use damping factors over 3 as well.
  • 79. Chapter 3. Lock-In amplifier OPLL architecture 60 Figure 3.9: Rise time of the optimal configurations for several loop main delays. Loop delay Setting time Rise time 10 ns 49.8 ns 45.2 ns 5 ns 31.6 ns 28.9 ns 1 ns 19.6 ns 18.6 ns Table 3.2: Convergence values for setting and rise times, at several loop delays. 3.2.3 Amplitude of the dithering signal Effects of amplitude of dithering signal variation on phase noise cancellation were also studied. Since dithering and phase noise are independent processes, degradation intro- duced by dithering amplitude itself is very low when phase error deviation tends to be high. So, for large loop delays, dithering impact is mostly on the loop gain, as shown in figure 3.11. However, loop gain can be optimized by modifying damping factor per natural frequency product, as in the previous subsections. For the other cases, it introduces an additional error phase deviation of 2◦ , limiting the noise influence tolerance to a maximum of 10.82◦ . 3.2.4 Comparison with other loops The Lock-In amplified oPLL performances were compared to other oPLL architectures: Balanced, Costas, and Sub-Carrier Modulated loops. The four oPLL configurations were simulated, estimating their phase noise cancellation. Several configurations of damping factor, natural frequency and dithering amplitude have
  • 80. Chapter 3. Lock-In amplifier OPLL architecture 61 Figure 3.10: Maximum overshoot of the optimal configurations for several loop delays. Figure 3.11: Phase dithering effect for large loop delays. been simulated in order to determine the loop limitations when cancelling the phase noise. Precisely, for each loop, the damping factor now was set to 9, assuring an overdamped performance, since it has been demonstrated to be optimum when designing loops with large delay [16]. This is the case for the Lock-In loop implemented, using non-integrated off the shelf components. In each case, the loop natural frequency was optimized in terms of output phase error. These simulations determined the optimal designs for each loop type. So, several sweeps
  • 81. Chapter 3. Lock-In amplifier OPLL architecture 62 of damping factor and natural frequency were made for different loop delays and laser linewidths. Since 10 ns (equivalent to 20 cm of optical fiber) is an easily implementable delay when regarding a laboratory prototype, results for optimal configurations at 10 ns loop delay are shown in figure 3.12. Figure 3.12: Phase error deviation evaluated at a loop delay of 10 ns. From the results, it can be seen that at low linewidths (below 1 MHz at 10 ns loop delay), the Lock-In amplified loop mostly has an intermediate performance between the Costas loop and the balanced loop. Thus when using the Lock-In amplified loop, for a 11◦ maximum phase error deviation the maximum linewidth limit is of 675 kHz, near the balanced loop limit. When using PSK modulation, that phase error of 11◦ limits to operate at a 1 dB penalty for BER of 10−9 , as shown in sections 2.3.1.2 and 2.3.1.3. On the other hand, if FEC codes are used, a BER of 10−3 is operable. Then, its 1 dB penalty point (19◦ of phase error deviation) can be found at a linewidth of 3.1 MHz. In this case, the Lock-In amplified loop clearly outperfoms the most advanced loops, such as the sub-carrier modulated loop. This behaviour is due to the inclusion of the Lock-In amplifier in the loop. At low linewidths, it is mostly limited by the dithering amplitude, so its performances are near the balanced loop. However, at high linewidths, when the dithering amplitude is relatively negligible, the loop performances are improved by the Lock-In amplifier, that ensures a better phase ranging. For proper comparison between loops, the values of the most important points of figure 3.12 have been written in table 3.3. After phase noise cancellation simulations, other important parameters were also evalu- ated: hold in and pull in ranges. The results are shown in figures 3.13 and 3.14.
  • 82. Chapter 3. Lock-In amplifier OPLL architecture 63 Figure 3.13: Pull in margins of the simulated architectures. Figure 3.14: Hold in margins of the simulated architectures. In these figures, it is depicted that the Lock-In amplified loop has low pull in and hold in ranges. This is its main drawback. Precisely, at 10 ns loop delay, the pull in range is found to be around 20 MHz in front of the 176 MHz achieved by the subcarrier architecture. Concerning the hold in range, the exact data is 896 MHz for the Lock-In amplified loop, and up to 7.68 GHz for the subcarrier.
  • 83. Chapter 3. Lock-In amplifier OPLL architecture 64 Balanced Costas SCM Lock-In amplified Linewidth tolerance BER 10−9 420 kHz 1.15 MHz 1.35 MHz 675 kHz Linewidth tolerance BER 10−3 1.2 MHz 2.65 MHz 2.75 MHz 3.1 MHz Pull in margin 19 MHz 72 MHz 176 MHz 20 MHz Hold in margin 1.28 GHz 2.55 GHz 7.68 GHz 896 MHz Table 3.3: Table summarizing results at 10 ns delay. 3.3 Experiments and discussion A laboratory prototype of the proposed PLL was developed and assembled into an ex- perimental setup (Figure 3.15). Figure 3.15: Experimental Setup. An external cavity tunable laser was placed at the transmitter (Tx) side, while at the receiver side (as local oscillator) there was a standard DFB laser (Panasonic LNFE03) running at 1544.07 nm, with a measured sensitivity of 1.4 GHz/V. The local oscillator laser linewidth was measured by using a self-homodining technique [38]. Results are shown in table 3.4. The optimum point was found to be 120 mA, giving a laser linewidth of 833 kHz. As the linewidth specification of the external cavity laser was of 150 kHz, the total linewidth for the operating OPLL was of 983 kHz. Laser current Linewidth Laser current Linewidth 30 mA 1.66 MHz 150 mA 1 MHz 50 mA 1 MHz 165 mA 1 MHz 70 mA 916 kHz 175 mA 1 MHz 90 mA 916 kHz 200 mA 1 MHz 100 mA 833 kHz 225 mA 1.1 MHz 120 mA 833 kHz 250 mA 1.17 MHz 135 mA 916 kHz 300 mA 1.36 MHz Table 3.4: Measured values of the local oscillator linewidth.
  • 84. Chapter 3. Lock-In amplifier OPLL architecture 65 Figure 3.16: Electrical power spectrum after photodetection. The Rx laser output was fusion spliced with the optical coupler and the photodetector (a standard PIN diode). The balanced detector was substituted by a single photodetector because of the need to monitor optical signals and also achieve relative low loop delay. A printed circuit board was fully engineered and prototyped, containing the electrical parts of the OPLL: bandpass and lowpass filters, electrical oscillator (VCO), electronic phase shifter, RF mixer and PI filter. The parameters for the Lock-In amplified loop prototype were optimized for the 700 MHz dithering frequency. Filters placed inside PLL board were designed and implemented to introduce the same delay as in simulations (around 3 ns). Finally, the total loop delay was measured using a vectorial network analyzer, and found to be 10 ns. Locking was observed by tuning one of the lasers until the main beat signal was about 20 MHz, in agreement with the pull in range simulation results. Concerning the hold in range, it was found to be 868.24 MHz, also in agreement with the simulations. Figures 3.16 and 3.17 depict the spectra at the output of the photodetector before and after the phase-locking is achieved. In figure 3.16 it is clearly shown that the peak due to the frequency difference between the local oscillator generates a replica modulated at the frequency of the electrical VCO. Nevertheless, the most interesting spectrum is the one depicted in figure 3.17. From this spectrum, the phase error standard deviation could be calculated using the procedure described in [39]. Using Matlab, the photodetected spectrum could be integrated up to 200 MHz, and afterwards divided by the proposed constant of RL PS PLO . As all these values where known, it was easy to find the phase error deviation. Precisely, it was of
  • 85. Chapter 3. Lock-In amplifier OPLL architecture 66 Figure 3.17: Electrical power spectrum after photodetection. 11.49◦ , for a measurement bandwidth of 200 MHz. This value fits perfectly into the Lock-In amplfied loop curve of figure 3.12, confirming again the theoretical calculations. So a BER near 10−9 could be achieved when working with this configuration. 3.4 Chapter summary In this chapter a novel OPLL concept was introduced and demonstrated, based in Lock-In amplification of optical phase error. This architecture uses simple optical components, being especially indicated for low linewidth DFB commercial lasers, and avoiding the use of the phase-critical optical 0◦ /90◦ hybrids. While phase noise swings quite slowly, phase governing band can fill under DFB laser FM response dip and dithering will be above it. After a theoretical analysis, simulations and experiments were performed: Loop was opti- mized in front of phase noise, taking into account the time response, dithering amplitude effects, acquisition parameters, and comparison with other loops. At first glance (theoretically) its phase noise performances are similar as the Decision- Driven optical PLL: With a main delay over 4.5-5 ns, the 10−9 BER 1 dB penalty point cannot be achieved for linewidths worse than 1 MHz. Nevertheless, after advanced simulations confirmed by experimental results, it was found that with a more realistic delay value of 10 ns this loop can achieve a 10−9 BER with 675 kHz linewidth. In case FEC codes are used, a BER of 10−3 can be tolerated, en- hancing the linwidth tolerance up to 3.1 MHz and clearly outperforming the other loops
  • 86. Chapter 3. Lock-In amplifier OPLL architecture 67 (Decision driven, balanced, costas and subcarrier modulated). Regarding the acquisiton parameters, pull in and hold in ranges were found to be 20 MHz and 868.24 MHz, respec- tively. Thus, a very low range guarantees acquisiton, whereas a much wider is ensuring the PLL tracking. Finally, it should be noted that the unique characteristics of such loop make it easy to embed it onto an integrated semiconductor optical circuit. In that case, the loop delay can be dramatically reduced, thus improving OPLL performances.
  • 87. Chapter 4 Advances in phase and polarization diversity architectures Even though the PLL proposed in chapter 3 has shown good performances against the loop delay and laser linewidth, its achievements are very poor for being used with the majority of commercially available lasers (linewidths in the order of tens of MHz). Thus, the need to explore the phase diversity techniques was found, as a solution to the phase noise problem. Nevertheless, the architectures proposed until now (shown in section 2.3.3) need a full 90◦ hybrid, which is still a fairly complex device. Nevertheless, in this chapter two main approaches will be enhanced for achieving the phase diversity: Full phase diversity and time-switched phase diversity. First the full phase diversity scheme (using a 90◦ hybrid) will be improved, proposing a novel algorithm for phase estimation based in the Karhunen-Lo`ve series expansion [40], e clearly outperforming the Wiener filter estimation seen in section 2.3.3.2. Secondly a new phase diversity technique based in time switching is proposed, using more simple optical devices [41, 42]. The idea is to divide each bit into two time slots: one for detecting the I component and the other one for detecting the I component. As it is a concept with enough novelty, it has been patented [43, 44]. For this second approach a first implementation based on differential detection will be experimentally demonstrated [41], while using a phase modulator (driven by the recovered data clock) after the local oscillator for achieving the time switching diversity. Furthermore, two variations of this time switching phase diversity will be analyzed: one based in Fuzzy logic data estimation and another one simplifying the receiver architecture. For the first one, a heuristic data estimation method based on Fuzzy logic will be intro- duced and extensive simulations will be performed, showing a phase noise robustness near the Wiener filter performances [45]. Regarding the simplified receiver approach, the idea is to drive the local laser directly with the filtered data clock to achieve the time switching 69
  • 88. Chapter 4. Advances in phase and polarization diversity architectures 70 diversity [46]. Also, simulations and proof-of-concept experiments will be performed for this case. Finally, the natural extension of the time switching concept to the polarization diversity will be theoretically analyzed, showing its feasibility [47]. 4.1 Full phase diversity 4.1.1 Karhunen-Lo`ve series expansion phase estimation e In chapter 2, the phase and polarization diversity architectures were introduced and several algorithms (including differential detection) for phase estimation have been seen. The most remarkable are the ones based in Wiener filter and Viterbi-Viterbi algorithms, that are becoming more popular due to the improvement of digital signal processors. With such algorithms one is allowed to use lasers on the order of tens of MHz, for a data rate of 1 Gb/s, representing a big improvement respect the OPLLs. Here a phase estimation method based on Karhunen-Lo`ve series expansion is presented. e It can be implemented using standard DSP devices, since its complexity is not very high. Also, it clearly outperforms the Wiener filter algorithm. 4.1.1.1 Receiver scheme A scheme of the receiver to be used is a typical intradyne architecture, shown in figure 4.1. There, the optical input signal is interfered with a free-running optical local laser in a 2x4 90◦ hybrid. The output signals of the hybrid are then detected by two balanced de- tectors, and the I and Q outputs are digitized by an Analog-to-Digital Converter (ADC). Finally, a Digital Signal Processing (DSP) module performs the phase estimation and data detection. Figure 4.1: Scheme for a standard intradyne receiver.
  • 89. Chapter 4. Advances in phase and polarization diversity architectures 71 4.1.1.2 Phase estimation algorithm In section 2.3.1.4 is shown that laser phase noise characterized by a certain spectral width (∆ν) can be modeled as a Wiener process (w(t)). As a random process, the Wiener process can be expanded into a Karhunen-Lo`ve (KL) series form: e ∞ w(t) = cn ϕn (t) (4.1) n=1 where ϕn (t) is a set of orthonormal functions (eigenfunctions) in the interval (0, T ), and cn are the series coefficents, being random variables. For the Wiener process case, it is shown that [14]: 2 ϕn (t) = sin(ωn t) (4.2) T T 2 cn = w(t) sin(ωn t) (4.3) T 0 2π∆ν (2n+1)π where ωn = λn = 2T being the eigenvalues defined as the variance of the series coefficients: 8T 2 ∆ν λn = E{c2 } = n (4.4) (2n + 1)2 π Since larger eigenvalues are those of lower n, the series can be truncated at a relatively short number of terms, M . Precisely, it can be easily shown that λ1 is almost 20 times λ5 , then truncating at M = 5 should be enough. Please note that for M = 5 and M = 15 the output phase error deviation of the estimator is almost the same, independently of the time interval squared per spectral width product. This is shown in figure 4.2, for the proposed performances evaluation. In order to estimate and cancel the laser phase noise, what is proposed is to process the received phase block by block. Thus, a priori, the observable interval length T is known, and the eigenfunctions can be easily calculated. Consequently, the phase noise can be estimated as:   ϕ1  ϕ2    w(t) = c ϕ = c1 c2 c3 . . . cn  ϕ3  ˆ ¯ ¯ (4.5)    .   .  . ϕn Equation 4.4 shows that the lower the interval squared per spectral width product is, the lower the eigenvalues. Also, inside a block, the phase noise will have a limited variance of
  • 90. Chapter 4. Advances in phase and polarization diversity architectures 72 2π∆νT [14]. So for small blocks and same spectral width, the estimator will work better. That is the reason why it is proposed to work with blocks of only 1 symbol. Figure 4.2: Phase error deviation as a function of time interval squared per spectral width product (T 2 ∆ν). From one block to another, phase noise is expected to be highly correlated as it is a slow process compared to data. Thus, the series coefficients also present small changes from one block to the next one. So, coefficients are going to be calculated by using an adaptive algorithm. In this case the Least Mean Square (LMS) method will be used. The estimation algorithm diagram used in the following section is the one depicted in figure 4.3. There, the M sinusoidal waveforms needed for phase estimation are stored in the DSP memory. Next, a coefficient (coming from the LMS estimation block) is applied to each of the waveforms. The sum of all the waveforms multiplied by the coefficients results to be the estimated phase, and the phase error is used as the LMS input. 4.1.1.3 Algorithm performances and discussion In order to evaluate the performances of the proposed receiver architecture, firstly a BPSK data stream was simulated, running at 10 Gb/s with variable phase noise. The linewidths ranged from 100 kHz to 10 GHz. Phase noise was modeled using the technique proposed in appendix E. Such a phase noise was used to compare the proposed algorithm performances with respect to a Wiener filter with a lag of 10 symbols (optimum), as described in [26] and in chapter 2. In both cases a data stream of 220 = 1048576 symbols was simulated, and the received signal was resampled at 16 samples per symbol. Then, the digitized I and Q signals were used to reconstruct the received optical signal and its phase was packet into blocks of 16 samples (one symbol). The phase error deviation after
  • 91. Chapter 4. Advances in phase and polarization diversity architectures 73 Figure 4.3: Block diagram of the phase estimation algorithm. phase estimation was used as a measure for the quality of the algorithm. Results are shown in figures 4.2 and 4.4. Figure 4.4: Phase error deviation as a function of the spectral width per bitrate ratio. Regarding the term where the KL expansion is truncated, in figure 4.2 is shown that there is almost no difference on the estimation for M = 5 and M = 15; so the best decision was to keep it at 5. From figure 4.4, it is shown that when using the KL series estimation for a 11◦ maximum phase error deviation (1 dB penalty for a BER of 10−9 ) the system is limited, working at a maximum spectral width of 4% of the bit rate. Also, if Forward Error Correction
  • 92. Chapter 4. Advances in phase and polarization diversity architectures 74 (FEC) codes are used and a BER of 10−3 is operable with only 7% overhead [11], and a maximum phase error deviation of about 19 is allowed for less than 1 dB penalty, leading to a maximum spectral width of 11% of the bit rate. 4.2 Time switched phase / polarization diversity This section presents a novel phase/polarization diversity receiver based on a time switch- ing between phase and (or) polarization orthogonal states. The general idea is to have the orthogonal states for phase and polarization in the same bit, not concurrently but sequentially. Thus, for the first half of the bit the signal relative to one orthogonal component (I or H) will appear, whereas for the second half of the bit the signal relative to the other orthogonal component (Q or V) will be seen. First the time switched phase diversity will be explained, with all of its approaches: phase diversity with differential detection, fuzzy logic data estimation, and simplified diversity with laser direct drive. Finally, analyze the time switching polarization diversity will be analyzed. 4.2.1 Phase diversity combined with differential detection The receiver has two main parts: the first is a homodyne coherent photo-receiver with an Automatic Wavelength Controller (AWC), instead of an OPLL, with an added phase modulator (PM) at the local laser output, and the second is the electrical post-processing. The coherent photo-receiver mixes the incoming optical field with the local laser carrier in the balanced photo-detector stage. The optical phase modulator at the local laser output is controlled by the data clock (50% duty cycle) producing a fixed 0◦ − 90◦ phase modulation, to obtain the I and Q signal components, at the first and second half part of each Tb (bit time) respectively, after the optical homodynation. By gating the photo-receiver output with the data clock signal and its inverse, the I and the Q components are obtained separately in the two branches, now with RZ shape. The electronic signal processing performs differential demodulation of both I and Q compo- nents, with a delay time equal to Tb ; and the synchronous combination of the I and Q components. Given the post-processing scheme, lets analyze how it can improve the data detection. Also, in order to synchronize data signals at the two branches, the I com- ponent is delayed by half of a bit time, and, finally, both demodulated components are synchronously combined with an adder. The signal power fluctuates between the I and the Q branches randomly, due to the phase noise, at a rate of the order of the laser linewidth, and the combination of both outputs
  • 93. Chapter 4. Advances in phase and polarization diversity architectures 75 Figure 4.5: Time-switched diversity differential homodyne receiver scheme. can assure its recovery. This operates like a phase-diversity system. Compared to an OPLL, the phase noise swing time has been shortened from the loop delay to the bit time, which, in contrast, reduces with the bit rate. The local laser does not need to be phase coherent with the incoming optical carrier, al- though an automatic wavelength controller is convenient to maintain the two wavelengths close each other. It can be regarded as intradyne or heterodyne receiver with near-zero intermediate frequency. At the transmitter, the optical carrier is feedforward PSK mod- ulated. The NRZ data is previously differentially coded; thus, the RX decoded data is d(t) = d(t) ·d(t−Tb ) . By performing the calculation of the optical homodynation process and the electrical post-procesing, the signal can be obtained at the receiver output, where the laser phase noise is considered as the limiting impairment, C2 Vout (t) = d(t) cos φe (t) − φe (t − Tb ) + cos φe (t) + φe (t − Tb ) + 2 + cos φe (t − Tb /2) − φe (t − 3Tb /2) − cos φe (t − Tb /2) + φe (t − 3Tb /2) (4.6) √ where C = 2 RL PS PLO is the peak voltage available at the output of the photo-receiver and φe (t) is the TX+LO laser total phase noise, which can be in the margin [-π, π], producing signal amplitude reduction. The result is governed by the noise decorrelation in the Tb /2 and Tb . This signal Vout (t) is fed back to the AWC of the local laser, with dithering electronics, to track the wavelength, maximizing the output amplitude. It is
  • 94. Chapter 4. Advances in phase and polarization diversity architectures 76 not affected by the loop delay: the I/Q post-processing reduces the error phase down to its bit time variations. 4.2.1.1 Expected system performances The scheme proposed is difficult to implement mainly because of the lack of high speed switches, the high number of four quadrant multipliers used for differential detection, and the electrical branches synchonization. Thus, a simplified version of the receiver has been developed (referred as second approach) and will be deeply analyzed in the next subsections. Its scheme is shown in figure 4.10. The proposed receiver (first and second approaches) is compared to a lock-in amplified OPLL homodyne receiver. In these two cases, the output parameter measured was the statistical eye-opening of the received data, accounted as the mean amplitude minus twice the standard deviation, which provides a fair measure of the sensitivity penalty. The eye-opening estimation was made at several linewidths from 225 kHz to 1 GHz. Bit rate was set at 10 Gb/s; and a 4th-order 7.5 GHz Bessel low-pass filter was placed before decision. Figure 4.6 shows an example of the time traces of the signals Vouti , Voutq and Vout , from figure 4.5. Figure 4.6: Example of the time diversity operation, from scheme shown in figure 4.5. Blue line is Vouti , green line is Voutq and red line is Vout after filtering. Figure 4.7 shows the resulting eye-diagrams for a large linewidth (50 MHz), for the I and Q branches and the combined output. While I and Q branches have their eye totally closed, the post-processing output has a similar opening as for very low linewidths. The resulting system tolerance to the laser phase noise is depicted in figure 4.8. It shows the normalized eye opening measured as a function of the laser linewidth, showing that the phase noise tolerance can be greatly extended, from 1 MHz to 100 MHz (for about 2dB sensitivity penalty).
  • 95. Chapter 4. Advances in phase and polarization diversity architectures 77 Figure 4.7: I, Q, and I+Q outputs Eye-diagrams, at 50 MHz total laser linewidth. Figure 4.8: Statistical normalized eye opening (20Log) for the I/Q receiver (both first and second approach) and a lock-in oPLL. Also, the I/Q processing demonstrates to be insensitive to the loop delay and to the wavelegth drift (up to 10%Rb ), and does not require fast tuning lasers. These results have been validated through extensive numerical simulations, and are shown in figure 4.9. 4.2.1.2 Simplified scheme and phase noise analysis A simplified version of the time-switch phase diversity receiver is shown in figure 4.10. In this case, the proposed diversity receiver has two main parts, too. The first is a coherent photo-receiver with added clock-synchronous phase switching (0◦ − 90◦ ). The second part is an electronic postprocessing of the signal demodulation and synchronous combination
  • 96. Chapter 4. Advances in phase and polarization diversity architectures 78 Figure 4.9: Statistical normalized eye-opening (20log) for the I/Q receiver (both first and second approach) as a function of the laser frequency drift. of the orthogonal components. At the transmitter, the digital data modulate the optical phase [0◦ -180◦ ], after a differential precoder (differential phase shift keing). Figure 4.10: Receiver scheme for phase noise analysis. The electronic signal processing stage performs the parallel differential demodulation. Next, a delay-and-add block is placed. It properly combines the orthogonal components. As a result, the same theoretical results as those obtained in [24] were retrieved in the present case, except for the extra Tb /2 delay. Besides, it must be taken into account that the reduced duty-cycle and the correspondingly increased electrical bandwidth produces a 3dB signal-to-noise ratio penalty. After the photo-detection stage, typical differential phase shift keying data that is cor- rupted by phase noise describe a circle in the IQ plane when the data are represented as a phasor. When differential demodulation is performed, a change of phasor bases takes
  • 97. Chapter 4. Advances in phase and polarization diversity architectures 79 place; data points affected by the phase noise then fall either at the first or the third quadrant of the IQ signal plane depending on the data value (1 or 0). The receiver decoded data are d(t) = d(t) · d(t − Tb ) , where d(t) is the differentially precoded version of data. The combination of the detected I and Q components after differential demodulation can be written as: C2 Vout = 2 d(t) cos φe (t) − φe (t − Tb ) − (4.7) − cos φe (t) + φe (t − Tb ) + + cos φe (t − Tb /2) − φe (t − 3Tb /2) + + cos φe (t − Tb /2) + φe (t − 3Tb /2) √ where C = 2 RL PS PLO is the peak voltage available at the output of the photo- receiver, RL is the receiver impedance, is the photodiode responsivity, and φe (t) is the total laser phase noise (transmitter plus local), which is in the margin [−π, π], producing signal amplitude reduction. Since the data are differentially coded, in this expression the relevant terms are those relative to the phase-noise difference. After some trigonometric algebra, this expression can be properly expanded for evaluating the phase noise effects: C2 ∆φ1 (t) + ∆φ2 (t) ∆φ2 (t) + ∆φ3 (t) Vout = d(t) 1 − sin2 − sin2 + (4.8) 2 2 2 1 ∆φ1 (t) + ∆φ3 (t) + cos 2φe (t) 2 cos ∆φ1 (t) + ∆φ2 (t) sin2 + 2 2 + sin ∆φ1 (t) + ∆φ2 (t) sin ∆φ1 (t) + ∆φ3 (t) + 1 ∆φ1 (t) + ∆φ3 (t) + sin 2φe (t) 2 sin ∆φ1 (t) + ∆φ2 (t) sin2 − 2 2 − cos ∆φ1 (t) + ∆φ2 (t) sin ∆φ1 (t) + ∆φ3 (t) where the phase error on each half bit has been defined as: Tb ∆φ1 (t) = φe (t) − φe t − 2 Tb ∆φ2 (t) = φe t − − φe (t − Tb ) 2 3Tb ∆φ3 (t) = φe (t − Tb ) − φe t − 2
  • 98. Chapter 4. Advances in phase and polarization diversity architectures 80 Figure 4.11: BER-floor of several cases: theoretical (dashed line), theoretical but including the penalty due to phase switching (dotted line), numerical simulation (con- tinuous line) and measurements (square points). As phase noise is a modeled as a Wiener process, its differences in time are Gaussian random processes [14]. So, the three arguments, ∆φ1 (t), ∆φ2 (t) and ∆φ3 (t); are indepen- dent identically distributed Gaussian random processes with zero mean and a variance proportional to the linewidth per half of bit-time product. Thus the terms containing φe (t), tend to 0 as they are multiplied by sine terms near 0, and the only remaining terms will be near 1. ∆φ3 (t) is an added delay due to the fact that time-switched architecture is used instead of the architecture proposed in [24]. So, in the ∆φ3 (t) = 0 limiting case, a BER-floor can be easily calculated in equation 2.29, and this theoretical limit is shown in figure 4.11. An early evaluation of the receiver performance was carried out by means of numerical simulations. This was calculated from the same equation 2.29 used before but including additive noise effects due to the use of the phase scrambler. As shown in figure 4.11, the calculated results differ by only 3 dB from the theoretical calculations, as expected. Since in this type of system the phase scrambling is done on the receiver side, the system is less affected by fiber impairments, such as chromatic dispersion, than an alternative approach [48]. It is potentially useful to recover the orthogonal I and Q components of the complex received signal (phase and magnitude), opening the door to the electronic compensation of transmission impairments by means of new digital signal processing techniques. An experimental prototype of the proposed receiver has been developed and was tested in a laboratory setup. Sensitivity and frequency drift impacts were evaluated. Although the proposed receiver architecture was demonstrated to be highly insensitive to phase
  • 99. Chapter 4. Advances in phase and polarization diversity architectures 81 noise effects, low linewidth lasers were initially used; the total linewidth was 350 kHz, much smaller than the bit rate. Two external-cavity tunable lasers were used, one on each branch (Transmitter, TX, and local oscillator, LO) of the setup (see figure 4.12). Figure 4.12: Experimental setup The TX branch was binary phase shift keying modulated with a Mach-Zehnder Modulator (MZM, a Fujitsu FTM7921ER) at 1 Gb/s. The local oscillator branch was 0◦ − 90◦ mod- ulated with a standard LiNbO3 phase modulator (PM, a JDSU IOAPMOD9183), driven by the clock signal (CLK) of the pseudo-random binary sequence (PRBS) generator used to provide data at the first branch. These two optical signals are coupled and detected by a balanced detector followed by an electrical processing stage (a delay-and-multiply mod- ule plus a delay-and-add stage, as shown in figure 4.10). A microwave double-balanced mixer (Marki Microwave M20004) was used as a four-quadrant multiplier. Firstly, sensitivity measurements were carried out, obtaining −38.7 dBm for BER= 10−9 (see figure 4.13). As the Tx laser used was a high power laser (about 9 dBm), a power budget of 47 dB could be reached. For the 10−3 BER, the sensitivity was found to be at −44.3 dBm, achieving 53.3 dB of power budget. Finally, in order to evaluate the impact of the critical phase noise on our system, the transmitting laser was changed by a standard linewidth distributed feedback laser (Fujitsu YM004). Varying the laser’s bias, the total laser linewidth ranged from 18 MHz up to 30 MHz, yelding a BER-floor of from 10−3 to 10−2 . These linewidth values (1.8%-3% Rb ) are lower than the theoretical limit (3.6%-5.9% Rb ) calculated in [13] and the simulation results, as shown in figure 4.11. The difference between the experimental points and the theoretical curve is explained by the frequency response distortion of the microwave mixers and the 3 dB penalty due to phase switching.
  • 100. Chapter 4. Advances in phase and polarization diversity architectures 82 Figure 4.13: Sensitivity results and output eye-diagram 4.2.1.3 Frequency drift For the frequency drift impact evaluation, equation 4.7 can be revisited. Assuming no phase noise, φe (t) will be only contributed by the frequency drift. For a generic frequency drift fd , equation 4.7 can be rewritten as: C2 Vout = 2 d(t) cos 2πfd Tb − (4.9) − cos 4πfd t + 2πfd (t − Tb ) + + cos 4πfd Tb + + cos 2πfd (t − Tb /2) + 2πfd (t − 3Tb /2) After some algebra, the expression can be rearranged as: C2 Vout = 2 d(t) 2 cos 2πfd Tb − (4.10) − cos 2πfd t + 2πfd Tb + + cos 4πfd t − 4πfd Tb Note that the cos(2πfd t + 2πfd Tb ) and cos(4πfd t − 4πfd Tb ) terms are running at much lower speed than the bit rate and, in the worst case condition, its argument can be approximated by a uniformly distributed random variable ranging from −π to π. In this case, these processes can be approximated by its mean [12]. Thus, these last cosine terms cancel each other. Then, the output voltage of the add block is approximated by: C2 Vout ≈ d(t)2 cos 2πfd Tb (4.11) 2
  • 101. Chapter 4. Advances in phase and polarization diversity architectures 83 Using the proposed model and taking into account that data is differentially encoded, the probability of error can be written as: 1 −SNR · cos2 2πfd Tb Pe = exp (4.12) 2 2 Numerical evaluation of such expression is shown in figure 4.14. There it is shown that for Pe = 10−9 and no frequency shift a minimum SNR of 16 dB is required, whereas for Pe = 10−9 and a frequency shift of 10% (100 MHz at 1 Gb/s) a minimum SNR of 17.9 dB is needed. Figure 4.14: Modeled BER as a function of the laser frequency drift per bitrate ratio. To experimentally evaluate the detuning tolerance between transmitter laser and the receiver local laser, the setup of figure 4.12 was also used. There, a manual frequency drift was carried out around 1550 nm. As it is shown in figure 4.15, a drift of about 60 MHz (6% of the bit rate) leads to 10−3 BER, that can be considered as the operating limit (using forward error correction codes, equivalent to 10−9 ).
  • 102. Chapter 4. Advances in phase and polarization diversity architectures 84 Figure 4.15: Measured BER as a function of the laser frequency drift. 4.2.1.4 Channel spacing Several adjacent channel interference studies have been carried out regarding coherent system [10, 49], but it has never been done in this homodyne system. It is shown that for a generic optical receiver, assuming only additive white Gaussian noise, the optical power penalty due to channel crosstalk can be expressed as: SNR penalty(dB) = −10 · log 1 − (4.13) SIR where SNR is the electrical signal to noise ratio for the reference Bit Error Ratio (BER), and SIR is the electrical signal to interference power ratio. This SIR mostly depends on the architecture of the receiver, the modulation format used, and the power difference between main channel and those undesired. In order to accurately calculate this SIR, next an analysis of the signals detected will be performed. Please note that in this analysis the phase and additive noises will not be taken into account. Figure 4.16 shows a block diagram of the time-switched phase-diversity differentially encoded BPSK receiver under study. In an equally-spaced multiple channel environment, it satisfies the following set of equations: N eS (t) = PS exp(j(ωt + 2πDit + φSi (t))) (4.14) i=−N eLO (t) = PLO exp(j(ωt + φLO (t))) (4.15)
  • 103. Chapter 4. Advances in phase and polarization diversity architectures 85 Figure 4.16: Time-Switched Phase-Diversity DPSK receiver for channel spacing study. π φLO (t) = p(t) (4.16) 2 where eS (t) is the transmitted optical field; eLO (t) is the local oscillator optical field; 2N + 1 is the total number of channels; PSi and φSi (t) are the transmitter optical power and the phase coded data [0, π] of the ith channel, respectively; D is the channel spacing; PLO is the local oscillator optical power; φLO (t) is the phase introduced by local laser; and p(t) is a pulse train [0, 1] at same frequency than bit-rate used for phase scrambling. Please note that this notation implies an odd number of channels, where the central channel is the desired one. After the balanced detection stage, the photo-detected current can be written as: N π Ip (t) = 2 PSi PLO cos φSi (t) + 2πiDt − p(t) + n(t) (4.17) i=−N 2 where is the detector responsivity and n(t) is the overall noise process. Let us as- sume that all channels have the same bit rate and are BPSK modulated by statistically independent sources, and further calculate the power spectral density (PSD) of the mod- ulated signal in the desired channel. Before calculating that PSD, the modulation signal pulses shape (including p(t)) should be carefully analyzed. Assuming ideal pulses and data alphabet [0, π], Ip (t) can be also expressed as: N Ip (t) = 2 PSi PLO cos g1 (t)φSi (t) · cos 2πiDt + n(t) (4.18) i=−N
  • 104. Chapter 4. Advances in phase and polarization diversity architectures 86 1 0≤t≤ T 2 where g1 (t) is defined as when t ∈ [0, T ], g1 (t) = 1 T g1 (t + T ) = g(t) 2 2 <t≤T being T the bit time (the inverse of the bit rate). Since cos(π) = −1, cos(0) = 1 and cos(±π/2) = 0, it can be rewritten as: N Ip (t) = 2 PSi PLO · g2 (t) · d · cos 2πiDt + n(t) (4.19) i=−N where d is defined to be 1 when φSi (t) = 0; and -1 when φSi (t) = 0. 1 0≤t≤ T 2 Also g2 (t) is defined as: when t ∈ [0, T ], g2 (t) = g2 (t + T ) = g(t) 0 T <t≤T 2 g1 (t) and g2 (t) waveforms are shown in figure 4.17. Please note that g2 (t) has a triangular shaped autocorrelation (R2 (t)), but defined in the [−T /2, T /2] interval. This is also shown in figure 4.17. The ideal autocorrelation of a differentially coherent BPSK signal is a triangular function defined in the [−T, T ] interval [9]. Since the PSD is defined as the Fourier transform of the autocorrelation function, Ip (t) has the same PSD as a general differentially coherent BPSK, but spread with a factor of 2. This is shown in figure 4.18. Figure 4.17: g1 (t) and g2 (t) pulse shapes and autocorrelation of g2 (t), R2 (τ ) Then, from equation 4.19, the double-sided PSD of Ip (t) is found to be: N 2 T πT (f − iD) Gp (f ) = 4 PS0 PLO sinc2 + N (f ) (4.20) i=−N 4 2 where N (f ) is the PSD of n(t). Note that all channels are assumed to have the same power. Then, the SIR is defined as the power of the useful channel versus the power of the interferent channels. Taking into account only the terms 0 and 1, the SIR can be
  • 105. Chapter 4. Advances in phase and polarization diversity architectures 87 Figure 4.18: Spectrum after photodetection: Ideal homodyne reception (a) and using time-switched phase-diversity (b) calculated from If (t): +∞ πT f −∞ |H(f )|2 sinc2 2 df SIR ≈ +∞ (4.21) πT (f −D) −∞ |H(f )|2 sinc2 2 df where H(f ) is the transfer function of the channel filter placed after photo-detection stage, depicted in figure 4.16. Here the total SIR is approximated to the signal-to-strongest- interference (due to the adjacent channel). If channel spacing penalty is calculated using equation 4.13 with SIR values provided by equation 4.21, results do not resemble so much to experimental results (Figure 4.20) because it is a very rough approximation. Thus, a more precise model must be developed in the time domain. Let us assume that a signal sample is taken at the output of the channel filter, plus an interference signal which has a phase θ(t), as shown in figure 4.19. This phase can be written as: π θ(t) = φS1 (t) + 2πDt + p(t) (4.22) 2 where the signal-to-strongest-interference approximation is applied. θ(t) varies rapidly during the symbol interval (at speed near D). So its increments between consecutive symbols represents a residual filtered phase that can be regarded as a random phase process uniformly distributed from −π to +π along the bits. In the proposed receiver, decision stage input samples are signal plus interference vector projections onto In-phase straight line. Thus, the decision variable, X, can be expressed as: 1 ˆ X ∝d 1+ √ cos(θ) (4.23) SIR
  • 106. Chapter 4. Advances in phase and polarization diversity architectures 88 Figure 4.19: Complex representation of signal samples including interference. Figure 4.20: Sensitivity penalty due to channel crosstalk. Square points are experi- mental data. ˆ where θ is the residual phase after filtering θ(t) increments between two consecutive symbols. Thus, for a DPSK signal [10], the error probability conditioned to the value of ˆ θ in the optimum sampling instant is found to be: 2 ˆ 1 SNR 1 ˆ Pe (θ) = exp − 1+ √ cos(θ) (4.24) 2 2 SIR
  • 107. Chapter 4. Advances in phase and polarization diversity architectures 89 Finally, the overall bit error rate can be written as: +π 2 1 SNR 1 ˆ ˆ Pe = exp − 1+ √ cos(θ) dθ (4.25) 4π −π 2 SIR This BER expression was used to evaluate by means of numerical calculations the pro- posed receiver performances. It was also compared to the Gaussian approach (based on equation 4.13), and to an ideal coherent receiver, described in [10]. In these three cases, the output parameter measured was the 10−9 BER sensitivity penalty due to channel crosstalk. The penalty crosstalk measures were made at several channel spacing, from 1 GHz to 6 GHz. Bit rate was 1 Gb/s; and a 4th-order 2 GHz Bessel low-pass filter was placed after photo-detection stage. For the ideal system, that filter was also 4th-order Bessel, but with 750 MHz bandwidth. The resulting system tolerance to the channel crosstalk is depicted in figure 4.20. It shows that for a 1 dB penalty the minimum spacing between channels is around 3 GHz for both, Gaussian model and the proposed approach. However, if channel spacing decreases, the Gaussian model becomes useless. For the ideal system the 1 dB penalty channel spacing is of only 1.25 GHz. Figure 4.21: Experimental setup. An experimental prototype of the proposed receiver has been assembled into a laboratory setup, shown in figure 4.21. Although the proposed receiver architecture has demon- strated to be highly insensitive to phase noise effects, low linewidth lasers (hundreds of
  • 108. Chapter 4. Advances in phase and polarization diversity architectures 90 kHz) were used. Precisely, total laser linewidth was 300 kHz, much smaller than bit rate. Three external cavity tuneable lasers were used, one on each branch of the setup. First, transmitter branch (TX) was modulated by a standard LiNbO3 phase modulator (PM, Avanex IMP10) at 1Gb/s. Next, interference signal (INT) was obtained by a Mach-Zehnder Modulator (MZM, Fujitsu FTM7921ER) properly biased to work in the 0◦ − 180◦ range. This modulator was driven by the complementary Pseudo-Random Binary Sequence (PRBS). Afterwards, these two branches were coupled and launched in to a 27 km standard G-652 fiber spool. On the other side, the local oscillator branch (LO) was 0◦ − 90◦ modulated by another phase modulator, now driven by the clock (CLK) signal of the generator used to provide data at the TX and INT branches. LO and interfered signals were finally coupled and detected by a balanced detector followed by a 4th-order 2 GHz Bessel-Thomson filter. Under these circumstances, the power penalty due to channel crosstalk was measured. Results are shown on figure 4.20, square points. While 3 GHz spacing was the minimum for a 1 dB penalty, the 3 dB point was found to be between 1.5 GHz and 2 GHz. Also, when channel spacing is greater than 6 GHz, adjacent channel interferences can be almost neglected. Thus, the curve depicted by experimental points is very close to the theoretical one. In summary, a channel crosstalk penalty model have been discussed and experimentally demonstrated, when using a time-switched phase-diversity homodyne receiver. 3 GHz minimum channel spacing was obtained for a 1 dB sensitivity penalty, and 1 Gb/s bit- rate. In other words, more than 1500 wavelength can be easily accommodated in the C band if no other non-linear effects are generated. 4.2.2 Fuzzy data estimation This section aims to change the combination (delay-and-add) block placed after differen- tial demodulation, shown in figure 4.10. This combination block is replaced by a digital signal processing block, containing a data estimation algorithm based on Fuzzy logic. Fuzzy logic techniques have been applied in several fields because of its performances with respect to non-linear systems and its easiness of implementation [50, 51], constituting a reliable approach to hard estimation problems. Thus, novel receiver concept is proposed, based on combining both techniques: time- switched phase diversity and fuzzy logic based digital processing. The first part allows making a homodyne receiver with simple optical and electrical components. This enables a receiver prototype implementation, as this digital processing can be made using off-the- shelf electrical components: digital gates and other simple digital components.
  • 109. Chapter 4. Advances in phase and polarization diversity architectures 91 4.2.2.1 Receiver scheme As previously shown, in the proposed phase diversity homodyne receiver, the coherent photo-receiver mixes the incoming optical field with the local laser carrier in the balanced photo-detector stage. The optical phase modulator at the local laser output is controlled by the data clock producing a periodic 0◦ − 90◦ phase change, to obtain the Inphase (I) and Quadrature (Q) signal components, at the first and second half part of each bit time (Tb ) respectively. After the conventional balanced photodetector, a differential decoder is placed. As the limiting impairment of homodyne systems is phase noise, this detector reduces the com- plexity of the problem to a possible phase cycle-slipping limited by the bounds of a bit time, achievable only when using extremely high linewidths. Figure 4.22: Receiver scheme. Next, an analog to digital converter (ADC) is placed. In order to digitize the I and Q components of the signal received, a minimum of two samples per bit must be taken. Afterwards, digitized signals are processed by the fuzzy-logic-based data estimator. A receiver scheme is shown in figure 4.22. 4.2.2.2 Data estimation Since data is DPSK coded and initial phase is random, after differential decoding, data contaminated with phase noise points fall either at the first or the third quadrant of the IQ signal plane depending on data value (1 or 0). After the photodection stage typical BPSK data only corrupted by a certain amount of phase noise describes a circle in the IQ plane when data is represented as a phasor. When differential decoding is performed the phasor bases changes. The detected I and Q
  • 110. Chapter 4. Advances in phase and polarization diversity architectures 92 component after differential decoding can be written as: VI (t) ∝ cos φd (t) − φd (t − Tb ) + φN (t) − φN (t − Tb ) + (4.26) + cos φd (t) + φd (t − Tb ) + φN (t) + φN (t − Tb ) Tb 3Tb Tb 3Tb VQ (t) ∝ cos φd t − − φd t − + φN t − − φN t − − 2 2 2 2 Tb 3Tb Tb 3Tb − cos φd t − + φd t− + φN t − + φN t − 2 2 2 2 (4.27) where φd (t) is the received coded data phase (either 0 or π ), and N is phase noise term due to the total laser linewidth (∆ν). Figure 4.23: IQ plane data plotting without differential decoding (left), and after differential decoding (right) for a signal corrupted by a phase noise due to 100 kHz of total laser linewidth As the laser phase noise is a Wiener process, after differential demodulation it is trans- formed to a Gaussian random process. Then a data remapping is achieved, between the first and the third quadrant, now clearly separated. This is shown in figure 2 for a small linewidth (100 kHz). For convenience, it is assumed 0 when it falls on the third quadrant and 1 when it falls on the first quadrant. Also I and Q digitized components are assumed to take values from -1 to +1. Thus, a reasonable way to delimit the areas of data value is to classify the universe of discourse of these two components into five membership functions: negative, moderate- negative, null, moderate-positive and positive. These functions are plotted in figure 4.24. In a similar way, the classification for the estimated data must be done in two membership functions: one and zero. As fuzzy membership functions are heuristically determined, their sigmoid shape and the values covered by anyone of them where optimized by means
  • 111. Chapter 4. Advances in phase and polarization diversity architectures 93 Figure 4.24: I and Q components membership functions of numerical evaluation of their robustness in front of phase noise, since this is the major limitation of homodyne systems. I/Q Negative Moderate Null Moderate Positive Negative Positive Negative Zero Zero Zero Zero - Moderate Negative Zero Zero Zero - One Null Zero Zero - One One Moderate Positive Zero - One One One Positive - One One One One Table 4.1: Fuzzy logic estimator rules base. Since membership functions are complex functions, the membership of the digitized signal after the ADC will be determined by means of a look-up table. Based on these member- ship functions, simple rules like ”If [(I is moderate-positive) and (Q is moderate-positive) then (data is one)]” construct a solid fuzzy rules base to estimate detected data. This is shown in table 4.1. It is shown in [51] that for the same membership functions and given a rule base, there are multiple methods to implement the fuzzy estimator. The key point of such methods is to work on the undefined zone of table 4.1, to achieve a higher phase noise tolerance. After extensive numerical simulations, the optimum method is found to be the so called Method of Maximum (MoM), which gives us a maximum linewidth tolerance. Just for summarizing and clarifying, a scheme for data estimation after analog to dig- ital conversion is shown in figure 4.25. First I and Q components are separated (not shown). Next, in the fuzzyfication module, the membership function of each component
  • 112. Chapter 4. Advances in phase and polarization diversity architectures 94 is determined by a look up table (LUT). Finally the decision is performed by the chosen defuzzyfication method (MoM). Figure 4.25: Data estimation scheme. 4.2.2.3 System performances When the optimum fuzzy logic data estimator was obtained, it was used to evaluate the proposed receiver performances by means of numerical simulations. It was compared to the time-switched phase diversity receiver with analog signal processing described before, which provides high phase noise tolerance compared to conventional homodyne systems. In these two cases, the output parameter measured was statistical counting of received bits, which provides the best measure of the system BER. The BER-floor measures were made at several linewidths from 22.5 kHz to 100 kHz. Bit rate was 1 Gb/s. As time- switching diversity introduces a penalty due to its nature and simulations of this kind require high computational resources, minimum BER simulated was 10−4 . The resulting system tolerance to the laser phase noise is depicted in figure 4.26. It shows the BER-floor measured as a function of the laser linewidth, showing that the phase noise tolerance can be extended, from 2.8% bit rate of the best known technique to 3.5% bit rate for a 10−3 BER-floor. This is a 26 % improvement of the bit rate per linewidth product. The use of this type of codes is inevitable mostly due to the penalty inherent to the nature of the time-switching diversity (3 dB penalty due to the half-bit 0◦ − 90◦ modulation).
  • 113. Chapter 4. Advances in phase and polarization diversity architectures 95 Figure 4.26: BER-floor as a function of the laser linewidth at 1 Gb/s 4.2.3 Direct drive time switching This subsection will deal with a third approach to the time diversity switching. Such approach tries to further simplify the optical parts of the receiver by driving directly the laser, and avoiding the phase scrambler module needed in the previous subsections. 4.2.3.1 Receiver scheme The proposed diversity receiver has also two main parts: the first is a coherent photo- receiver with added clock-synchronous sinusoidal phase switching (0◦ − 90◦ ). The sec- ond part is an electronic post-processing performing the signal demodulation and a syn- chronous combination of the orthogonal components, too. Furthermore, a polarization controller is assumed to compensate signal fluctuations due to SOP changes, although a polarization diversity scheme can be implemented with increased receiver complexity. The coherent photo-receiver mixes the incoming optical field with the local laser carrier in the balanced photo-detector stage. The local laser is controlled by the filtered data clock producing a sinusoidal phase change, to obtain the In-phase (I) and Quadrature (Q) signal energy, at the first and second half part of each bit time (Tb ) respectively. The laser is driven by VLO = Ac γ cos(2πfc t); where Ac is the main amplitude, γ is a gain factor, and fc is the fundamental harmonic frequency of the recovered clock. Adjusting the amplitude and the gain factor is critical in order to obtain the desired performances.
  • 114. Chapter 4. Advances in phase and polarization diversity architectures 96 Figure 4.27: Generic receiver module The photocurrent present at the output of the balanced detector can be written as: π Ac γKLO Ipd (t) = 2 PS PLO cos φS (t) − φLO (t) + φd (t) + + sin(2πfc t) 4 2πfc (4.28) where A2πfLO is assumed to be π ; is the photodetector responsivity, PS the transmitter cK c 4 power, PLO the local laser power, KLO is the local laser FM sensitivity, φd (t) is the phase containing the data information, φS (t) is the generic phase at the transmitter side (including phase noise) and φLO (t) is the phase but at the local laser side. This phase modulation can be easily obtained even with a standard DFB laser, with few mA driving amplitude. Regarding the gain factor, as a first approach it can be fixed to 1, in order to achieve a full 0◦ − 90◦ peak-to-peak modulation; but a better performance is obtained when gain √ factor of about 2 is used (Figure 4.28). It is assumed to be the best-fit of the sinusoid energy when comparing to a square wave. After the conventional balanced photodetector, an electrical filter is placed to properly reduce the noise and reject the interference from adjacent WDM channels. Due to the synchronous 0◦ − 90◦ phase scrambling, at this point the bandwidth obtained is two times the low-pass equivalent data power spectrum. As a consequence, this first filter must be twice as broad as usual. This produces 3 dB Signal-to-Noise Ratio penalty with respect to an ideal phase-locked homodyne receiver. However, as seen in the previous subsections, it has much higher linewidth tolerance, about two orders of magnitude. After filtering, a post-processing module is placed in order to make data decision. This is the main core of this receiver and it can be more or less complex depending on the modulation format used. For example, when differentially coded BPSK is detected, the post-processing module is implemented by a delay and-multiply block followed by a delay- and-add block to combine the I and Q infos, as done in the previous subsections. Thus, it reduces the complexity of the phase noise problem to a possible phase cycle-slipping in
  • 115. Chapter 4. Advances in phase and polarization diversity architectures 97 Figure 4.28: BER floor versus the linewidth per bitrate ratio the bounds of a bit time, possible only when using extremely high linewidths. A possible scheme is depicted in figure 4.29, similar to the previously evaluated one. Figure 4.29: Differential BPSK receiver scheme
  • 116. Chapter 4. Advances in phase and polarization diversity architectures 98 4.2.3.2 Phase noise analysis For the phase noise analysis of the DPSK case, lets start the analysis with the current present at the output of the multiplier (Im (t)): 2 π Ac γKLO Im (t) = 4 PS PLO d(t) cos φe (t) + + sin(2πfc t) · (4.29) 4 2πfc π Ac γKLO · cos φe (t − Tb ) + + sin 2πfc (t − Tb ) 4 2πfc = 2 2 PS PLO d(t) cos φe (t) − φe (t − Tb ) + π + cos φe (t) + φe (t − Tb ) + + β sin(2πfc t) 2 where the phase error is expressed as φe (t) = φS (t) − φLO (t). Then, the expression can be rearranged as, ∞ 2 Im (t) = 2 PS PLO d(t) cos θ1 (t) − Jn (β) sin θ2 (t) (4.30) n=−∞ with β = Ac γKLO ; θ1 (t) = φe (t) − φe (t − Tb ) and θ2 (t) = φe (t) + φe (t − Tb ) + 2πnfc t. πfc √ Here, the undesired effects of sin θ2 (t) can be minimize by making γ near 2, meaning that β takes a value that makes J0 (β) very low. As a useful example the positive Bessel coefficients have been plotted for different orders at the desired β value. It is shown in figure 4.30. There, the Bessel coefficients that are above −20 dB respect the unity, are the corresponding to orders 1, 2 and 3; while the other terms can be neglected. Please note that for the odd order coefficients J−n (β) = −Jn (β), being n positive. Oppositely, for the even coefficients J−n (β) = Jn (β), being n positive. As a sum from −∞ to +∞ is obtained, the only surviving terms will be those corresponding to even orders of the Bessel function. Thus, for the assumption made, the only surviving terms are those corresponding to J−2 (β) = J2 (β). Taking into account such approximation, the current after the delay-an-add module, can be written as: Tb Id (t) = Im (t) + Im t − (4.31) 2 2 Tb = 2 PS PLO d(t) cos θ1 (t) + cos θ1 t − − 2 Tb −2J2 (β) sin θ2 (t) − 2J2 (β) sin θ2 t− 2
  • 117. Chapter 4. Advances in phase and polarization diversity architectures 99 √ Figure 4.30: Bessel coefficients for γ = 2 In a similar way as equation 4.8, Id (t) can be rewritten as: Id (t) = Io (t) + Io (t − Tb /2) (4.32) 2 = 2 PS PLO d(t) cos(∆φ1 + ∆φ2 ) + cos(∆φ2 + ∆φ3 ) + 3∆φ1 + 2∆φ2 + ∆φ3 ∆φ1 + ∆φ2 +2J2 (β) sin 2φe (t) ± 2π2fc t − cos 2 2 where Tb ∆φ1 (t) = φe (t) − φe t − 2 Tb ∆φ2 (t) = φe t − − φe (t − Tb ) 2 3Tb ∆φ3 (t) = φe (t − Tb ) − φe t − 2 as when talking about the first time-switching case, shown in figure 4.10, and expressed in the same equation 4.8. Here they are also independent and identically distributed gaussian random processes, with zero mean and a variance proportional to the linewidth per half of bit-time product. In principle, when phase noise can be neglected, performing decision on Im (t), is slightly better than deciding on Id (t). This is shown in figure 4.31. It is due to the fact that low linewidth can be cancelled by the differential detection itself and the interference of the Bessel function terms. On the other hand, when phase noise becomes important, the cos(∆φ2 + ∆φ3 ) and cos(∆φ1 + ∆φ2 ) terms grow, and are the main phase noise
  • 118. Chapter 4. Advances in phase and polarization diversity architectures 100 contribution. To see it more clearly, three aspects should be highlighted regarding the Bessel coefficients terms: 1. The ideal receiver bandwidth of such a diversity receiver is 1.5 times data rate. 2. The phase noise will vary more slowly than data rate (i.e. its bandwidth is much narrower than data signal bandwidth). 3. Noise terms contributed by the Bessel function are over a carrier running at a frequency two times the data rate. Figure 4.31: Comparison between decision on Id (t) (using delay-and-add, DAD) and Im (t) (NDAD). Thus, it can be assumed that these terms are going to be filtered by the matched filter placed before decision block, and will not interfere in data detection. Thus, Id (t) can be approximated by: 2 Id (t) ≈ PS PLO d(t) cos(∆φ1 + ∆φ2 ) + cos(∆φ2 + ∆φ3 ) (4.33)
  • 119. Chapter 4. Advances in phase and polarization diversity architectures 101 4.2.3.3 Frequency drift analysis Now, the goal is to have a look on what happens when a frequency drift 2πfd is applied. In this case, Im (t) can be rewritten as: 2 π Ac γKLO Im (t) = 4 PS PLO d(t) cos φe (t) + 2πfd t + + sin(2πfc t) · 4 2πfc π Ac γKLO · cos φe (t − Tb ) + 2πfd (t − Tb ) + + sin(2πfc (t − Tb )) 4 2πfc = 2 2 PS PLO d(t) cos φe (t) − φe (t − Tb ) + 2πfd Tb + π + cos φe (t) + φe (t − Tb ) + 4πfd t − 2πfd Tb + + β sin(2πfc t) 2 ∞ 2 = 2 PS PLO d(t) cos θ1 (t) − Jn (β) sin θ2 (t) (4.34) n=−∞ being β = Ac γKLO ; θ1 (t) = φe (t) − φe (t − Tb ) + 2πfd Tb and θ2 (t) = φe (t) + φe (t − Tb ) + πfc 2π(2fd + nfc )t − 2πfd Tb . For the correct development of the model, no phase noise is assumed. Then: 2 π Im (t) = 2 PS PLO d(t) cos 2πfd Tb ) + cos 4πfd t − 2πfd Tb + + β sin(2πfc t) 2 ∞ 2 = 2 PS PLO d(t) cos 2πfd Tb − Jn (β) sin 2π(2fd + nfc )t − 2πfd Tb n=−∞ (4.35) Next, the output of the delay-and-add module can be calculated, obtaining: Id (t) = Im (t) + Im (t − Tb /2) (4.36) ∞ 2 = 2 PS PLO d(t) cos(2πfd Tb ) − Jn (β) sin 2π(2fd + nfc )t − 2πfd Tb − n=−∞ ∞ − Jn (β) sin 2π(2fd + nfc )t − 2π(6fd + nfc )Tb /2 n=−∞ 2 = 2 PS PLO d(t) cos(2πfd Tb ) − ∞ 2π(4fd + 2nfc )t − 2π(10fd + nfc ) Tb 2 −2π(2fd + nfc )Tb −2 Jn (β) sin( ) cos( ) n=−∞ 2 4 Revisiting the assumptions made in the last sub-subsection, here the main cotributing Bessel term is also the corresponding to order 2, and will produce a carrier running at twice the data rate plus the correspondent tones due to the frequency drift. As such a frequency drift can be also assumed to be much lower than data rate, Id (t) can be
  • 120. Chapter 4. Advances in phase and polarization diversity architectures 102 approximated as: 2 Id (t) ≈ PS PLO d(t) cos(2πfd Tb ) (4.37) That is the same expression than the one reported in equation 4.11, and the same effects on the overall BER are expected, shown in figure 4.14. 4.2.3.4 Simulations In order to evaluate the performances of the proposed receiver architecture, a BPSK configuration was used (Figure 4.29). It was compared to a classical time-switching phase diversity receiver by means of numerical simulation with Monte-Carlo BER estimation. First of all, the linewidth per bit rate tolerance was evaluated at 1 Gb/s. The output parameter measured was error counting of received bits. The BER measures were made at several linewidths from 10 MHz to 100 MHz and assuming an infinite signal to noise ratio (SNR) in order to properly evaluate the phase noise effects. Due to high computational resources required by simulations of this kind, minimum BER simulated was 10−4 . The resulting system tolerance to the laser phase noise is depicted in figure 4.28. When a sinusoidal phase switching is used with a 0◦ −90◦ peak-to-peak phase swing, the maximum tolerated linewidth at 10−3 BER is shown to be as low as 2% bit rate, whereas using square phase switching it is 2.8% bit rate. This is mainly due to the penalty when moving from square waveform to a sinusoidal wave. Note that FEC codes must be used to increase this 10−3 BER up to 10−9 . When the amplitude of the sinusoidal wave is increased to its optimum, the spectral broadening is near-optimum, increasing the maximum tolerated linewidth up to 3.5% bit rate. It is slightly higher than the 3.2% reported in subsubsection 4.2.1.2, but now using a much simpler scheme. Next, the effects of changing the gain factor were evaluated. The optimum value was √ found to be 2, as expected. This is shown in figure 4.32. Afterwards, the sensitivity penalty was obtained when using this new configuration con- sidering the additive noise. Here the laser linewidth was disabled. Hence, the probability density function of the output bits was found to be the same as when having a pure DPSK, but with a certain penalty. Results are shown in figure 4.33. As expected, a certain penalty is observed when replacing the square signal by the sinusoid: 0.5 dB √ when γ = 2 , and 0.7 dB when 1 for a 10−9 BER. Almost no penalty is observed when comparing at 10−3 BER (−40 dBm). 4.2.3.5 Experiments An experimental prototype of the proposed receiver has been developed and was tested in a laboratory setup. It is shown in figure 4.34. Sensitivity penalty for gamma parameter (gain factor) and frequency drift impacts were evaluated. Although the proposed receiver
  • 121. Chapter 4. Advances in phase and polarization diversity architectures 103 Figure 4.32: Maximum tolerated linewidth per bit rate ratio at BER 10−3 as a function of the gain factor. Figure 4.33: Receiver sensitivity for several configurations. architecture was demonstrated to be highly insensitive to phase noise effects, an external cavity tunable laser was used featuring low linewidth. The total linewidth was 300 kHz (2x150 kHz), much smaller than the bit rate (1 Gb/s). The TX branch was binary phase shift keying modulated with a Mach-Zehnder Modu- lator (MZM, Fujitsu FTM7921ER) at 1 Gb/s. The local oscillator branch was 0◦ -127◦ modulated with a standard LiNbO3 phase modulator (PM, Avanex IMP10), driven by the clock signal (CLK) of the pseudo-random binary sequence (PRBS) generator used to provide data at the first branch. These two optical signals are coupled and detected by a
  • 122. Chapter 4. Advances in phase and polarization diversity architectures 104 balanced detector followed by an electrical processing stage (a delay-and-multiply mod- ule plus a delay-and-add stage). A microwave double-balanced mixer (Marki Microwave M20004) was used as a four-quadrant multiplier. Figure 4.34: Experimental setup for the direct drive time-switching. Some preliminar experiments were carried out, measuring the SNR penalty at BER 10−3 . First, for the γ factor (from γ = 1 to γ = 2), and afterwards for a laser fluctuation of up to ±200 MHz. The gain factor results are shown in figure 4.35. Where it is shown that the optimum is √ achieved at γ = 2. Also the 1 dB penalty points were mesured to be at γ = 1.3 (117◦ ) and γ = 1.8 (162◦ ), meaning 0.5 (45◦ ) tolerance. Regarding frequency drift tolerance results are shown in figure 4.36. There it is shown that 1 dB penalty was achieved for 75 MHz detuning. Meaning that the system is very sensitive to the laser wavelength fluctuations. 4.2.4 Searching for a polarization diversity In this case, the diversity receiver has two main parts: the first is a coherent photo-receiver with added clock-synchronous phase (0◦ − 90◦ ) and polarization (H-V) scrambling at the local laser output. The second part is an electrical post-processing performing the signal demodulation and the synchronous combination of the orthogonal components. The local laser does not need to be coherent with the incoming optical carrier, although an automatic frequency controller is convenient to maintain the two wavelengths close. It can be regarded as a heterodyne receiver with near-zero intermediate frequency.
  • 123. Chapter 4. Advances in phase and polarization diversity architectures 105 Figure 4.35: SNR factor penalty at 10−3 BER vs gain factor γ. Figure 4.36: SNR factor penalty at 10−3 BER vs frequency drift. The coherent photo-receiver mixes the incoming optical field with the local laser carrier in the balanced photo-detector stage. These components are no coincident with the Tx-generated ones due to the unlocked transmission phase, but, with post-processing performing the operations in [24], the phase-modulated information is fully recovered. The optical phase scrambler at the local laser output is controlled by the data clock (50% duty cycle) producing a 0◦ − 90◦ phase modulation, to obtain the I and Q signal components, at the first and second half part of each bit time (Tb ) respectively, after the optical mixing. In a similar way, the polarization components V and H, are at even and odd quarters of the bit time, respectively. This is indicated in figure 4.37.
  • 124. Chapter 4. Advances in phase and polarization diversity architectures 106 Figure 4.37: I, Q, H, V time distribution of each bit Correspondingly, the photo-receiver output is gated with the data clock signal, its in- verse, and a doubled frequency version of them to obtain the I and the Q separately for every polarization component in the four branches, now with RZ shape. In order to re-synchronize the data clock signal with the signals introduced to the post-processing block and phase modulator, the respective relative delays are introduced in the branches, and a variable delay is added to compensate for the RX propagation delays. The electronic signal processing stage performs the differential demodulation of the four components, separated by electronic switching, with a delay time equal to Tb . All demod- ulated components are synchronously combined with an adder. Due to the phase noise and the SOP random variation, the signal power fluctuates between the four branches randomly, at a rate in the order of the laser linewidth and SOP fluctuation, and the detailed combination of all the outputs assures its recovery (Figure 4.38). In terms of phase noise, compared to an oPLL, the phase-swing time has been shortened from the loop delay to only a bit time, which, in contrast, reduces while bit rate increases. Figure 4.38: Intradyne differential receiver with polarization and phase diversity.
  • 125. Chapter 4. Advances in phase and polarization diversity architectures 107 To achieve the best performances in polarization diversity terms, the scrambler must in- troduce a clock-synchronous 90◦ rotation. In is implemented with a highly birefringent phase modulator, with its input 45◦ linearly polarized. The results obtained after com- bining all the branches are the same theoretical results obtained in [24, 52]. However, it must be taken into account that the reduced duty-cycle and the correspondingly increased electrical bandwidth produces 6dB SNR penalty. The analysis of such receiver is complex, and it would be good to take into account the phase noise and polarization effects separately. First, lets deal with the polarization. In fact, polarization varies very slowly (in the order of ms) when compared to data rates. So, the parameters regarding polarization can be assumed to be constant and the equations governing the receiver can be set as: eS (t) = PS (cos(ϕ)ˆ + sin(ϕ)e−jθ y ) exp jω0 t + φS (t) x ˆ (4.38) eLO (t) = PLO (ˆ + y ) exp jω0 t + φLO (t) x ˆ (4.39) φe (t) = φS (t) − φLO (t) (4.40) As now the point under study is the polarization the phase noise term φe (t) can be neglected, and the currents after photodetection can be described as: IHI (t) = Cd (t) cos(ϕ) (4.41) IV I (t) = Cd (t) sin(ϕ) cos θ (4.42) IHQ (t) = Cd (t) cos(ϕ) (4.43) IV Q (t) = Cd (t) sin(ϕ) sin θ (4.44) Thus, after data demodulation and delay alignment of the components: C2 IHIo (t) = d(t) cos2 (ϕ) (4.45) 2 C2 IV Io (t) = d(t) sin2 (ϕ) 1 + cos(2θ) (4.46) 2 C2 IHQo (t) = d(t) cos2 (ϕ) (4.47) 2 C2 IV Qo (t) = d(t) sin2 (ϕ) 1 − cos(2θ) (4.48) 2 And after combination it leads to: C2 Io (t) = IHIo (t) + IV Io (t) + IHQo (t) + IV Qo (t) = d(t) (4.49) 2 having no polarization influence on data decision. Regarding phase diversity, it should be noted that time distribution of the I and Q components are the same as the first time-switching diversity receiver analyzed. So it can be better implemented with the simplified scheme presented in figure 4.39. Please note
  • 126. Chapter 4. Advances in phase and polarization diversity architectures 108 Figure 4.39: Alternative implementation for achieving time-switched phase and po- larization diversities. that in such receiver the signals that are added at each Tb /4 are the ones corresponding to polarization diversity, while the ones corresponding to phase diversity are combined each Tb /2. Then, the assumptions made in sub-subsection 4.2.1.2 are still valid for this receiver, as well as the results of equation 4.8. 4.3 Chapter summary In this chapter four novel techniques were introduced for achieving high phase noise and SOP tolerances: a phase estimation algorithm based on Karhunen-Lo`ve series expan- e sion for standard full phase diversity receivers, a simple time-switching phase diversity architecture, a Fuzzy logic data estimation algorithm, and a time-switching polarization diversity architecture. For the phase estimation algorithm based on Karhunen-Lo`ve series expansion, the con- e cept was developed theoretically and performed some simulations. Due to its spectral properties, it remarkably increases the phase noise tolerance of conventional coherent ho- modyne receivers, up to a linewidth of 4 % bit rate for the 1 dB penalty point at a BER of 10−9 , and it also avoids the need for oPLL. Next, the time-switching phase diversity architecture has been discussed. Here the objec- tive was to obtain a phase diverse coherent receiver, using standard off-the-shelf optical components. This is achieved by introducing a phase dithering that covers the entire bit time with a [0◦ , 90◦ ] switch. Two variations have been presented: using a phase scrambler/switch, and directly driving the local laser. With them the obtained phase noise tolerance was of 1.8 % and 5.4 % of the bitrate, respectively (BER-floor 10−3 ). Also channel spacing penalty has been analyzed and measured to be of 1 dB for 3 GHz spacing at 1 Gb/s. Nevertheless, the fact of switching between 0◦ and 90◦ in the same bit
  • 127. Chapter 4. Advances in phase and polarization diversity architectures 109 broadens the signal bandwidth and introduces a non-negligible penalty of 3 dB respect to a theoretical coherent receiver. In between, the concept of combining Fuzzy logic data estimation and time-switching was proposed and analyzed. It is an interesting approach that also combines low-complexity field programmable arrays, avoiding the need for a digital signal processor that would be needed for the first proposed phase estimation algorithm. In this case, the linewidth tolerance was of 3.5% bit rate for a 10−3 BER-floor, equivalent to a 10−9 when FEC codes are used. Finally, a time-switching polarization diversity receiver architecture has been presented. In this case the concept has been theoretically demonstrated, achieving a total SOP insensitivity. However, it must be taken into account that the reduced duty-cycle and the correspondingly increased electrical bandwidth produces a 3dB SNR penalty, like in the time-switching phase diversity architecture.
  • 128. Chapter 5 ONU and OLT architectures 5.1 Summary of techniques and issues to take into account Homodyne systems today are mainly investigated toward long-haul WDM applications and not being seriously considered for use in access passive optical networks (PONs). As these networks have multiple low capacity channels, a major concern is the use of optical filters in order to delimitate these channels in direct-detection based systems, mainly because of the commercial filter’s low selectivity. Thus, if demand increases, a coherent receiver using electrical filtering is an alternative way to solve this problem. Heterodyne optical receivers can be a first approach, as in [5], but due to its inherent image frequency problem, a better solution is homodyne reception. Nevertheless, the implementation of homodyne transceivers has not been commercially deployed in part because of its stringent requirements in terms of laser spectral width (laser’s phase noise) and polarization mismatch. Also, other issues more related to generic transceiver’s bidirectional transmission over a single fibre are relevant: Rayleigh backscat- tering and modulation formats to be used. 5.1.1 Phase noise In order to properly synchronize local laser and received signals, early homodyne coher- ent systems used an optical Phase-Locked Loop (oPLL) module. Several architectures were proposed and analyzed: Decision driven [19], Costas [18], Balanced [21], Subcarrier modulated [23], and heterodyne loops [33]. However, they had the same problem: optical path between local laser and optical mixer (e.g. optical hybrid + photo-detection stages) introduces a non-negligible loop delay [17], resulting in a significant penalty. In order to avoid it, extremely low linewidth external cavity lasers had to be used. 111
  • 129. Chapter 5. ONU and OLT architectures 112 In chapter 2, it has been shown that another approach towards homodyne reception came later, with the concept of zero-IF/intradyne diversity receivers. The main goal of these receivers is to replace the feedback loop (oPLL) by a feed-forward post-processing. For full phase diversity schemes (using a 90◦ Hybrid), the architecture is based on an intradyne system, using a free-running local oscillator which is nominally at the same fre- quency as the incoming signal. Several relevant post-processing approaches, depending on the estimation/decoding performed are: differential detection in the analog domain [24], digital Wiener filtering [26], regenerative frequency dividers [28], Viterbi-Viterbi algo- rithm [30, 31], and Fuzzy logic data estimation [45]. The core component is the 90◦ hybrid, which can be integrated as a polymer waveguide device [53], offering a fairly sim- ple and low cost fabrication involving low-temperature processes and low cost packaging based on passive alignment [54]. A third approach is time-switching phase diversity, where the receiver has two main parts: the first is an intradyne coherent receiver, with an added phase modulator at the local laser output; the second is the electrical post-processing. The optical phase modulator at the local laser output is controlled by the data clock (50% duty cycle) producing a fixed 0◦ − 90◦ phase modulation, to obtain the I and Q signal components, at the first and second half part of each bit time respectively, after the optical homodynation [41, 45]. As seen in section 4.2, a further simplification consists in driving directly the local laser with a sinusoidal wave at the bit frequency [46]. So, if the laser is working in saturation mode, one can take advantage of its adiabatic chirp and perform a sinusoidal frequency modula- tion resulting in a phase modulation 0◦ -127.3◦ , obtaining a further tolerance enhancement of the time-switching diversity performances. Technique Linewidth Penalty Required key Complexity tolerance component Decision-drive loop 5 MHz 0 dB 90◦ hybrid High Costas loop 4.9 MHz 0 dB 90◦ hybrid Medium/High Subcarrier loop 5.1 MHz 0 dB 90◦ hybrid High Balanced loop 2.4 MHz 2 dB Optical coupler Low Heterodyne loop 6.4 MHz 1 dB Optical coupler Low Full phase diversity 5% bitrate 0 dB 90◦ hybrid Medium Time-switch (Scrambler) 1.8% bitrate 4 dB Phase modulator Medium Time-switch (Direct drive) 5.4% bitrate 4 dB High-chirp laser Low Table 5.1: Phase noise cancellation techniques summary table. The linewidth toler- ance is for a 10−3 BER-floor, whereas the penalty is respect to an ideal system. 5.1.2 Polarization mismatch Polarization is one of the major problems for the coherent systems, as the optical mixing of the LO and received signals needs to be aligned in polarization. There are four different
  • 130. Chapter 5. ONU and OLT architectures 113 approaches to deal with a possible polarization mismatch: full polarization diversity, time- switching polarization diversity, polarization scrambling and local polarization control. A full polarization diversity scheme, using a polarization beam splitter (PBS) and double receiver, has been already investigated [24, 26, 55]. However, it means doubling the number of optical components and precise match. A decisive enhancement would be to integrate it with the 90◦ hybrid in polymer technology, as for reduced fabrication costs would be possible. In time-switching polarization diversity, each bit time would be split into two, one con- taining H polarization information and the other one containing V polarization infor- mation [47, 56]. This diversity is achieved by driving the clock signal to a polarization switch/scrambler. Afterwards, at the electronic post-processing block, H and V compo- nents can be properly combined. A drawback can be the loss of 3 dB of sensitivity, but it can be afforded due to the inherent high sensitivity of the receiver. For the polarization scrambler, it can be implemented by a highly birrefringent phase modulator, with its input light beam polarization properly aligned. It can be done either at the customer premises, or better at the central office to share the cost of the device, that would not have to be integrated. Additional options can be investigated [57, 58]. Local polarization control requires a polarization actuator (Liquid crystal, Fibre squeezer, Faraday rotator, to mention a few) at the customer’s equipment, with a related high cost. It also means an additional electronic control (at low frequency) should be placed at the electric part of the receiver. Local control Polarization diversity Polarization switching Penalty 0 dB 0 dB 3 dB Key component Polarization actuator Pol. beam splitter Pol. scrambler/switch Response time 1 ms – 1 s < 10 µs < 10 µs Complexity High Med./high Low (if placed at CO) Table 5.2: Polarization handling methods summary table [56]. 5.1.3 Modulation techniques and Rayleigh backscattering Even though it is not an objective of the present thesis, one of the key issues in PONs is the full bidirectionality of a transmission system over a single fiber. If down-stream and up-stream spectra overlap, the Rayleigh backscattering may become a substantial interference to the received signal, for fiber lengths of more than 10 km [59]. A double fiber scenario will be asumed for the architectures under discussion, while keeping in mind a couple of approaches to solve this in future lines: dual laser configuration (one used as a reference for detection, and the other for upstream transmission), and inclusion of a wavelength shifter at the ONU, using a wavelength shifting device [60]. For the
  • 131. Chapter 5. ONU and OLT architectures 114 targeted low cost in access, to achieve phase modulation with laser direct modulation, the technique in [61] can be adopted. 5.2 ONU and transceiver architectures In this section, some coherent ONU schemes are presented. Concerning the modulation data rate, will be 1 Gb/s to directly transmit the common EPON protocol over fiber. In general terms, the downstream modulation format to be used is PSK, because of its good trade-off between performances and simplicity. For upstream modulation, PSK is also preferred, but the simpler IM option can be also used with an asymmetrical up/down data rate, in order not to penalize optical power budget. In principle it is assumed that a modulator can be placed at each transceiver and leave other implementations as open issues for future lines. Precisely, recently it has been demonstrated a coherent receiver integration with an optical phase modulator [62]. Regarding electronics, digital and analog signal processing will be the adopted solutions against the impairments. A key element is the broadband 4-quadrant multiplier required for the differential demodulation of I and Q components. 5.2.1 Transceivers based in a full phase diversity scheme 5.2.1.1 Transceiver with 90 degree hybrid and digital processing The basic implementation uses 90◦ hybrid with 2 pairs of balanced photo-detectors. The electronic post-processing combines both I and Q detected components optimally. Scheme is shown in figure 5.1. The targeted modulation format is BPSK, although other multilevel Figure 5.1: Transceiver with 90◦ hybrid and digital processing. modulations can be used and investigated, e.g. QPSK.
  • 132. Chapter 5. ONU and OLT architectures 115 Channel selection is performed by tuning the local laser to the right wavelength and fil- tered by the electrical low pass filters. In the digital implementation, data is sampled and converted into digital domain by high-speed analog-to-digital converters (ADCs). Inside digital I and Q post-processing several basic operations are performed: Phase es- timation, frequency estimation and control, data estimation, and polarization switching combination. Phase and data estimations can be performed taking as a basis well-known techniques, like Wiener filter [26] or Viterbi-Viterbi [31], which have proved their fea- sible implementation. Also other new algorithms can be used, as the one proposed in section 4.1.1. Regarding frequency estimation, standard digital controls can be used, like the ones used in RF communications [6] or the most advanced recently demonstrated for optical communications [63, 64]. Polarization is managed at the OLT side, performing a switching that covers the two H and V orthogonal states in the same bit time period. 5.2.1.2 Transceiver with 90 degree hybrid and analog processing This is the analog version of the previous transceiver, with lower consumption require- ments, data is differentially encoded and demodulated (DPSK), for higher tolerance against phase noise. Its scheme is shown in figure 5.2. The post-processing is now composed by two parallel delay-and-multiply blocks, for differential detection. The com- bination of the I and Q component output by a standard electrical combiner may require optimal weightening (not shown in the figure) for optimal performance. Next, a delay and add module may perform the proper polarization switching combination, also not shown in the figure. Finally, before data decision, a 4th order Bessel low pass filter is placed. Figure 5.2: Transceiver with 90◦ hybrid and analog processing. 5.2.1.3 Transceiver including 90 degree hybrid and PBS, with digital pro- cessing This is the natural extension of architecture 5.2.1.1, using polarization beam splitters for achieving polarization diversity. The transceiver architecture is shown in figure 5.3. With
  • 133. Chapter 5. ONU and OLT architectures 116 them, a full polarization diversity is achieved and the sensitivity penalty can be neglected. However, the number of optical and electronic components is doubled, increasing the complexity and cost of the transceiver. Figure 5.3: Digital configuration scheme using 90◦ hybrid combined with PBS. 5.2.1.4 Transceiver including 90 degree hybrid and PBS, with analog pro- cessing Similarly to the previous one, this is the natural extension of architecture 5.2.1.2, using polarization beam splitters for achieving polarization diversity. Also, the only differences are the duplicity of optical and electrical components. Now the post-processing is com- posed by four parallel delay-and-multiply blocks, for differential detection, as proposed in [24]. Figure 5.4: Analog configuration scheme using 90◦ hybrid combined with PBS.
  • 134. Chapter 5. ONU and OLT architectures 117 5.2.2 Transceivers based in time-switching phase diversity 5.2.2.1 Transceiver including phase switch with digital processing and stan- dard balanced detector As an alternative to 5.2.1.1, 5.2.1.2, 5.2.1.3 and 5.2.1.4 it is proposed to implement a time-switching scheme for data detection, described in section 4.2 and reported in [41]. Here, by means of fast phase scrambling, the 90◦ hybrid can be avoided and a simple 3dB coupler can be used. For the digital version, the targeted modulation format is also PSK. Polarization is managed at CO, performing an H and V alternate switching, too. A remarkable difference is the data rate clock recovery unit, placed before the A/D conversion block; its output drives a phase modulator/scrambler, performing the 0◦ − 90◦ phase switching needed. Also, a similar structure has been integrated in an InP substrate [62]. Figure 5.5: Digital configuration scheme using phase switch. 5.2.2.2 Transceiver including phase switch with analog processing and stan- dard balanced detector For the analog version of 5.2.2.1, the targeted modulation format is DPSK. Then, after channel selection filter, a couple of switches perform the I and Q components distribution, and a delay and multiply block is placed at each branch, for differential detection. 5.2.2.3 Transceiver including direct laser switching with digital processing and standard balanced detector Here, by means of fast phase scrambling of the local laser, the phase scrambler can be avoided simplifying even more the transceiver architecture. For the digital version, the targeted modulation format is PSK, too. Polarization is also managed at CO. A slight difference with architecture 5.2.2.1 is that the data rate clock recovery output is filtered
  • 135. Chapter 5. ONU and OLT architectures 118 Figure 5.6: Analog configuration scheme using phase switch. for obtaining a sinusoidal signal, which drives directly the laser, performing the 0-127 phase modulation needed [46]. Figure 5.7: Digital configuration scheme using standard balanced detector. 5.2.2.4 Transceiver including direct laser switching with analog processing and standard balanced detector For the analog version of 5.2.2.3, the targeted modulation format is DPSK. So, a delay and multiply block is placed at each branch, for differential detection. Next, a delay-and-add block (T /2 delay) can be placed (not shown), for combining the orthogonal polarization components of the received signal. Finally, before data decision, a 4th order Bessel low pass filter is placed.
  • 136. Chapter 5. ONU and OLT architectures 119 Figure 5.8: Analog configuration scheme using standard balanced detector. 5.2.3 Transceiver based in Optical Phase-Locked Loop 5.2.3.1 Transceiver with OPLL and analog processing This architecture has a more advanced electronics implementing homodyne detection with optical coupler. In this case, the targeted modulation format is DPSK, and polarization is managed by switching at OLT side. It is a homodyne balanced receiver with fast lock- in dithering and proportional-integral optical tuning loop. This leads to an amplitude modulated error phase ranging signal after balanced receiver. Next, a squaring module (or an electrical double rectifier) removes data modulation, and its output is filtered and down-converted in a proper way [33]. At data detection branch, a delay-and-multiply block performs differential demodulation, and next a delay-and-add block can be placed (not shown), for combining the orthogonal polarization components of the received signal. Figure 5.9: Analogue configuration scheme for the oPLL transceiver prototype.
  • 137. Chapter 5. ONU and OLT architectures 120 5.2.4 Transceiver comparison Table 5.3 summarizes the requirements of the different architectures presented, in terms of optical phase handling, polarization handling, electronic processing type, sensitivity penalty with respect an ideal homodyne system and the laser linewidth tolerance at 1 Gb/s (these values are approximated, since depend on the precise implementation). Arch. Phase Polarization Electronic Sens. Linewidth Cost handling handling processing penalty tolerance 5.2.1.1 90◦ hybrid Switch at CO Digital 3 dB 5 MHz Med./High 5.2.1.2 90◦ hybrid Switch at CO Analog 4 dB 5 MHz Med./High 5.2.1.3 90◦ hybrid PBS Digital 0 dB 5 MHz Very high 5.2.1.4 90◦ hybrid PBS Analog 1 dB 5 MHz Very high 5.2.2.1 Switch (Scr.) Switch at CO Digital 6 dB 1.8 MHz Medium 5.2.2.2 Switch (Scr.) Switch at CO Analog 7 dB 1.8 MHz Medium 5.2.2.3 Switch (Dir.) Switch at CO Digital 6 dB 5.4 MHz Low 5.2.2.4 Switch (Dir.) Switch at CO Analog 7 dB 5.4 MHz Low 5.2.3.1 OPLL Switch at CO Analog 4 dB 525 kHz Low Table 5.3: Transceiver architectures summary table. The linewidth tolerance is for 1 dB penalty at 10−10 BER, whereas the penalty is respect to an ideal system. Among them, it should be remarked that the architecture presenting higher performances is the one combining 90◦ hybrids, a PBS and digital signal processing ( 5.2.1.3). It presents no additional penalty with respect to an ideal system, while achieving high linewidth tolerance. But it is costly because it implies the duplication of many components needed. Another architecture that should be highlighted is the one that uses a direct-drive time- switching in combination with time-switching and digital signal processing (5.2.2.3). Even though it presents a high penalty (6 dB), it has a high linewidth tolerance and a reduced complexity and, thus, cost. Nevertheless, it requires a fully engineered laser, capable to be phase modulated. In principle a standard DFB laser can be used, but it would have an undesired residual intensity change. Finally, OPLL approach (5.2.3.1) should not be forgotten, as it is an architecture featuring low complexity, but with lower linewidth tolerance (525 kHz). An additional handicap, is the delay associated to the optical path length. As this delay dramatically limits the OPLL performances, the local laser should be ideally embedded with the optical reception front-end (couplers and photodetector). 5.3 OLT architecture Regarding the OLT implementation it will be strongly linked with the network archi- tecture to be deployed. For example, the case of having only a power splitting stage at
  • 138. Chapter 5. ONU and OLT architectures 121 the external plant is different than when a wavelength routing device is present at the distribution network. Nevertheless, there are several parts that will be common to all the networks, and they are going to be discussed in this section. Here our purpose is to give an overview of the problems that can arise when designing the OLT, not giving a detailed analysis. First of all the transceivers used at the OLT should be commented. For the proposed double fiber network approach the same architectures suggested for the customer premises equipment could be used. Figure 5.10: OLT scheme with double fiber and including the birefringent polarization switch. Apart from the transceivers, an additional polarization actuator should be placed in order to perform the high speed polarization switch needed for architectures 5.2.1.1, 5.2.1.2, 5.2.2.1, 5.2.2.2, 5.2.2.3, 5.2.2.4, and 5.2.3.1. A first approach can be the use of a birefringent component to perform this switch [56]. The problem with these devices is that for performing an aproppiate switching they need at their input a SOP corresponding to equal powers in the two principal modes of the birefringent component. Usually, 45◦ linearly polarized light is used. Thus, in order to deliver the needed SOP, the transceivers should be carefully designed, combining optical waveguides and polarization mantaining fibers. So, it should be not feasible to place a single polarization switch stage after combining all the OLT transceivers. The most viable but less affordable approach would be to place the birefringent modulator at the output of each transceiver. Also, as the SOP of the ONU’s signals is completely random, a polarization controller should be used for each one. A scheme of the OLT with a polarization switch based in birefringent modules is shown in figure 5.10, where an external plant having only a power splitting stage is assumed. There it is shown that the transceiver optical output is connected to the birefringent high-speed modulator, which is driven with a high-speed signal (e.g. the data clock). There the retardation is performed and the two orthogonal polarizations are obtained in the same bit slot. For example if a 45◦ linearly polarized beam of light is present at the input, a ±45◦ linear polarization alternate will be present at the output. For upstream signals, only a polarization control is needed, with a polarization
  • 139. Chapter 5. ONU and OLT architectures 122 actuator (Liquid crystal, Fiber squeezer, Faraday rotator,. . . ) driven by an electrical signal provided by the reception part of the transceiver. Of course, this signal type will depend on the modulation format used and the transceiver architecture. Figure 5.11: OLT scheme with double fiber and including the FRM based polarization switch. A more cost effective alternative to the birrefringent based polarization switch can be the use of a single switch for all the transceivers of the OLT, shown in figure 5.11. Here it is also assumed to have an optical network having only a power splitting distribution stage. In that scheme, first an intensity modulator shapes the input signals, giving at its output pulses with half of the bit time length. Next, an optical isolator is placed for ensuring unidirectionality. After the isolator, these pulses enter a planar lightwave circuit (PLC) composed of 3 optical couplers, a controlled length delay, and a Faraday Rotator Mirror (FRM). At the output of this structure a secondary pulse is obtained, with 90◦ polarization rotation respect to first, plus the pass-through result of the introduced pulse. As the output power for the first half of a bit would be higher than for the second half, a gain controlled EDFA [65] should be placed after the PLC, for maintaining a fixed output power during all the bit time. These two alternatives are so complex and have to be further investigated. The first one is based in an architecture that has been demonstrated a long time ago [56], whereas the second one is more background breaking and has to be carefully examined before practical implementation. In fact, one should pay special attention to the the delay difference between the two possible paths of the optical couplers network present in figure 5.11.
  • 140. Chapter 5. ONU and OLT architectures 123 Nevertheless it was proposed here for future lines of research, as it can be an interesting way to follow. 5.4 Chapter summary As the thesis is focused on access networks, in this chapter a discussion has been performed for the possible transceiver architectures for OLT and ONU. After discussing the main problems and summarizing the results obtained after the pre- vious work of the thesis, 9 new transceivers were presented for being used at both sides, OLT and ONU. They are all based in a double fiber network approach, as a first step for avoiding Rayleigh backscattering, leaving it for further investigations. Furthermore, since recently a coherent receiver integration with an optical phase modulator has been demon- strated [62], a phase modulator has been assumed to be integrated at each transceiver leaving other implementations as open issues for future research, too. As a first conclusion, the ONU transceiver architecture that has demonstrated to be the most appropriate for the first network tests which are going be performed in this thesis, is the one that uses a direct-drive time-switching in combination with time-switching and analog signal processing. Even though it presents a high penalty (6 dB), it has a high linewidth tolerance and a reduced optical complexity. Thus, potential cost is very low. Also, the sensitivity demonstrated (−38.7 dBm) is enough for achieving the high power budget required for PONs. Regarding the particular designs of the OLT, the implementation of the polarization switching has been discussed. A first approach including birefringent phase modulators has been analyzed and found to be not optimum due to the amount of components needed. Thus, another architecture, based on Faraday rotating mirrors has, been proposed. As it uses few optical components and its cost can be shared among all network users, it is a more affordable way to achieve the desired polarization switch.
  • 141. Chapter 6 Network topologies Usually, the architecture of a PON has some flexibility depending on the balance between distance reach, number of users served and its geographical distribution. In our ultra dense (UD) WDM approach, the passive losses can be balanced by combining coupler splitting and wavelength multiplexing with different density or granularity levels. 6.1 Pure coupler splitting A first topology to work with can be one of the most simple: a network with one splitting stage (only a power splitter), typically used in GPON [2]. In this topology, the Central Office (CO) is provided with an OLT that serves all the users. Then, a feeder fiber connects the power splitter with the OLT. Each user has a customer premises equipment (CPE), and is connected to the splitting stage by a distribution fiber. The typical fiber length between CO and CPE is of about 10-20 km. This is compatible with currently deployed PONs, but with high extension and splitting capability with the proposed transceivers. 6.2 Subband WDM tree To reduce the overall external plant loss, and maybe get a better adaptation to the laser tuneability range, WDM demultiplexers can be introduced in the PON, subdividing the optical spectrum in several sub-bands, that fit several ultra-densely spaced channels allocated in the sub-PON with coherent ONUs or narrow filtering ONUs. In this case, the network outside plant is composed by a classical dense WDM routing stage, which routes a set of ultra-dense wavelengths to a power splitter. Such a power splitter distributes the signal to each of the ONUs connected to the branch. 125
  • 142. Chapter 6. Network topologies 126 Figure 6.1: Pure coupler splitting network scheme. Figure 6.2: Network scheme and routing profile. Ultra-dense WDM wavelengths are generated at the central office by a set of tunable lasers which are modulated with downstream data. At the ONU side, the tuneable local oscillator at the customer premises equipment (CPE) selects the assigned channel and decodes downstream data. In terms of available bandwidth, the proposed network offers huge transmission capabil- ities. For example, in the experimental testbed that will be shown in section 6.4, using 2-GHz channel spacing and 1 Gb/s data rate, 32 channels can be easily accommodated in an ITU-T G.694.1 100-GHz DWDM channel. So the network can potentially serve 1280 users, offering a total capacity of slightly more than 1 Tb/s.
  • 143. Chapter 6. Network topologies 127 6.3 Advanced: WDM ring-tree SARDANA network This network is based in a WDM double-fibre-ring with single-fibre wavelength-dedicated trees connected to the main ring at the Remote Nodes (RN) (Figure 6.3); switchless add&drop is simply performed by broadband optical couplers at the ring, to maximize cascadability in terms of optical transfer function at the RNs. To hold increased power budged, remote amplification is introduced at the RN by means of Erbium Doped Fibres (EDFs), remotely pumped by 1480 nm lasers located at the CO [66]. Figure 6.3: SARDANA network architecture. Downstream and upstream signals are coupled into the corresponding ring fibers by means of the Resilient Network Interface, that allows for adjustment of the transmission direc- tion, providing always a path to reach all the RNs even in case of fiber failure. Two pump lasers are WDM coupled for bidirectional, balancing pumping and resilience against fiber failure. At the RNs, simple Optical Add-Drop Multiplex (OADM) is accomplished by two 90/10 couplers (for the OADM function) and a 50/50 coupler (for protection function). Fixed filters determine the dedicated wavelength of each Network Tree. Pump is pre- viously demultiplexed and led to the EDFs for amplification of up/downstream of each tree (in-line EDFs are also possible). Two single-fiber PON trees are connected to each RN; a 1:32 TDM PON is considered at each tree-PON network section. Using standard 2.5 Gb/s transmitters and receivers, an average bandwidth close to 100 Mb/s can be offered to each ONU. A SARDANA network with 16 RNs and 32 wavelengths serves up to 1024 ONUs. A goal on this architecture would be to upgrade it using some of the already proposed transceiver, serving the same number of ONUs but increasing the guaranteed bandwidth up to 1 Gb/s. As the users would have an assigned wavelength, a full transparent re- mote node could be foreseen, using only couplers. Also, by using homodyne technology,
  • 144. Chapter 6. Network topologies 128 remote amplification could be shut down, because of the relaxed sensitivity requirements associated to coherent systems. 6.4 Case studies After taking a look at the transmission/reception techniques developed, a full working prototype has been patented and build for the most promising one: the coherent system with time-switching phase diversity reception. With such a prototype two cases of future deployment were tested in the laboratory: • Subband WDM tree PON, featuring wavelength grooming [67]. • Ring-tree ultra-dense WDM-PON, with transparent remote nodes [68]. Both networks are based on the ultra-dense WDM concept introduced before, aiming to give service to a high number of users (around 1000), at very high speed (1 Gb/s). 6.4.1 Subband WDM tree PON In this case, the network outside plant is based on two splitting stages (see figure 6.2). Firstly, a classical dense WDM routing stage routes a set of UD wavelengths to the secondary splitting stage where a power splitter distributes the signal to each of the Optical Network Units (ONUs) connected to the branch. A stack of thermally controlled lasers located at the CO generate all the UD wavelengths which are modulated with downstream data and transmitted to the ONUs. Each ONU re- ceives the set of wavelengths that have been passed through the first WDM demultiplexer (typically a 1xN AWG). Then, the tuneable local oscillator at the CPE selects the ONU’s assigned channel and decodes downstream data by means of optical homodyning. This same optical oscillator carrier is then used for upstream transmission. It is modulated using again BPSK modulation format. At the CO a symmetric receiver is implemented (see figure 6.4). In terms of available bandwidth, the proposed network offers huge transmission capa- bilities. Using 2 GHz channel spacing and 1 Gb/s data rate, 32 channels can be easily accommodated in an ITU-T G.694.1 100-GHz D-WDM channel. On the C-band, 40- channel AWGs are commercially available so the network can potentially serve 40 x 32 = 1280 users, offering a total capacity of more than 1 Tb/s. In terms of transmission robustness, the use of a local oscillator and homodyne detection increases the system sensitivity thus enhancing the tolerance of the network to power
  • 145. Chapter 6. Network topologies 129 Figure 6.4: OLT and CPE transmission modules. losses and relaxing power budget restrictions. This permits to use the time switching phase diversity homodyne receiver in both OLT and ONU instead of a typical homodyne receiver (with an optical phase-locked loop). A network testbed was implemented to demonstrate the feasibility of the proposed net- work design. Two consecutive channels were transmitted while measuring upstream and downstream sensitivity with 2 GHz channel spacing. The impact of the interference signal was also evaluated obtaining the sensitivity penalty as a function of the channel spacing. Although the proposed receiver architecture has demonstrated to be highly insensitive to phase noise effects, sub-MegaHertz (hundreds of kHz) linewidth lasers were used in order not to have linewidth limitations and explore the full potential of the network. A DFB tuneable laser was modulated using DPSK modulation at 1 Gb/s and coupled by means of a 3 dB coupler. A 25 km fiber spool simulated the access trunk fiber and a 1 x 40 AWG acted as the first remote node (RN) and routed both signals to the output port corresponding to the D-WDM band input wavelength. Losses at the RN were measured to be 6.47 dB. Finally, the power splitter stage was implemented with a 1:32 power splitter and added 16 dB losses. Firstly, sensitivity measurements of downlink and uplink were carried out with no adjacent interfering channel. −38.7 dBm of sensitivity (BER=10−9 ) were measured (see figure 6.5). To analyse penalties due to adjacent channels, two consecutive downstream channels were transmitted adjusting the channel spacing between them. The reference wavelength was 1549.70nm. With the proposed 2-GHz spacing sensitivity worsened by about 2 dB, as shown in figure 6.6. When the interference was separated just 1 GHz, sensitivity was degraded more than 3 dB. On the other hand, when channel spacing was greater than 6 GHz, sensitivity was barely affected so penalty due to adjacent transmission could be neglected. So, almost no change was appreciated when comparing to the study performed in section 4.2.1.4.
  • 146. Chapter 6. Network topologies 130 Figure 6.5: Up-and Down-stream transmission results. Figure 6.6: Sensitivity penalty as a function of channel spacing. 6.4.2 Ring-tree ultra-dense WDM PON As shown in figure 6.7, the network topology fiber, is based on a two-stage outside plant, with a primary coupler-based feeder ring and a secondary power-splitter-based distribu- tion stage. The central ring multicasts all the wavelengths transmitted on the trunk fiber to the different networks sub-segments. Subsequently, the secondary power-splitters mul- ticast the signals to the end users. At the customer premises equipment (CPE) side, the receiver selects the wavelength that has been assigned to establish a virtual point-to-point link. Even though it is not explicitly shown in figure 6.7, this is a double fiber topology, with separate fibers for upstream and downstream. This topology has some interesting features: total transparency and scalability. The addition of a new user simply requires the installation of an optical 1:2 coupler and the
  • 147. Chapter 6. Network topologies 131 Figure 6.7: Network topology and wavelength plan. interface at Central Office (CO). Also, resilience is achieved naturally with the central ring, the bidirectional design of the RN and by means of an optical switch that connects the interfaces at the head-end with the branch of the ring that offers the best connectivity. This is shown in figure 6.8. In case of a fiber break, there is always a light path to reach all the remote nodes. Figure 6.8: Central office scheme. Transmission losses are increased due to the use of power couplers instead of wavelength multiplexers. This is the price to pay for total transparency and scalability. Homodyne detection can be used to overcome this problem, as it is known to offer much better sensitivity than direct detection systems.
  • 148. Chapter 6. Network topologies 132 A stack of thermally controlled lasers located at the CO generate all the wavelengths, which are Differential Phase Shift Keying (DPSK) modulated with downstream data and transmitted to the ONUs. Then, at the CPE, the tunable local oscillator selects the ONU’s assigned channel and decodes downstream data by means of optical homodyning, IQ phase switching, differential demodulation and post-processing, following the steps described in chapter 5. This same optical oscillator carrier is then used for upstream transmission. It is modulated using again DPSK modulation format. At the CO a symmetric transceiver is implemented similarly to the previous section (see figure 6.4). The main feature of the transceiver design used is its potential low cost. According to [62], the optical circuit used for reception (and transmission) can be integrated into an InP substrate. This may enable a potential low cost implementation of such solution. The network implemented is designed to offer connectivity to 1024 users (U ) per fiber in a 4-node (N ) configuration with 8 secondary trees with 1:128 (K) splitting factor. Other topologies are also possible taking into account that the total number of users is U = 2 · N · K and that total link losses are: L = Lf iber + (N − 1) · 10 · log(x) + N · Lex + 3 + 3 log2 (K) (6.1) where x and y are the remote node coupling factors (typically x = 0.9; y = 0.1), Lf iber is the attenuation along the fiber, and Lex are excess losses due to alignment and manufac- turing imperfections of the components per remote node. A priori, the number of users is mainly limited by the available power budget. Regarding available bandwidth in the proposed network, a 4 GHz channel spacing can be used, at 1 Gb/s data rate, as demonstrated in section 4.2.1.4. Thus, 1024 channels can be easily accommodated in the overall C-band. So the network can potentially serve those 1024 users at the same time, offering a huge total capacity, of slightly more than 1Tb/s. Please note that, unlike in section 6.4.1, in this special case the channels doesn’t have to fit in the 100 GHz WDM space, having more freedom to design the network spectrum. A possible limit of the network is the total optical power that can be launched into the ring fiber. According IEC safety rules [69], that power should not exceed 1270 mW (approx. 31 dBm) at 1550 nm. In the network plan described in figure 6.7, all the operative wavelengths are launched into the ring fiber at the same time. In the worst case, working in resilient mode, all 1024 wavelengths would be launched on the same side of the ring. Thus, the maximum power of each wavelength should not exceed 0.9 dBm. To demonstrate the feasibility of the network, proof-of-concept network experiments were carried out using the network setup detailed in figure 6.9. At the CO, a 1550 nm Distributed Feed-Back (DFB) commercial laser was externally DPSK modulated at 1 Gb/s, with a LiNbO3 phase modulator. The output power was of 0 dBm. To simulate the case in a UD-WDM scenario, a 4-GHz-spaced signal was
  • 149. Chapter 6. Network topologies 133 Figure 6.9: Experimental network testbed also inserted, using another continuous wave laser. A 30 km fiber spool was used as the access trunk fiber to reach the first remote node (L=5.2 dB). This remote node (RN) was based on two 90/10 couplers and a 3-dB coupler to connect to two access trees. Insertion losses of this device were measured to be 1.6 dB for pass-through signals and 13.2 dB for dropped signals. Finally, the second distribution stage, emulated by means of a variable optical attenuator (VOA), added 21 dB losses to the link, corresponding to 1:128 splitting. The network tested had four remote nodes (RN) thus total outside measured plant losses ranged from 39.4 dB on RN1 to 44.2 dB on RN4. In standard operation, the optical switch at the head-end would be configured to connect the users to the light path with less lossess. The case of passing through three remote nodes to reach RN4 simulates the network working in resilient mode after a fiber break in the worst case (between the head-end and RN 4). Three different cases were investigated: RN1, RN2 and RN4. RN2 represents the worst case in standard operation and RN4 the worst case in resilient mode. Results are shown in figure 6.10. Regarding sensitivities, for a BER of 10−9 in RN1 sensitivity was measured to be −43 dBm and in RN2 a sensitivity of −41.3 dBm was obtained. Thus, the system performed correctly in standard operation mode. In resilient mode, a BER of 10−9 at RN4 could not be reached due to a BER-floor. That floor is explained as deriving from the linewidth of the system (1MHz), and the frequency response distortion of the microwave mixers used at the reception side of the transceiver. Nevertheless, a BER of 10−6 was measured at −44.3 dBm. To compare with the other cases, in RN 1 BER of 10−6 was measured at −48.1 dBm, whereas in RN2 it was measured at −46.3 dBm. A Forward Error Correction (FEC) coding scheme, with only 7% overhead, is able to recover (quasi) error-free data for a raw BER below 10−3 [11]. So,
  • 150. Chapter 6. Network topologies 134 Figure 6.10: Sensitivity results Normal operation Resilient mode RN1 RN2 RN4 RN1 RN2 RN4 Sensitivity −43 dBm −41.3 dBm - −49.1 dBm −49.3 dBm −49.1 dBm Link Losses 39.4 dB 41 dB 44.2 dB 39.4 dB 41 dB 44.2 dB Power Budget 42.9 dB 41.2 dB - 49 dB 49.2 dB 49 dB Table 6.1: Power budget summary FEC strategies should be applied to increase BER over 10−9 in resilient mode. In that case one can work with an increased sensitivity of −49.1 dBm. Penalty to adjacent channels at 4-GHz channel spacing was measured to be 0.1 dB, as expected. Table 6.1 I shows a summary of the experiments carried out. There one can see that the power budget for the best case (RN1) is of 42.9 dB, including the channel spacing penalty. For the worst case in normal operation, power budget is of 41.2 dB. At a first glance, one could think on doubling the number of users supported in RN1 and RN4 in order to balance such power budget, while increasing the total number of users up to 1536. In that case, to maintain the total output power at the CO, each wavelength power should be reduced to −0.8 dBm. Consequently each power budged would be reduced by 0.8 dB, and in RN2 link losses would become higher than the power budget. So, even the power budget is not balanced at all for the different cases, the maximum number of users to serve is 1024, due to safety power limits.
  • 151. Chapter 6. Network topologies 135 6.5 Chapter summary In this chapter, several advanced UD-WDM access solutions have been presented. Among them, two case studies have been experimentally demonstrated: A subband WDM tree, and a Ring-tree ultra-dense WDM network architectures. In both cases, the networks potentially offers terabit transmission capabilities, and are based on the transmission of narrowed-spaced dedicated wavelengths to each end user and optical homodyning to tune and decode each transmission channel. For the first case, transmission experiments showed a network sensitivity of −38.7 dBm after 25 km, and a sensitivity penalty of about 2 dB in the case of 2 GHz channel spacing, meaning that it can potentially serve up to 1280 users at 1 Gb/s. The Ring-Tree ultra-dense WDM network is capable to provide a flexible and scalable architecture with large capabilities in terms of number of users (up to 1024) or in capacity (more than 1 Tb/s). Also, it features a completely passive outside plant, wavelength- transparent remote nodes and high transmission capabilities and resilience. Transmission experiments at 1 Gb/s show a sensitivity of −43 dBm in the first RN, after 30 km and a power budget of 42.9 dB. For the worst case (RN4), when the network is operating in resilient mode, BER better than 10−9 could be reached by applying FEC strategies. In that case a sensitivity of −49.1 dBm was achieved, showing a power budget of 49 dB. Thus, 1 Gb/s per user could be guaranteed, achieving a total network capacity of 1024 Gb/s.
  • 152. Chapter 7 Conclusions and future work 7.1 General conclusions The use of coherent systems in access networks appear as a promising mid to long-term solution to the high-speed and ultra dense PONs. By replacing the transceivers at both the central office and customer premises, a dedicated wavelength per user is allowed. On this basis, the present thesis identifes and addresses the inherent problems of coherent transceivers, while mantaining low cost. The overall goals of this study were homodyne OLT and ONU designs for upgrading the current standard passive optical networks. In order to provide some background, a survey of the evolution of optical homodyne systems was performed, dealing with their fundamental problem: the phase noise. Special attention was paid to BPSK and DPSK modulation formats, due to their simplicity and robustness. Furthermore, the main OPLL and phase diversity techniques have been analyzed. Therein, the phase noise impact has been the main impairment to report on. Next, several coherent detection techniques have been proposed, improving the perfor- mances of the receivers that shape the current state of the art. A novel OPLL has been analyzed and prototyped, reaching an improved phase noise tolerance of 3.1 MHz with low cost optical components. Also, several new phase diversity systems and phase estimation algorithms have been analyzed: Karhunen-Loeve series expansion phase es- timation, Fuzzy logic data estimation, time-switched phase diversity, and time-switched polarization diversity. Among them, the Karhunen-Lo`ve series expansion phase estimation should be high- e lighted, which can tolerate linewidth up to 11% of the bitrate, for the 1 dB penalty point at a BER of 10−3 . A very simple and robust architecture has been prototyped, featuring time-switching phase diversity and tolerating linewidths up to 1.8% of the bitrate. i. e. at 1 Gb/s it 137
  • 153. Chapter 7. Conclusions and future work 138 can tolerate linewidths up to 18 MHz (BER-floor 10−3 ). For such architecture, channel spacing has been evaluated in both senses, theoretically and experimentally, achieving a 3 GHz spacing for 1 dB penalty at 10−9 BER. Another architecture that should be taken into account, is the direct drive time switching. With it, the linewidth tolerance is enhanced, permitting linewidths up to 5.4% of the bitrate (BER-floor 10−3 ). A first experimental tuning has been made, confirming the √ expected optimum value of 2 for the gain factor. Fuzzy logic data estimation and time-switched polarization local diversity have been analyzed, demonstrating the concept theoretically. Furthermore, several transceiver architectures have been proposed and discussed, some of them including well-known techniques (as the full phase diversity receivers) whereas others have been designed using the novel approaches previously reported in this thesis. The trade-off between performances and cost has been difficult to overcome, but finally the decision of implementing the time-switched diversity transceiver has been made. Ad- ditionally, OLT designs searching for a time-switched polarization diversity have been discussed, proposing a couple of alternative implementations, one based in birefringent modulators and the other one based in Faraday rotating mirrors. Finally, the upgrading of PON architectures has been discussed for implementing full ultra-dense WDM networks. Laboratory testbeds have been developed for two of the presented schemes (subband WDM tree and ring tree transparent ultra-dense WDM networks), which demonstrate that the proposed topologies and transceivers are feasible, achieving transmission of up to 1 Gb/s in links higher than 25 km. 7.2 Future lines During this Ph.D. thesis, research has been developed in access networks towards an optimal coherent transmission and reception technique for reaching high density, power budget, and bandwidth efficiency; points that were mentioned in the introductory chap- ter. Moreover, proof of concept experiments were performed. However, there are some improvements to be achieved, that will result in a significant step forward: 1. Compact coherent transceiver. 2. Full bidirectionality over a single fiber. 3. Spectrum management. All of them are going to be briefly commented in the next subsections, as well as the approaches that can be followed.
  • 154. Chapter 7. Conclusions and future work 139 7.2.1 Compact coherent transceiver A compact and simple version of the transceiver can be developed, in the sense that less number of devices should be used. This can be achieved by using a digital signal processor in order to perform all the operations to cancel the noises, and recover data. In principle it should contain the key modules: Phase recovery and/or estimation, and automatic frequency control of the local laser (in order to keep it in the same wavelength as the received signal). However this presents some problems to solve before implementation: 1. State of polarization mismatch between local oscillator and received signal. In order to have an optimal reception, they have to be coincident. This is critical issue. A consequence may be that more optical components are needed at the user premises (higher cost). 2. Careful design of the modulation formats to be used. In principle the optimum mod- ulation format is PSK for both, upstream and downstream, as stated before. But, reusing the local laser may entail an additional external device. So other modulation formats that can be generated using only a laser (no external modulator), can be investigated. Along this line, also coherent optical Orthogonal Frequency Division Multiplexing (OFDM) techniques can be studied, as they have been demonstrated to be more robust against fiber dispersion and other impairments. 7.2.2 Full bidirectionality over a single fiber This is a key issue for simplifying the access network architecture and save costs. In prin- ciple, using only one fiber per user in the last mile, the cost can be reduced significantly; since it is the part of the external plant of the network that is not shared among all users. Thus, a big effort should be done when dealing with bidirectionality over a single fiber. Nevertheless, there are a couple of main undesired fiber effects to mitigate: 1. Rayleigh backscattering. This refers to the amount of backscattered light when propagating a light beam over a fiber. It basically depends on the amount of fiber on which the light beam is launched. This backscattered light becomes a substantial interference to the received signal, for fiber lengths of more than 10 km [59]. 2. Light reflections. They are mainly due to optical connectors, which may reflect a part of the incoming optical power. It also represents a non-negligible interference at the receiver side. However, it can be mitigated by fusion splicing as many network components as possible, and using angled connectors. Please note that these impairments are strongly related to the modulation format used for upstream and downstream, so they have to be taken into account when designing the modulation format to be used.
  • 155. Chapter 7. Conclusions and future work 140 7.2.3 Spectrum management In the case of 1000 different wavelengths propagating over a single fiber, before the dis- tribution stage of the access network, they can be interfering one to each other. Also, depending on the modulation format used, upstream and downstream spectra can be interfering at the same wavelength. Thus, not only a careful design of the modulation format is a must, and there is a need to work in additional topics: 1. Spectral efficiency maximization. This includes the proper dimensioning of the spacing between channels, as well as the maximum number of users to serve simul- taneously. 2. Wavelength monitoring, control and stabilization, in order to limit the maximum wavelength drifts of the lasers used, that can produce an added penalty in the transmission system. Commercially available lasers emit at 1550 nm ( 193 THz, when propagating over fiber), thus a high spectral purity of 1 ppm means that the laser should be stable in a 200 MHz range. This is a substantial drift for the intended low channel spacing ( 3 GHz). This monitoring and control should be centralized at the CO premises. An important problem can be the characterization of the lasers that are going to be used at each side (central office, and user premises). At least, at central office side, lasers are expected to switch from one wavelength to another, distributing the traffic among the users, and each type of laser has its switching transient characteristics.
  • 156. Appendix A Passive optical network solution using a subcarrier multiplex A.1 Introduction Access Passive Optical Networks (PONs) have emerged as an effective platform to pro- vide advanced bandwidth demanding services to the final users. Thus, next generation PONs have to match several issues, including high bandwidth delivering and Wavelength Division Multiplexing (WDM). An interesting specification is the use of one single fiber for both, upstream and downstream transmission, to reduce the size of the network’s external plant and the complexity of the Optical Network Unit (ONU). In the past, some advanced designs which avoid the generation of light at the ONU by using different modulation schemes for downlink and uplink transmission have been demonstrated [70–74]. However, these methods may not be cost-effective due to the components needed for the modulation and detection of the upstream and/or downstream signals. A more cost-effective solution is the use of SubCarrier Multiplexing (SCM) techniques with direct modulation [75, 76]. But they have been limited to 1.25 Gb/s downstream, with the electrical schemes not optimized. In this appendix, a 20 km full duplex PON will be analyzed operating at a rate of 2.5 Gb/s downstream and 1.25 Gb/s upstream, using the SCM technique and giving service to a maximum of 1280 users. Aiming to reach the user with single fiber-scheme, the tolerance against Rayleigh backscattering ratios is evaluated. 141
  • 157. Appendix A. Passive optical network solution using a subcarrier multiplex 142 A.2 Receiver scheme The subcarrier transmission system has two main parts: The first is an electrical 2.5 Gb/s DPSK upconverting module, inside the transmitter of the OLT, prior to optical modulation. The second part is an electrical delay-and-multiply DPSK detector placed after photodetection, at the ONU. This scheme avoids the need of an electrical oscillator at the receiver, increasing its simplicity. For a downstream rate of 2.5 Gb/s, the subcarrier frequency is fixed at 5 GHz, allowing a good band margin between downstream and upstream. As the 3dB bandwidth of the data stream is approximately 4 GHz (double-sided), the photo-detection bandwidth needed is of almost 7 GHz. Thus, for maintaining the BPON/GPON link compatibility in terms of sensitivity at the specified bandwidth, an Avalanche Photo-Detector (APD) should be used instead of standard PIN diodes. It is shown in [77] that the electrical SNR after an APD can be written as: 1 2 (mM PS ) 2 SNR = 4kT F ∆f (A.1) RL + 2qFa ( PS + Id )∆f where m is the modulation index, M is the multiplication factor of the APD, is the photodiode responsivity, PS is the received optical power, k is the Boltzmann constant, T is the room temperature (in K), F is the electronic receiver amplifier noise factor, RL is the photodiode load resistor with a 50 Ω nominal value, ∆f is the signal bandwidth (one-sided), q is the electron charge (in C), Fa is the excess noise factor, and Id is the photodiode dark current. The modulation index can also be expressed as [78]: 1 − ER m= (A.2) 1 + ER where ER is the Extinction Ratio, in linear units, defined as the off-state power over the on-state power. On the other hand, according to the SNR definition of equation A.1, the probability of error can be easily found for an electrical DPSK signal as [9]: 1 Pe = exp(−SNR) (A.3) 2 The next section shows the experimental implementation of the SCM-system by means of a simplified electrical scheme, showing a good agreement with equations A.1 and A.3. Also, some deployment scenarios are proposed and discussed.
  • 158. Appendix A. Passive optical network solution using a subcarrier multiplex 143 A.3 Experiments and discussion Two kinds of measurements were performed for the system modelled. First of all, only SCM downstream measurements were carried out, for validating the theoretical model. Afterwards, full-duplex transmission over a single fiber was characterized. A.3.1 Downstream characterization Figure A.1 shows the experimental setup. For the 2.5 Gb/s downstream signal, a Mach- Zehnder Modulator (MZM) is used for modulating a CW laser signal, as a first approach that allows a better control of the modulation. However, a more cost-effective solution can be the use of direct modulated lasers. In our case, an ER of 8.4 dB can be easily achieved, enough for our purposes as it will be shown. Figure A.1: Half-duplex experimental setup. The downstream signal, a PRBS with a length of 231 −1, is precoded inside the first Pulse Pattern Generator (PPG1) and mixed with a 5 GHz electrical oscillator by using a stan- dard double balanced mixer. The electrical oscillator is not synchronized with the PPG1 clock, constituting a more realistic platform and allowing the analysis of frequency drift tolerances. Along a different line, the band-pass filter typically used at the transmitter side is not required, thanks to the mixer’s frequency response, which ensures a band-pass filtering as shown in figure A.2. The mixer’s bandwidth has been measured by using the three-mixer method [79], showing: a bandwidth of ± 1.9 GHz, centred at 5 GHz, enough for 2.5 Gb/s; and a rejection better than 20 dB for frequencies beyond ± 2.9 GHz. At the electrical side of the ONU receiver a delay-and-multiply scheme [9] was imple- mented using a double balanced mixer. Since it does not require any electrical oscillator placed in the ONU, data detection becomes simpler and phase-locking between detected carrier and electrical oscillator is avoided. The sensitivity and extinction ratio penalty measurements, at a reference wavelength 1550 nm, are shown in figures A.2 and A.3.
  • 159. Appendix A. Passive optical network solution using a subcarrier multiplex 144 Figure A.2: Low pass equivalent of the mixer’s response for a 5 GHz carrier. Figure A.3: Sensitivity results for setup described on figure A.1 Applying the specific characteristics of the APD employed and applying the model de- scribed in section A.2, a theoretical sensitivity of −28.3 dBm for a BER of 10−10 has been obtained. It did not fully match with the measurements at a PRBS length of 231 − 1. In this case the sensitivity was −26.6 dBm for a BER of 10−10 . Using a shorter length of 27 − 1, a sensitivity of −28.2 dBm is obtained, only 0.1 dB away from theoretical val- ues. The sensitivity differences between different PRBS lengths are explained as deriving from the ripple of the mixers response and the low frequency response of the base-band amplifiers used (cutting off at 30 kHz). Regarding extinction ratio penalty for BER at 10−10 , shown in figure A.4, the baseline ER of 8.4dB provided by the OLT-TX generates a penalty of 1.5 dB, both experimentally or theoretically. Also, experimental points follow quite well theoretical calculations, just with a deviation of less than 1 dB.
  • 160. Appendix A. Passive optical network solution using a subcarrier multiplex 145 Figure A.4: Downstream power penalty at BER 10−10 due to extinction ratio. Square points are experiments, whereas continuous line is derived from Eqs. 1 and 2. A.3.2 Full-duplex measurements For this case, the experimental setup is shown in figure A.5. Downstream parameters for this configuration were exactly the same used before, when testing only downstream transceivers’ interfaces. The splitting ratio at the input of the ONU was 50:50 and the reference wavelength was 1550 nm. For the 1.25 Gb/s upstream signal, another pulse pattern generator (PPG2) was used, independent from PPG1 and driving an RSOA in the ONU. Data injected was operating with a PRBS length of 231 −1. The RSOA small signal gain was of 17 dB. The downstream signal is modulated by the RSOA with the upstream data, and sent back to the OLT. Note that the downstream signal is not constant power, but sinusoidal. At the OLT side, after photo-detection, a low pass filter was placed in order to properly reject re-modulation noise from downstream signal. Figure A.6 shows the detected power spectrums at the ONU and OLT sides before and after filtering. The residual downstream signal detected could be rejected by more than 20 dB. The sensitivity has been also evaluated and shown in figure A.7. Both directions were measured (upstream and downstream) at the reference wavelength of 1550 nm. For the downstream −23.4 dBm wer achieved for a BER of 10−10 , whereas for the upstream, −22.6 dBm were measured. Please note that downstream sensitivity is 3.3 dB away from the value obtained in figure A.3. This is explained by the 3 dB coupler present between the ONU input and the APD of the SCM receiver (shown in figure A.5). Please note that upstream curve is much steeper than downstream. This is due to the fact that the VOA is placed in between OLT and ONU, attenuating not only the signal transmitted by the
  • 161. Appendix A. Passive optical network solution using a subcarrier multiplex 146 Figure A.5: Experimental setup for single fibre full-duplex measurements. ONU, but also the RSOA input power. Therefore, the OSNR is degraded at the ONU output. A.4 Network measurements Based on the proposed ONU/OLT designs shown in figure A.5; three different PON scenarios were tested, trying to keep a minimum system margin of 3 dB: first a pure TDM network with low coverage (8 users) used as a reference; secondly an standard medium hybrid TDM/WDM PON covering 160 users; and finally an optimized network capable to serve 1280 users. It must be pointed that the 3 dB splitter present at ONU stage in figure A.5, was replaced by a 70:30 coupler, with 30 % output connected to the RSOA. In this first case, total network losses were measured to be 16 dB. The ONU input power was of −16 dBm, and its output power was measured to be of +1 dBm, meaning that the ONU net gain was of 17 dB and the RSOA gain was of 20.5 dB. 0 dBm were injected from the OLT to the feeder fibber, generating a measured Rayleigh Backscattering (RB) of −33.5 dBm (having a 17.5 dB of upstream optical Signal to Rayleigh-backscattering Ratio, OSRR). Under these conditions, upstream and downstream transmission curves were measured. Results are shown in figures A.9 and A.10. For the downstream a sensitivity of −20.1 dBm was found, while for the upstream the sensitivity found was of
  • 162. Appendix A. Passive optical network solution using a subcarrier multiplex 147 Figure A.6: Electrical power spectrums after photo-detection at the receiver side: (a) before electrical filtering at the ONU, (b) after electrical filtering at the ONU; (c) before electrical filtering at the OLT, and (d) after electrical filtering at the OLT. Figure A.7: Sensitivity results for the proposed OLT and ONU architectures. −18.4 dBm, both for a BER of 10−10 . This means that the achieved downstream power budget was of 20.1 dB (system margin of 4.1 dB), whereas the upstream power budget was of 19.4 dB (system margin of 3.4 dB). The second test-bed was composed of a 16 km feeder spool, followed by a 1:40 AWG demultiplexer. Next to it, a 2.4 km distribution spool was preceding a 1:4 power splitter and a 2.2 km drop spool. This is shown in figure A.11. This scenario was for an area with medium density of users, capable to serve up to 160 users.
  • 163. Appendix A. Passive optical network solution using a subcarrier multiplex 148 Figure A.8: Scenario 1 schematic. Figure A.9: Downstream sensitivity curves for the three different network scenarios. Total network losses were measured to be 16.4 dB. The ONU input power was −16.4 dBm, and its output power was measured to be of +0.8 dBm, meaning that the RSOA gain was of 20.7 dB. Also, 0 dBm were injected from the OLT to the feeder fibber, generating a RB of −33.5 dBm; and having a 17.5 dB of upstream OSRR, too. Again, upstream and downstream transmission curves were measured. Results are shown in figures A.9 and A.10. For the downstream the sensitivity found was of −20.6 dBm, while for the upstream the sensitivity found was of −19 dBm, both for a BER of 10−10 . This means that the achieved downstream power budget was of 20.6 dB (system margin of 4.2 dB), whereas the upstream power budget was of 19.8 dB (system margin of 3.4 dB). Finally, an upgraded third test-bed was assembled. In this case, a double fibre feeder is proposed, so the RB amount generated was very low; only due to de distribution and drop stages. Now a couple of 16 km feeder spools were implementing a double fibber path from the OLT to the AWG. Next to it, a 2.4 km distribution spool was preceding the 1:32 power splitter and a 2.2 km drop spool. This is shown in figure A.12. Also, an optical preamp was used for upstream detection, since there was no RB limitation, and OLT costs are shared among all users. This scenario was intended as an improved access network, covering as many users as possible (up to 1280). Now, total network losses were measured to be 26.6 dB. The ONU input power was of
  • 164. Appendix A. Passive optical network solution using a subcarrier multiplex 149 Figure A.10: Upstream sensitivity curves for the three network scenarios. Figure A.11: Schematic of scenario 2. −16.6 dBm, and its output power was measured to be of +0.8 dBm, as in the second scenario. Please note that the Rayleigh backscattered signal was negligible, and the OLT output power could be increased to +10 dBm. Under these optimized conditions, upstream and downstream transmission curves were measured. Results are shown in figures A.9 and A.10. For the downstream the sensitivity found was of −19.6 dBm, while for the upstream the sensitivity found was of −28.8 dBm, both for a BER of 10−10 . This means that a symmetrical power budget of 29.6 dB (system margin of 4 dB) was achieved. A.5 Conclusions The bi-directional full-duplex 2.5 Gb/s / 1.25 Gb/s was demonstrated in a SCM sin- gle fibre PON. The downstream signal is DPSK coded and up-converted by using a 5 GHz subcarrier, while the upstream data is transmitted in burst-mode NRZ. A simplified
  • 165. Appendix A. Passive optical network solution using a subcarrier multiplex 150 Figure A.12: Scheme for scenario 3. scheme for the electrical parts of the ONU and OLT has been proposed and demon- strated. Also a theoretical model for SCM downstream is proposed and experimentally validated. Furthermore, three different deployment scenarios are evaluated: Large cov- erage area and low density of users; area with medium density of users; and improved access network, covering as much users as possible. In the first and second test-beds, transmission experiments have shown a network power budget of 20 dB, for a single fibre configuration, combined with the capacity to serve up to 160 users. For the third one, the power budget could be increased up to 29 dB, matching clearly the typical values of GPON deployments, and serving up to 1280 users. In all the cases a minimum system margin of 3 dB was achieved.
  • 166. Appendix B Automatic wavelength control design B.1 Introduction Homodyne coherent optical reception received great attention at the beginning of the 90s. It presents many advantages with respect to the conventional direct detection be- cause of its excellent wavelength selectivity, high sensitivity and tuneability performances. However, there have been never found practical application, mostly due to the complex receiver structure and the stringent linewidth and loop delay requirements [3]. Recently, some approaches to solve this have been proposed using ultra-fast digital signal pro- cessing to estimate and track the carrier phase [27, 28]. However, all these techniques tolerate only a rather small frequency difference between LO and transmit laser. Thus, a very important point regarding receiver implementation is the frequency estimation and locking. Until now, for the analog domain several designs have been proposed, coming from RF techniques [6], and being tested in optical communications [80]. Also, two main lines have been followed for the digital domain: the off-line processing and the real-time approach. For the off-line approach, the same receiver architecture is used, almost all times [63, 64]. There the challenge is the development of feed-forward processing algorithms that can compensate the frequency offset between local laser and received signal. This study presents a reliable, fast and real-time wavelength control for intradyne co- herent reception, based on a single side-band optical VCO; intended for, at least, 8-PSK modulation format. 151
  • 167. Appendix B. Automatic wavelength control design 152 B.2 Loop design and performances The overall loop proposed is shown in figure B.1. It has a 90 optical hybrid coherent reception front-end, driven by a local optical Voltage Controlled Oscillator (VCO); a delay-and-multiply frequency discriminator, and a loop control filter. The VCO device used is an optical VCO that can be controlled in a stable and fast way, like an electrical one. Its conceptual scheme is detailed on figure B.2. The objective is to control a Single Side Band (SSB) tone modulation of the local laser. For that reason an IQ modulator is used. Details of its performance and characterization for using it in optical phase-locking can be found in [81]. Figure B.1: Scheme of the proposed analog frequency estimation loop. Figure B.2: Optical SSB-modulation VCO. Regarding the frequency discriminator, it was found that a good and reliable implemen- tation can be just a delay-and-multiply architecture that has been deeply studied and reported in [6, 35]. In order to evaluate the performances of the proposed receiver architecture, several sets of simulations were run for different parameters: frequency discriminator performances, loop delay impact and phase noise impact. Regarding frequency discriminator performances, a
  • 168. Appendix B. Automatic wavelength control design 153 first set of simulations was configured for an RZ-8-PSK scheme, with a coherent receiver followed by a digitization stage. Since it is digitized, the proposed frequency discriminator could be compared with a theoretical solution that can be used in off-line processing: a linear fit of the phase curve. Results are shown in figure B.3. There it can be observed that, while the proposed scheme gets almost unaffected, the other has a discontinuity around +-1250 MHz. This is because in the last case, a power-of-eight block is required to remove the 8-PSK modulation, and this puts a cycle-slipping indetermination on that frequency, that has to be taken into account when estimating the frequency. Figure B.3: Frequency discriminator output vs. frequency difference between LO and received signal. After characterizing the frequency discriminator with the proposed modulation, a nu- merical analysis was made, departing from the theoretical equations that arise from loop linearization [6, 35]. Regarding this numerical analysis, the transfer function was found to be expressed as: K · Kp · S + K · Ki HS (S) = (B.1) (1 + K · Kp )S + K · Ki K = 8πKLO 2 RL PS PLO 2 (B.2) 1 + K · Kp T = (B.3) K · Ki where S is the complex angular frequency, KLO is the sensitivity of the optical VCO module (Hz/V), is the photo-detector’s responsivity, RL is the load impedance, PS is the received optical power, PL O is the local oscillator optical power, Kp is the proportional
  • 169. Appendix B. Automatic wavelength control design 154 gain of the loop filter, Ki is the integral gain of the loop filter, and T is the time constant of the system. From this model, the loop delay impact was evaluated. The output parameter evaluated was the setting time at 10% error from the final value, for a low frequency step (100 MHz). Results are shown in figure B.4. There it is shown that when loop delay is about one fifth the time constant, the system performances decrease dramatically. Figure B.4: Loop delay impact on loop setting time. Afterwards module for the full loop (no data) was implemented using the software VPI- transmissionMaker. With this model, the phase noise impact to the frequency control could be evaluated. Results are shown in figure B.5. As expected, it introduces a residual error, with gaussian statistics, that cannot be tracked nor compensated at all. The vari- ance of such error is proportional to the laser linewidth, as expected. Thus, this error can not be compensated at all, and a separate phase compensation method has to be used. B.3 Practical implementation In figure B.6 a schematic of the overall loop to be implemented is shown. Of that loop the optical VCO was characterized and afterwards, with the SSB VCO, some preliminary measurements (based on off-line processing) were carried out for the frequency estimator. Regarding VCO characterization, a result of the maximum hold function for the output spectrum of the optical VCO is shown in figure B.7. There, it is shown that such an optical
  • 170. Appendix B. Automatic wavelength control design 155 Figure B.5: Error signal variance vs. laser linewidth. Figure B.6: Schematic to be implemented. VCO has a residual amplitude tilt, with a slope of 2.8 dB. Also, when approaching to the limit of 4 GHz, a high peak appears at the opposite side of the spectrum. This peak corresponds to a worst case condition, and is 12.7 dB below the principal component. Precisely, this is shown in figure B.7 for the -4 GHz case. That range of 4 GHz is set by the RF components used (hybrid, amps and filters) and is enough for our purposes. Of course, by carefully setting these RF components, the performances of the VCO can be substantially improved. Afterwards, it was compared to other two optical VCO approaches. Results are shown in table B.1. One approach is based on tuning the current of the phase section of a GCSR laser. It presents some difficulties for setting the appropriate working point, and could present some hysteresis, depending on the operation wavelength range [82]. Data
  • 171. Appendix B. Automatic wavelength control design 156 Figure B.7: Max hold function for the output spectrum of the optical VCO. presented here has been obtained from [82], where a GCSR laser model NYW 30-009 from ALTITUN, was characterized. A second alternative VCO is a DFB laser, working in a saturation point and modulating its adiabatic chirp for changing the wavelength. However it could give a high residual amplitude change. For this approach, data was retrieved from basic measurements carried out of a Panasonic LNFE03YBE1UP. Laser Tuning Tuning Frequency Residual linewidth speed range slope amplitude SSB-IQ mod. 100 kHz 10 MHz 2 GHz 260 MHz/V 2.8 dB GCSR 60 MHz 100 MHz 18 GHz 320 MHz/V 0.1 dB DFB 1 MHz 1 GHz - 1.4 GHz/V 7.4 dB Table B.1: Comparison between possible optical VCO approaches. In order to see the potential when using the proposed frequency discriminator, some off-line experiments were carried out. The setup used is depicted in figure B.8. It is a back-to-back version of the 10 Gbaud RZ-8PSK (30 Gb/s) transmission system reported in [83], with the proposed optical VCO. There, the transmitter consists of a laser that for RZ pulse carving a Mach-Zehnder Modulator (MZM) is used. Regarding the 8-PSK signal, it was generated by an IQ-modulator, which generated an optical QPSK signal, and a consecutive phase modulator, which was used for the additional π/4 phase modulation. The transmitted data signal was a 211 − 1 pseudo-random bit sequence, given by a Pulse- Pattern Generator (PPG). These signals were given to the modulator inputs with different delays to ensure their independence.
  • 172. Appendix B. Automatic wavelength control design 157 At the receiver side, the RZ-8PSK signal was interfered with the optical VCO in a LiNbO3 2x4 90 hybrid. The output signals of the hybrid were detected by two balanced detectors, and the I and Q outputs were sampled using a 50 GSa/s digital storage oscilloscope. Figure B.3 shows the estimation results, for two different discriminators: Linear curve fitting, and delay and multiply method. 3 different cases were compared: No frequency difference, 3 GHz of frequency difference, and 3.5 GHz of frequency difference. There it is shown that for all cases of frequency difference, both methods, linear fitting and delay-and-multiply, work as expected, but with a non-negligible error. Those differences are attributable to the laser instability. Figure B.8: Experimental setup. B.4 Conclusions An automatic frequency control design has been proposed and demonstrated. It is based on a high-speed optical VCO. First proof-of-concept experiments have been performed, showing the feasibility of the main components of the loop and a good agreement with simulation results.
  • 173. Appendix C Static and dynamic wavelength characterization of tunable lasers Tunable laser is a relevant critical device that nowadays is being pursued by several vendors worldwide. The purpose of this study is to analyze the static and dynamic behaviors of new generation tunable lasers for advanced applications like dynamic WDM- PON networks and Optical Burst Switching networks. In this case, the tuning process of tunable lasers is a critical issue in the performance of these lasers, usually not straightforward to obtain and to optimize. The main results of the present test were two: first a working points map for all possible reflectors and phase currents of the tunable laser (a modulated-grating Y-branch (MG-Y) laser, Syntune 3500) was obtained; and next a high-resolution 3-dimensional wavelength-time-power measurement of the tuning process of the laser was measured, obtaining the optical spectrum at every instant and its evolution along the tuning transient. With this, it is easy to identify, not only the wavelength temporal drift, but also the transitory mode hopping or interferences over other wavelength channels. Subsequently, this tool can be used to optimize the driving currents waveforms to improve the tuning process of the laser module. C.1 Experiments and discussion The laser characterized was a tunable modulated-grating Y-branch laser (Syntune 3500). First, a complete map of wavelengths versus reflectors and phase currents was measured. Next, switching between different wavelengths using the reflectors electrodes has been characterized. 159
  • 174. Appendix C. Static and dynamic wavelength characterization of tunable lasers 160 C.1.1 Static characterization: wavelength map The aim of this experiment was to obtain a wavelength map as a function of the tuning currents (phase and reflectors). Such a map was used afterwards, to locate the different wavelength modes and visualize the regions at which they remain stable. C.1.1.1 Static characterization setup The test setup used for characterizing the wavelength modes location and to find the stable regions of the laser is depicted in figure C.1. As one can notice in the diagram, a multi current source was used to drive the laser currents: active layer (IACT ), phase (Iph ), reflector 1 (Iref 1 ), reflector 2 (Iref 2 ) and SOA (ISOA ). Moreover, a wavelength meter was employed to measure the output beam wavelength. Both instruments were controlled by a PC through GPIB port. During the experiment, the active layer and SOA currents were kept constant, while phase and reflectors currents did a complete sweep within their range. Precisely, Iph was working in the 0-5mA range, while Iref 1 and Iref 2 were between 0 mA and 30 mA. Figure C.1: Experimental setup for stability regions characterization. C.1.1.2 Static characterization results An example of the so called wavelength map can be observed in figure C.2 (a). In this figure the wavelength (colour scale) is plotted as a function of both reflectors currents for a fixed phase current (2.4 mA). In the image one can clearly identify large conical regions (super-modes). Such super-modes start at coordinates (2;2) mA and show abrupt wavelength changes among them. In order to obtain a most accurate knowledge of the mode regions, the wavelength maps were processed using morphological image techniques. The regions shown in figure C.2 (b) were obtained after processing the wavelength map for Iph = 2.4 mA. In this figure
  • 175. Appendix C. Static and dynamic wavelength characterization of tunable lasers 161 the super-modes can be easily identified. Furthermore, one can also depict that each super-mode contains various scale shaped regions. Each of these regions correspond to a different mode. Figure C.2: (a) Wavelength map: Plot of the wavelength (colour scale) in function of reflector currents. (b) Logic stable regions map in function of reflector currents. The phase current for a) and b) is Iph = 2.4 mA. The wavelength spacing between consecutive modes is much bigger between two consecu- tive modes placed in different super-modes (∆λ in the order of 5 nm), than between two consecutive modes placed in the same super-mode (∆λ in the order of 0.5 nm). It can be also observed that the wavelength decreases within a mode region by increasing both reflectors currents (approximately at a ratio of 0.08nm/mA). The fact of changing the phase current has much smaller consequences than changing the reflector currents. However, two effects can be observed while increasing Iph . One of the effects is depicted in figure C.3 (a). In this figure, the wavelength of a fixed point (Iref1= 22.8 mA, Iref2= 8.6 mA) is plotted as a function of the phase current. As can be noticed in the plot, the wavelength of a fixed point decreases as the phase current increases (approx. 0.1nm/mA). The other effect arises when comparing figure C.3 (b) and (c). These figures represent the wavelength maps for Iph = 1.8 mA and Iph = 2.2 mA. Into each wavelength map, several fixed points (in terms of reflector currents) are plotted in order to see more easily the effect produced. If the points of figure C.3 (b-c) are observed carefully, one can realize that in figure C.3 (b) they are located just above the limit of a certain region, while in figure C.3 (c) they are placed under that limit. Therefore, when phase current increases the complete map shifts towards higher reflector currents. Consequently, if Iph is increased enough (2-3 mA) a fixed reflector’s bias point jumps from one mode to another, producing an abrupt change of wavelength (see figure C.3 (a) around Iph = 2 mA). Finally, the wavelength drift was measured for a change of the gain current. Three plots of the wavelength versus the gain current for different modes can be observed in figure C.4. In all the graphs, one can depict a quasi-linear wavelength increment while increasing the
  • 176. Appendix C. Static and dynamic wavelength characterization of tunable lasers 162 Figure C.3: (a) Plot of the wavelength in function of the phase current. Reflector currents are biased at Iref 1 = 22.8 mA and Iref 2 = 8.6 mA. (b, c) Wavelength region map as a function of both reflector currents for a phase current of 1.8 and 2.2 mA, respectively. gain current. Please note that the laser is highly stable, so the drift of wavelength per unit of current is quite low ( 0.001 nm/mA). Figure C.4: Plots of the wavelength as a function of the gain current for different reflector currents (Iph = 2.4 mA): (a) Ir ef 1 = 10.8 mA, Iref 2 = 29 mA; (b) Iref 1 = 12.4 mA, Iref 2 = 8.9 mA; (c) Iref 1 = 10.2 mA, Iref 2 = 11.9 mA. C.1.2 Dynamic characterization The aim of this experiment was to perform a high resolution measurement of the laser response while it was switching between two working points. So, for each case the power was measured in function of time and wavelength. The result of this experiment was a 3-D wavelength-power-time graph. From such graph, various aspects of the laser response were calculated: time evolution of the optical spectrum, Side Mode Suppression Ratio (SMSR) and switching time.
  • 177. Appendix C. Static and dynamic wavelength characterization of tunable lasers 163 C.1.2.1 Dynamic characterization setup The test setup for characterizing the laser transient wavelength response is shown in figure C.5. A function generator excited each laser reflector with a switching square signal. After the tuneable laser, a Variable Optical Attenuator (VOA) was placed, in order to not saturate the photo-detector, while being inside the range of the filter. After the VOA, the optical filter filtered the laser output, centred at the wavelength specified by the PC and with a bandwidth of 0.08 nm (10 GHz), from 1532 nm until 1562 nm. The photo-detector output was connected to the oscilloscope and its data was acquired by the laptop PC by means of a GPIB interface. Figure C.5: Experimental setup for transient response characterization. C.1.2.2 Dynamic characterization results The experiment was performed for different kind of transitions. Figure C.6 (a) depicts the wavelength map for Iph = 2.4 mA; from which the working points were chosen to carry out the switching measurements. Most of the measured transitions are denoted in figure C.6 (a) with a white line and an identification number. An example of the currents applied to the reflectors in function of time is shown in figure C.6 (b). Each reflector was excited with a square signal running at 20 MHz, with a rise/fall times of less than 100 ps. Along the following lines the response of the MG-Y laser will be discussed for different kinds of wavelength switching. In the first experiment (depicted in figure C.6 (a-1)), the laser behaviour was measured when having a change between two working points located in the same mode (from 1560.24 nm to 1560.46 nm). Figure C.7 (1.a) illustrates a 3-D Wavelength-Power-Time plot (WPT plot). Such plot represents the wavelength versus time and power (gray scale). One can observe in this figure that the transition from one wavelength to the other within the same mode is totally continuous, and mainly due to thermal effect. The switching time of the laser during this kind of transient is quite long (3 ns).
  • 178. Appendix C. Static and dynamic wavelength characterization of tunable lasers 164 Figure C.6: (a) Stable regions map for Iph = 2.4 mA. The black points denote working points used to measure the transition between two modes. The white lines denote such transitions, and the number is used as experiment identifier. (b) Voltage versus time plot of the signals driving reflector sections for experiment 4 (see table C.1). The lines shown in figure C.6 (a-2) and figure C.6 (a-3) denote the next two experi- ments. The first transition was measured between two consecutive modes (from 1550.12 to 1550.92 nm), whereas the second was measured between two non-consecutive modes (from 1559.49 to 1561.93 nm). Please note that the laser modes corresponding to each test are located in the same super-mode. The WPT plots obtained from these experi- ments are depicted in figure C.7 (2.a) and (3.a), respectively. In both figures, the same effect shown in figure C.7 (1.a) can be observed: the wavelength shifts continuously when the working point changes inside the same mode. Afterwards, when the transient reaches the limit of the mode, an abrupt (non-continuous) switch of wavelength occurs. This behaviour is also related to the thermal effect [84]: the gain of the material shifts continuously towards higher wavelengths when increasing current (what produces the continuous shifting); however, when the gain curve approaches the following longitudinal mode, a longitudinal mode-hopping occurs and such mode becomes the main mode (what produces the non-continuous jump to higher wavelengths). As expected, the switching time increased with the distance between working points (see table C.1). In the next set of experiments, which working points can be seen in figure C.6 (a-4, 5, 6 and 7), the measured transitions were between different super-modes: experiment 4 consists in a transition between consecutive super-modes; in experiment 5 the transition measured was between two super-modes separated by one super-mode; the working points in experiment 6 are separated by 4 super-modes; and in experiment 7 the two working points measured gave neighboring wavelengths (see table C.1), but located several super- modes apart. WPT plots of experiments 4, 5, 6 and 7 are depicted in figure C.7 (4.a, 5.a, 6.a, 7.a), respectively. A common characteristic can be observed in all WPT plots: there is no continuous wavelength shift as in the previous set of experiments. This is due to an additive Vernier effect, typical from the MG-Y laser [85]. Thus, when the reflector currents are not increased/decreased along the same direction, the wavelength remain constant until it arrives to a mode limit. Then, the wavelength changes suddenly.
  • 179. Appendix C. Static and dynamic wavelength characterization of tunable lasers 165 Figure C.7: (Id.a) WPT plot: Plot of the wavelength versus time, and power (gray scale) versus both wavelength and time for experiment ’Id’ (see table C.1 and/or fig- ure C.6 (a)). (Id.b) SMSR versus time plot for experiment ’Id’ (see table C.1 and/or figure C.6 (a)). Please note that the device lases for a certain period of time with the wavelength that belongs to each super-mode that crosses during the transient. Such effect slows down the time response of lasers. Therefore, in a similar way than in the previous set of experiments, the switching time increases with the number of super-modes between both working points (see table C.1). Consequently, the fact of not having the continuous wavelength shift makes the transitions between consecutive super-modes much faster than the transition between consecutive modes (inside the same super-mode). Transitions between working points placed at different phase currents were also measured. The WPT plot of a transition between two working points located in the same mode and the WPT of a transition between points placed in different modes are depicted in
  • 180. Appendix C. Static and dynamic wavelength characterization of tunable lasers 166 figure C.7 (8.a, 9.a), respectively. As it has been said previously, the change in wavelength is much smaller when changing the phase current than when varying the reflector currents. Nevertheless, a new important fact was observed during these experiments: the switching time was substantially much slower when the phase current was changed (see table C.1). Thus, the frequency of the signal driving the laser had to be low, from 20 to 2.5 MHz, in order to better appreciate the transition. Each of the plots placed below the WPTs -see figure C.7 (Id.b)- represent the Side Mode Suppression Ratio (SMSR) calculated for each of the experiments as a function of time (please note that the SMSR was calculated within the complete C-band range). One can observe in the plots that the SMSR undergoes the 20 dB only during the transition between different modes. The mean SMSR for each case can be seen in table C.1. In order to calculate the SMSR, the power of the main and secondary modes had to be measured along the transitions. In figure C.8 (a) it is shown the WPT of experiment 4 in logarithmic scale to be able to visualize the secondary mode. One can observe that such mode is located at the ”non-lasing” working point of the transition. However, as shown with the SMSR plots, the power of the secondary mode is 20 dB lower than the main mode power (see also figure C.8 (b), which depicts the power of the main and secondary modes versus time). Figure C.8: (a) WPT plot: Plot of the wavelength versus time, and power (logarithmic colour scale) versus both wavelength and time for experiment 4. (b) Main mode and secondary mode power versus time (in logarithmic scale). As said before, during the experiment shown in figure C.8, the secondary mode was located always in the non-lasing working point of the transition. However, in much complex transients, the laser does not have such behaviour. In figure C.9 (a) shows a zoom of the experiment 5 WPT. Such experiment was a transition between working points located in two different super-modes (mode 1 at 1534.16 nm and mode 2 at 1545.2 nm) separated by one super-mode (inter-mode at 1539.8 nm). In order to appreciate more easily the behaviour of the device during the transition, the wavelength versus time of the main and secondary modes was plotted in figure C.9 (b) (black and green lines, respectively). In this figure, one can see that the secondary exhibits two kinds of mode- hopping: a mode-hopping between mode 1 and mode 2 when the device (main mode) is
  • 181. Appendix C. Static and dynamic wavelength characterization of tunable lasers 167 lasing in the inter-mode; and a mode-hopping between the inter-mode and mode 1/mode 2, when the main mode is located in mode 2/mode 1. In addition, a mode-hopping of the main mode is also observed in figure C.9 (c), which is a zoom of the transition between the inter-mode and mode 2 of figure C.9 (b). Please note that this mode-hopping has a short duration ( 100 ps). Figure C.9: (a) WPT plot zoom of experiment 5: Plot of the wavelength versus time, and power (logarithmic colour scale) versus both wavelength and time. (b) Wavelength versus time of the main and secondary modes of depicted in (a). (c) Zoom of (b) during the transition between inter-mode (1539.8 nm) and mode 2 (1545.2 nm). Experiment Kind of transition Wavelengths (nm) Switching SMSR Identifier time (ns) (dB) 1 InM 1560.08 - 1560.32 3.0 20.14 2 CM 1550.00 - 1550.80 4.0 23.63 3 NCM 1559.68 - 1562.08 7.0 22.28 4 CS 1550.00 - 1555.60 0.40 20.83 5 NCS (1 super-mode in between) 1534.16 - 1545.20 5.0 22.27 6 NCS (4 super-modes between) 1534.16 - 1559.68 12.0 23.85 7 NCS (close wavelengths) 1534.16 - 1535.80 13.0 22.14 8 Iph : InM 1554.78 - 1554.86 12.0 23.15 9 Iph : CM 1554.80 - 1554.84 40.0 23.9 Table C.1: The acronyms read in ”kind of transition” column, have a brief explanation of the working points location: InM (Inside the same Mode); CM (Consecutive Modes in the same super-mode); NCM (Non-Consecutive modes in the same super-mode); CS (Consecutive Super-modes); NCS (Non-Consecutive Super-modes); Iph (change in phase current). C.2 Conclusions The topology of the wavelength modes has been fully characterized. The 3-dimensional (Iref 1 , Iref 2 , Iph ) wavelength map built, is essential to locate the wavelength modes needed for certain applications (e.g. the channels for WDM). Moreover, a good knowledge of the wavelength map topology let us find easily the most stable operation points.
  • 182. Appendix C. Static and dynamic wavelength characterization of tunable lasers 168 In addition, through the high-resolution wavelength-power-time measurement, the dy- namic behaviour of the laser while switching between modes has been also characterized. The optical spectrum at every instant and its evolution along the tuning transient was obtained. With this, it was easy to identify, not only the wavelength temporal drift, but also the transitory mode hopping or interferences over other wavelength channels.
  • 183. Appendix D Phase noise digital modeling As seen in chapters 3 and 4, the main phase noise simulations were done in Mat- lab/Simulink. Here the objective is to obtain an expression in order to generate the phase of phase noise at several linewidths, because the Matlab/Simulink software has no modules to generate it. Also the phase of phase noise must be generated in a discrete domain (Matlab), while the simulation with Simulink has to be in the analog domain. By definition, the phase introduced by an oscillator’s phase noise can be modeled as: t φN (t) = ϕN (τ )dτ (D.1) 0 where ϕN (t) is a Gaussian process, with zero mean and variance 2π∆ν; and φN (t) is the phase noise (rad.). Furthermore, it is known that in the frequency domain, this phase will have a power spectral density defined by the following equation: ∆ν S(f ) = (D.2) 2πf 2 where ∆ν is the laser linewidth. From equation D.1, the integral can be made in a discrete time domain, using the rect- angles method: j φN [j] = ∆t · ϕN [i] (D.3) i=1 For every j from 1 until M . where ϕN is a vector containing random numbers with a Gaussian distribution and zero mean, ∆t is the time step, M is the number of noise samples generated, and φN is a vector containing the phase noise. 169
  • 184. Appendix D. Phase noise digital modeling 170 In order to see the transfer function, this expression can be rewritten and transformed, first into the z domain and secondly into the DFT domain: φN [i] = φN [i − 1] + γN [i] (D.4) so ΦN (z) H(z) = (D.5) ΓN (z) and ΦN (Ω) 1 H(z) = = (D.6) ΓN (Ω) 1 − e−jΩ where ΦN (z) is the z transform of φN . ΦN (Ω) is the discrete Fourier transform of φN . γN [i] is ∆t · ϕN [i]. ΓN (z) is the z transform of γN . ΓN (Ω) is the discrete Fourier transform of γN . Ω is the normalized frequency expressed in rad/s. Thus, as γN is also white noise, it is an order 1 Auto-Regressive (AR) process. Now, one can work without normalized frequencies and manipulate the expressions, resulting in: Gγ (f ) Gγ (f ) Sφ (f ) = −j2πf ∆t |2 = (D.7) |1 − e 2 1 − cos(2πf ∆t) Where Gγ (f ) is the power spectrum density of γN . So, when f decreases and takes values near 0, Sφ (f ) grows towards +∞, and becomes similar to equation D.2. Assuming that the integration step is very small (sampling frequency very high) and that the PSD expression of the AR process is evaluated near 0 (base band), the cosine term can be approximated by its Taylor expansion truncated after its second term. Then, equation D.7 can be written as: Gγ (f ) Sφ (f ) = (D.8) 4π 2 f 2 ∆t2 And this has the exact shape of the theoretical expression. Taking into account the sampling reviewed theorem [86], the last expression from DFT domain can be moved to analog Fourier domain and terms between D.2 and D.8 can be identified. In this way the phase noise can be approximated by an AR process if it accomplishes: Gγ (f ) = 2π∆ν∆t (D.9)
  • 185. Appendix D. Phase noise digital modeling 171 But Gγ (f ) is the PSD of γN ; while γN has an autocorrelation of: γN [i] = ∆tϕN [i] (D.10) rγ (l) = ∆t2 rϕ (l) = ∆t2 σϕ δ(l) 2 (D.11) being rg amma the autocorrelation of γN and rϕ the autocorrelation of ϕN . Thus, as Gγ (f ) is the DFT of rγ : Gγ (f ) = DF T {rγ } = ∆t2 · σϕ 2 (D.12) And such expression can be made the same as equation D.9: ∆t2 · σϕ = 2π∆ν∆t 2 (D.13) 2 2π · ∆ν σϕ = (D.14) ∆t And this is it. So, to make a phase approximation of phase noise in order to simulate the proposed systems (oPLLs and diversity schemes) with Matlab/Simulink software it is needed to: 1. Generate a white Gaussian noise with variance 2π∆f /∆t. 2. Check it is itself uncorrelated. 3. Integrate it using the rectangles method. 4. Estimate its PSD. 5. Check that this PSD fits the theoretical curve. Figure D.1 show the AR modified covariance method spectrum estimation of a noise gen- erated by our method (thin black line), scaled discrete theoretical spectrum for the same noise (square points with dotted line), and analog theoretical spectrum to be approxi- mated (dashed line). From this figure, it should be noted that below 10 GHz the AR curve is very close to the analog theoretical curve. So the phase of phase noise generated is valid at the working frequencies.
  • 186. Appendix D. Phase noise digital modeling 172 Figure D.1: Phase of phase noise spectrum.
  • 187. Appendix E Lock-In OPLL prototype scheme and printed circuit board Figure E.1: Printed circuit board outline of the Lock-IN OPLL prototype. 173
  • 188. Appendix E. Lock-In OPLL prototype scheme and printed circuit board 174
  • 189. Appendix E. Lock-In OPLL prototype scheme and printed circuit board 175
  • 190. Appendix F Research publications F.1 Patents 1. J. Prat, J. M. F`brega, ”Homodyne receiver for optical communications with post a processing,” app. number US-12521619, date 10/20/2009. 2. J. Prat, J. M. F`brega, ”Receptor homodino para comunicaciones ´pticas con a o procesado a posteriori,” app. number ES-P200700041 (PCT/ES2007/000778), date 12/29/2006. 3. J. Prat, J. M. F`brega, J. M. Gen´, ”Receptor coherente homodino para comuni- a e caciones opticas con demodulacion diferencial,” app. number ES-P200500998, date 04/21/2005. F.2 Book contributions 1. Contribution to the book ”Next Generation FTTH Passive Optical Networks,” Springer-Verlag 2008 (Josep Prat Ed.), in the transmission techniques chapter. F.3 Journal publications 1. J. M. F`brega, J. Prat, ”Experimental Investigation of Channel Crosstalk in a a Time-Switched Phase Diversity Optical Homodyne Receiver,” OSA Optics Letters, vol. 34, No. 4, February 2009. 2. J. M. F`brega, J. Prat, ”Homodyne receiver prototype with time-switching phase a diversity and feedforward analog processing,” OSA Optics Letters, vol. 32, No. 5, March 2007. 177
  • 191. Appendix F. Research publications 178 3. J. M. F`brega, J. Prat, ”Fuzzy Logic Data Estimation Based PSK Receiver with a Time-switched Phase Diversity,” IEE Electronics Letters, vol. 42, no. 16, August 2006. F.4 Conference publications 1. J. M. F`brega, J. Prat ”Digital Phase Estimation Method based on Karhunen-Lo`ve a e series expansion for Coherent Phase Diversity Detection,” Optical Fiber Commu- nication OFC/NFOEC 2010, paper JThA3, San Diego (CA), USA, March 2010. 2. J. M. F`brega, A. El Mardini, V. Polo, J. A. L´zaro, E. T. Lopez, R. Soila, Josep a a Prat ”Deployment Analysis of TDM/WDM Single Fiber PON with Colourless ONU Operating at 2.5 Gbps Subcarrier Multiplexed Downstream and 1.25 Gbps Up- stream,” National Fiber Optic Engineers Conference OFC/NFOEC 2010, paper NWB5, San Diego (CA), USA, March 2010. 3. J. M. F`brega, V. Polo, E. T. L´pez, J. A. L´zaro, J. Prat ”Optical Network a o a Based on Reflective Semiconductor Optical Amplifier Using Electrical Subcarrier Multiplex for Enhancing Bidirectional Transmission,” 6a Reunion espanola de Op- toelectronica OPTOEL’09, Malaga, July 2009. 4. J. M. F`brega, J. Prat, ”Ultra-Dense, Transparent and Resilient Ring-Tree Access a Network using Coupler-based Remote Nodes and Homodyne Transceivers,” Inter- national Conference on Transparent Optical Networks ICTON’09, Paper Th.B3.3, Azores, Portugal, June 2009. 5. J. M. F`brega, J. Prat, L. Molle, R. Freund, ”Design of a wavelength control for a coherent detection of high order modulation formats,” International Conference on Transparent Optical Networks ICTON’09, Paper We.P.21, Azores, Portugal, June 2009. 6. J. M. F`brega, J. Prat, ”Low Cost Homodyne Transceiver for UDWDM Access Net- a works,” European Conference on Networks and Optical Communications NOC’09, Invited Paper, Valladolid, Spain, June 2009. 7. J. M. F`brega, E. T. L´pez, J. A. L´zaro, M. Zuhdi, J. Prat, ”Demonstration of a a o a full duplex PON featuring 2.5 Gbps Sub Carrier Multiplexing downstream and 1.25 Gbps upstream with colourless ONU and simple optics,” European Conference on Optical Communications ECOC’08, Paper We.1.F.6, Brussels, Belgium, September 2008. 8. J. M. F`brega, L. Vilabru, J. Prat, ”Experimental Demonstration of Heterodyne a Phase-locked loop for Optical Homodyne PSK Receivers in PONs,” International Conference on Transparent Optical Networks ICTON’08, Paper We.C1.5, Athens, Greece, July 2008.
  • 192. Appendix F. Research publications 179 9. J. M. F`brega, J. Prat, ”Simple Low-Cost Homodyne Receiver,” European Confer- a ence on Optical Communications ECOC’07, Paper 7.2.5, Berlin, Germany, Septem- ber 2007. 10. J. Prat, J. A. L´zaro, J. M. F`brega, V. Polo, C. Bock, C. Arellano, M. Omella, a a ”Next Generation Architectures for Optical Access and Enabling Technologies,” 5a Reunion espanola de Optoelectronica OPTOEL’07, Bilbao, July 2007. 11. J. M. F`brega, J. Prat, ”Channel Crosstalk in ultra-dense WDM PON using Time- a Switched Phase Diversity Optical Homodyne Reception,” International Conference on Transparent Optical Networks ICTON’07, Paper Tu.A1.3, Rome, Italy, July 2007. 12. J. M. F`brega, J. Prat, ”Homodyne PSK Receiver with Electronic-Driven Phase a Diversity and Fuzzy Logic Data Estimation”, European Conference on Optical Com- munications ECOC’06, Paper We3.P.100, Cannes, France, September 2006. 13. C. Bock, J. M. F`brega, J. Prat, ”Ultra-Dense WDM PON based on Homodyne a Detection and Local Oscillator Reuse for Upstream Transmission”, European Con- ference on Optical Communications ECOC’06, Paper We3.P.168, Cannes, France, September 2006. 14. J. M. F`brega, J. Prat, ”Homodyne Receiver Implementation with Diversity Switch- a ing and Analogue Processing”, European Conference on Optical Communications ECOC’06, Paper We3.P.100, Cannes, France, September 2006. 15. J. M. F`brega, J. Prat, ”Optimization of Heterodyne Optical Phase-Locked Loops: a Loop Delay Impact and Transient Response Performances”, International Confer- ence on Telecommunications ICT’06, Paper Thu.O2, Funchal (Madeira), Portugal, May 2006. 16. J. M. F`brega, J. Prat, ”New Intradyne Receiver with Electronic-Driven Phase a and Polarization Diversity”, Optical Fiber Communication OFC/NFOEC’06, paper JThB45, Anaheim (CA), USA, March 2006. 17. J. Prat, J. M. F`brega, ”New Homodyne Receiver with Electronic I&Q Differential a Demodulation”,European Conference on Optical Communications ECOC’05, paper We4.P.104, Glasgow, UK, September 2005.
  • 193. Appendix F. Research publications 180 F.5 Submitted publications F.5.1 Book contributions • J. A. L´zaro, J. M. F`brega, A. L. J. Teixeira and J. Prat, ”Laboratory Set-ups a a and Methods for Teaching and Characterizing Optical Polarization Modulators,” in EuroFOS project Handbook of Experimental Fiber Optics. F.5.2 Journal publications • J. M. F`brega, A. ElMardini, V. Polo, E. T. L´pez, J. A. L´zaro, M. Zuhdi, and J. a o a Prat, ”Demonstration of a single fibre PON featuring 2.5 Gbps Sub-Carrier Mul- tiplexing Downstream and 1.25 Gbps Upstream with Colourless ONU,” submitted to IET proceedings in optoelectronics. F.5.3 Conference publications • M. Mestre, J. M. F`brega, J. A. L´zaro, V. Polo, A. Djupsj¨backa, M. Forzati, P.-J. a a o Rigole and J. Prat, ”Tuning Characteristics and Switching Speed of a Modulated Grating Y Structure Laser for Wavelength Routed PONs,” submitted to OSA-ANIC 2010.
  • 194. Bibliography [1] C. K. Kao and G. A. Hockham. Dielectric-fibre surface waveguides for optical fre- quencies. Proceedings of the IEE, 13(7):1151, June 1966. [2] A. Girard. FTTx PON Technology and Testing. EXFO Electrical Eng., 2005. [3] L. G. Kazovsky, G. Kalogerakis, and W. T. Shaw. Homodyne phase-shift-keying sys- tems: Past challenges and future opportunities. IEEE / OSA Journal of Lightwave Technology, 24(12), 2006. [4] S. Narikawa, N. Sakuray, K. Kumokazi, and T. Imai. Coherent WDM-PON based on heterodyne detection with digital signal processing for simple ONU structure. In Proceedings of 32th European Conference on Optical Communications (ECOC 2006), Cannes (France), September 2006. [5] H. Rohde, S. Smolorz, K. Kloppe, and E. Gottwald. Next generation optical ac- cess: 1 Gbit/s for everyone. In Proceedings of 35th European Conference on Optical Communications (ECOC 2009), Vienna (Austria), September 2009. [6] F. D. Natali. AFC tracking algorithms. IEEE Transactions on Communications, 32(8), 1984. [7] John M. Senior. Optical fiber communications. Principles and practice. Prentice Hall, 1985. [8] S. Betti, G. De Marchis, and E. Iannone. Coherent Optical Communications Systems. J. Wiley, 1995. [9] J. G. Proakis. Digital Communications. McGraw-Hill, 2001. [10] K.-P. Ho. Phase-Modulated Optical Communication Systems. Springer-Verlag, 2005. [11] T. Mizuochi, Y. Miyata, T. Kobayashi, K. Ouchi, K. Kuno, K. Kubo, K. Shimizu, H. Tagami, H. Yoshida, H. Fujita, M. Akita, and K. Motoshima. Forward error correction based on block turbo code with 3-bit soft decision for 10-Gb/s optical communication systems. IEEE Journal of Selected Topics in Quantum Electronics, 10(2):376–386, February 2004. 181
  • 195. Bibliography 182 [12] W. C. Lindsey and M. K. Simon. Telecommunication Systems Engineering. Dover, 1991. [13] K. Kikuchi, T. Okoshi, M. Nagamatsu, and N. Henmi. Degradation of bit-error rate in coherent optical communications due to spectral spread of the transmitter and the local oscillator. IEEE / OSA Journal of Lightwave Technology, 2(6), December 1984. [14] A. Papoulis and S. U. Pillai. Probability, Random Variables and Stochastic Processes. McGraw-Hill, 2002. [15] F. M. Gardner. Phaselock Techniques. J. Wiley, 2005. [16] S. Norimatsu and K. Iwashita. Damping factor influence on linewidth requirements for optical PSK coherent detection systems. IEEE / OSA Journal of Lightwave Technology, 11(7), July 1993. [17] S. Norimatsu and K. Iwashita. PLL propagation delay-time influence on linewidth requirements of optical PSK homodyne detection. IEEE / OSA Journal of Lightwave Technology, 9(10), October 1991. [18] S. Norimatsu, K. Iwashita, and K. Sato. PSK optical homodyne detection using external cavity laser diodes in costas loop. IEEE Photonics Technology Letters, 2(5), 1990. [19] L. G. Kazovsky. Decision-driven phase-locked loop for optical homodyne receivers: performance analysis and laser linewidth requirements. IEEE / OSA Journal of Lightwave Technology, 3:1238, 1985. [20] M. Seimetz and C.-M. Weinert. Options, feasibility, and availability of 2x4 90 hybrids for coherent optical systems. IEEE / OSA Journal of Lightwave Technology, 24(3), 2006. [21] L. G. Kazovsky. Balanced phase-locked loops for optical homodyne receivers: per- formance analysis, design considerations, and laser linewidth requirements. IEEE / OSA Journal of Lightwave Technology, 4:182, 1986. [22] C.-C. Huang and L. Wang. Optimum deviations between mark and space-state bits in long-haul communication systems with balanced PLL receivers. IEEE / OSA Journal of Lightwave Technology, 13(10), 1995. [23] S. Camatel, V. Ferrero, and P. Poggiolini. 2-PSK homodyne receiver based on a decision driven architecture and a sub-carrier optical PLL. In Proceedings of the Conference on Optical Fiber Communication and the National Fiber Optic Engineers Conference, 2006 (OFC/NFOEC 2006), Anaheim (CA), March 2006.
  • 196. Bibliography 183 [24] Y. H. Cheng, T. Okoshi, and O. Ishida. Performance analysis and experiment of a homodyne receiver insensitive to both polarization and phase fluctuations. IEEE / OSA Journal of Lightwave Technology, 7:368–374, 1989. [25] M. Seimetz, M. Noelle, and E. Patzak. Optical systems with high-order DPSK and star QAM modulation based on interferometric direct detection. IEEE / OSA Journal of Lightwave Technology, 25(6), 2007. [26] M. G. Taylor. Phase estimation methods for optical coherent detection using digital signal processing. IEEE / OSA Journal of Lightwave Technology, 27:901–913, 2009. [27] M. G. Taylor. Accurate digital phase estimation process for coherent detection using a parallel digital processor. In Proceedings of 31st European Conference on Optical Communications (ECOC 2005), Glasgow (Scotland), September 2005. [28] R. Noe. Phase noise-tolerant synchronous QPSK/BPSK baseband-type intradyne receiver concept with feedforward carrier recovery. IEEE / OSA Journal of Lightwave Technology, 23(2), 2005. [29] T. Pfau, S. Hoffmann, R. Peveling, S. Bandhare, S. K. Ibrahim, O. Adamczyk, M. Porrmann, R. Noe, and Y. Achiam. First real-time data recovery for synchronous QPSK transmission with standard DFB lasers. IEEE Photonics Technology Letters, 18(18), 2006. [30] D. Van den Borne, C. R. S. Fludger, T. Duthel, T. Wuth, E. D. Schmidt, C. Schulien, E. Gottwald, G. D. Khoe, and H. de Waardt. Carrier phase estimation for coherent equalization of 43-Gb/s POLMUXNRZ-DQPSK transmission with 10.7-Gb/s NRZ neighbours. In Proceedings of 33th European Conference on Optical Communications (ECOC 2007), Berlin (Germany), September 2007. [31] M. Seimetz. Laser linewidth limitations for optical systems with high-order modu- lation employing feed forward digital carrier phase estimation. In Proceedings of the Conference on Optical Fiber Communication and the National Fiber Optic Engineers Conference, 2008 (OFC/NFOEC 2008), San Diego (CA), March 2008. [32] J. M. Fabrega and J. Prat. Optimization of heterodyne optical phase-locked loops in homodyne PSK detection. In Proceedings of International Conference on Telecom- munications ICT’06, Funchal (Madeira, Portugal), April 2006. [33] J. M. Fabrega, L. Vilabru, and J. Prat. Experimental demonstration of heterodyne phase-locked loop for optical homodyne PSK receivers in PONs. In Proceedings of International Conference on Transparent Optical Networks ICTON’08, Athens (Greece), July 2008. [34] R.W. Smith. Design of a phase-sensitive detector to maximize signal-to-noise ratio in the presence of gaussian wideband noise. Measurement Science Technology, 3:1052– 1062, August 1992.
  • 197. Bibliography 184 [35] H. Meyr and G. Ascheid. Synchronization in digital communications. John Wiley and Sons, 1990. [36] A. J. Viterbi. Principles of coherent communications. McGraw-Hill, 1966. [37] C.J. Wolesja. Effects of oscillator phase noise on PSK modulation. Comsat technical review, 6(1), March 1976. [38] D. Derickson. Fiber Optic Test and Measurement. Prentice Hall, 1998. [39] L. G. Kazovsky. A 1320- nm experimental optical phase-locked loop : Performance investigation and PSK homodyne experiments at 140Mb/s and 2Gb/s. IEEE / OSA Journal of Lightwave Technology, 8(9), 1990. [40] J. M. Fabrega and J. Prat. Digital phase estimation method based on karhunen-lo`ve e series expansion for coherent phase diversity detection. In Proceedings of Optical Fiber Communication OFC 2010, San Diego (CA), March 2010. [41] J. M. Fabrega and J. Prat. Homodyne receiver prototype with time-switching phase diversity and feedforward analog processing. OSA Optics Letters, 32(5), 2007. [42] J. M. Fabrega and J. Prat. Experimental investigation of channel crosstalk in a time- switched phase diversity optical homodyne receiver. OSA Optics Letters, 34(4), 2009. [43] J. Prat and J. M. Fabrega. Receptor homodino para comunicaciones opticas con ´ procesado a posteriori, 2006. [44] J. Prat and J. M. Fabrega. Homodyne receiver for optical communications with post processing, 2009. [45] J. M. Fabrega and J. Prat. Fuzzy logic data estimation based PSK receiver with time-switched phase diversity. IET Electronics Letters, 42(16), 2006. [46] J. M. Fabrega and J. Prat. Simple low-cost homodyne receiver. In Proceedings of 33th European Conference on Optical Communications (ECOC 2007), Berlin (Germany), September 2007. [47] J. M. Fabrega and J. Prat. New intradyne receiver with electronic-driven phase and polarization diversity. In Proceedings of the Conference on Optical Fiber Commu- nication and the National Fiber Optic Engineers Conference, 2006 (OFC/NFOEC 2006), Anaheim (CA), March 2006. [48] I.M.I Habbab, J. M. Khan, and L. J. Greenstein. Phase-insensitive zero-IF coherent optical system using phase switching. IET Electronics Letters, 24:975–976, 1988. [49] G. Jacobsen and I. Garret. Crosstalk and phase noise effects in multichannel co- herent CP-FSK systems with tight IF filtering. IEEE / OSA Journal of Lightwave Technology, 9(6), June 1991.
  • 198. Bibliography 185 [50] R. O. Duda, P. E. Hart, and D. G. Stork. Pattern Classification. J. Wiley, 1998. [51] G. J. Klir and B. Yuan. Fuzzy sets and fuzzy logic. Theory and applications. Prentice Hall, 1995. [52] L. G. Kazovsky. Phase- and polarization-diversity coherent optical techniques. IEEE / OSA Journal of Lightwave Technology, 7:279–292, 1989. [53] N. Keil, H. Yao, C. Zawadzki, W. Doldissen, W. Schlaak, M. Mohrle, and D. Schmidt. Polymer as integration platform for low-cost devices in future optical networks. In Proceedings NOC06, Berlin (Germany), July 2006. [54] A. Rogner and H. Pannhoff. Mass fabrication of passive polymer multimode and single-mode waveguide devices. In Proceedings of the Conference on Optical Fiber Communication, (OFC), Dallas(TX), March 1994. [55] M. Seimetz, L. Molle, M. Gruner, D.-D. Gross, and R. Freund. Transmission reach attainable for single-polarization and polmux coherent star 16QAM systems in com- parison to 8PSK and QPSK at 10Gbaud. In Proceedings of the Conference on Opti- cal Fiber Communication and the National Fiber Optic Engineers Conference, 2009 (OFC/NFOEC 2009), San Diego (CA), March 2009. [56] R. Noe, H. Rodler, A. Ebberg, G. Gaukel, B. Noll, J. Wittmann, and F. Auracher. Comparison of polarization handling methods in coherent optical systems. IEEE / OSA Journal of Lightwave Technology, 9(10), August 1991. [57] P. Poggiolini and S. Benedetto. Theory of polarization spreading techniques - part I. IEEE Transactions on Communications, 42(3), 1994. [58] S. Benedetto and P. Poggiolini. Theory of polarization spreading techniques - part II. IEEE Transactions on Communications, 42(6), 1994. [59] M. O. van Deventer. Fundamentals of Bidirectional Transmission over a Single Optical Fibre. Kluwer academic press, 1996. [60] M. Omella, J. Lazaro, and J. Prat. Driving requirements for wavelength shifting in colorless ONU with dual-arm modulator. IEEE / OSA Journal of Lightwave Technology, 27(17), 2009. [61] R. S. Vodhanel. 5 Gbit/s direct optical DPSK modulation of a 1530-nm DFB laser. IEEE Photonics Technology Letters, 1(8), 1989. [62] A. Ramaswamy, L. A. Johansson, J. Klamkin, C. Sheldon, H. F. Chou, M. J. Rodwel, L. A. Coldren, and J. E. Bowers. Coherent receiver based on a broadband phase-lock loop. In Proceedings of the Conference on Optical Fiber Communication and the National Fiber Optic Engineers Conference, 2007 (OFC/NFOEC 2007), Anaheim (CA), March 2007.
  • 199. Bibliography 186 [63] A. Leven, N. Kaneda, and K. Ut-Va Koc. Frequency estimation in intradyne recep- tion. IEEE Photonics Technology Letters, 19(6), March 2007. [64] L. Li, Z. Tao, S. Oda, T. Hoshida, and J. C. Rasmussen. Wide-range, accurate and simple digital frequency offset compensator for optical coherent receivers. In Proceedings of the Conference on Optical Fiber Communication and the National Fiber Optic Engineers Conference, 2008 (OFC/NFOEC 2008), San Diego (CA), March 2008. [65] H. G. Krimmel, T. Pfeiffer, B. Deppisch, and L. Jentsch. Hybrid electro-optical feed- back gain-stabilized EDFAs for long-reach wavelength-multiplexed passive optical networks. In Proceedings of 35th European Conference on Optical Communications (ECOC 2009), Vienna (Austria), September 2009. [66] J. A Lazaro, R. I. Martinez, V. Polo, C. Arellano, and J. Prat. Hybrid dual-fiber-ring with single-fiber-trees dense access network architecture using RSOA-ONU. In Pro- ceedings of the Conference on Optical Fiber Communication and the National Fiber Optic Engineers Conference, 2007 (OFC/NFOEC 2007), Anaheim (CA), March 2007. [67] C. Bock, J. M. Fabrega, and J. Prat. Ultra-dense WDM PON based on homodyne detection and local oscillator reuse for upstream transmission. In Proceedings of 32th European Conference on Optical Communications (ECOC 2006), Cannes (France), September 2006. [68] J. M. Fabrega and J. Prat. Ultra-dense, transparent and resilient ring-tree access network using coupler-based remote nodes and homodyne transceivers. In Proceed- ings of International Conference on Transparent Optical Networks ICTON’09, Ponta Delgada (Azores, Portugal), July 2009. [69] IEC. Criteria on safety for optical fiber communication systems, 2008. [70] W. Hung, C.-K. Chan, L.-K. Chen, and C. Lin. System characterization of a robust re-modulation scheme with DPSK downstream traffic in a WDM access network. In Proceedings of 29th European Conference on Optical Communications (ECOC 2003), Rimini (Italy), September 2003. [71] M. Schneider, T. Reimann, A. Stohr, S. Neumann, F.-J. Tegude, , and D. Jager. Monolithically integrated optoelectronic circuits using HBT, EAM, and TEAT. In Proceedings of international topical meeting on MicroWave Photonics, MWP 2002, Awaji (Japan), 2002. [72] T. Koonen, K. Steenbergen, F. Janssen, and J. Wellen. Flexibly reconfigurable fibre- wireless network using wavelength routing techniques: The ACTS project AC349 PRISMA. Photonics Network Communications, 3:297–306, July 2001.
  • 200. Bibliography 187 [73] H. Takesue and T. Sugie. Data rewrite of wavelength channel using saturated SOA modulator for WDM metro/access networks with centralized light sources. In Pro- ceedings of 28th European Conference on Optical Communications (ECOC 2002), Copenhagen (Denmark), September 2002. [74] J. Prat, C. Arellano, V. Polo, and C. Bock. Optical network unit based on a bidi- rectional reflective semiconductor optical amplifier for fibre-to-the-home networks. IEEE Photonics Technology Letters, 17(1), 2005. [75] C. Arellano, C. Bock, J. Prat, and K. D. Langer. RSOA-based optical network units for WDM-PON. In Proceedings of the Conference on Optical Fiber Communication and the National Fiber Optic Engineers Conference, 2006 (OFC/NFOEC 2006), Anaheim (CA), March 2006. [76] K. Y. Cho, H. Murakami, Y. J. Lee, A. Agata, Y. Takushima, and Y. C. Chung. Demonstration of RSOA-based WDM PON operating at symmetric rate of 1.25 Gb/s with high reflection tolerance. In Proceedings of the Conference on Optical Fiber Communication and the National Fiber Optic Engineers Conference, 2008 (OFC/N- FOEC 2008), San Diego (CA), March 2008. [77] H. Al-Raweshidy and S. Komaki. Radio over Fibre Technologies for Mobile Commu- nication Network. Artech House, 2002. [78] G. P. Agrawal. Fibre-Optic Communication Systems. J. Wiley, 2002. [79] C. J. Clark, A. A. Moulthrop, M. S. Muha, and C. P. Silva. Transmission re- sponse measurements of frequency-translating devices using a vector network ana- lyzer. IEEE Transactions on Microwave Theory and Techniques, 44(12):2724–2737, December 1996. [80] R. Noe, W. B. Sessa, R. Welter, and L. G. Kazovsky. New FSK phase-diversity re- ceiver in a 150 Mbit/s coherent optical transmission system. IET Electronics Letters, 24(9), April 1988. [81] V. Ferrero and S. Camatel. Optical phase locking techniques: an overview and a novel method based on single side sub-carrier modulation. OSA Optics Express, 16(2), January 2008. [82] V. Polo. Contribution to the dynamic tuning of tunable laser sources for modulation and optical packet routing in WDM and access networks. PhD thesis, Universitat Politecnica de Catalunya, 2006. [83] R. Freund, D.-D. Gross, M. Seimetz, L. Molle, and C. Caspar. 30 Gbit/s RZ-8- PSK transmission over 2800 km standard single mode fibre without inline dispersion compensation. In Proceedings of the Conference on Optical Fiber Communication and the National Fiber Optic Engineers Conference, 2008 (OFC/NFOEC 2008), San Diego (CA), March 2008.
  • 201. Bibliography 188 [84] Jens Buus, Markus-Christian Amann, and Daniel J. Blumenthal. Tuneable Laser Diodes and Related Optical Sources. J. Wiley, 2005. [85] J.-O. Wesstrom, G. Sarlet, S. Hammerfeldt, L. Lundqvist, P. Szabo, and P.-J. Rigole. State-of-the-art performance of widely tunable modulated grating Y-branch lasers. In Proceedings of the Conference on Optical Fiber Communication and the National Fiber Optic Engineers Conference, 2004 (OFC/NFOEC 2004), Anaheim (CA), March 2004. [86] J. G. Proakis and D. Manolakis. Digital Signal Processing. Prentice Hall, 1996.

×