University of Southern California Dissertations and Theses

Applications of all optical signal processing for advanced optical modulation formats

(USC Thesis Other)

Applications of all optical signal processing for advanced optical modulation formats

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content APPLICATIONS OF ALL OPTICAL SIGNAL PROCESSING FOR
ADVANCED OPTICAL MODULATION FORMATS

by

Scott R. Nuccio

_________________________________________________

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ELECTRICAL ENGINEERING)

May 2011

Copyright 2011 Scott R. Nuccio

ii

Dedication

To my parents
Robert and Jill Nuccio
and my entire family and friends
for their everlasting love, support and understanding.

iii

Acknowledgements

I would like to thank my academic advisor and dissertation committee chairman,
Dr. Alan E. Willner, for his guidance and mentorship throughout my graduate
studies. I would also like to extend my great appreciation to Professor William Steier
and Professor Andrea Armani for serving on my dissertation and qualifying exam
committees. I would also like to thank Professor John O’Brien and Professor
Gerhard Kramer for their support and guidance during my qualifying examination.
I would like to extend my warmest thanks to the many members of the Optical
Communications Lab (OCLab) for many years of insightful discussions and
collaboration. In particular, Dr. Louis Christen, Dr. Yannick Lizé, Omer Faruk
Yilmaz, Irfan Fazal and Xiaoxia Wu were all instrumental to the completion of the
experimental results presented in this dissertation. Thanks team! I enjoyed all the
days and sleepless nights we spent in the lab. I appreciate all the discussions over the
years and wish you all the best in your careers.
Finally, I would like to extend my greatest thanks to the Aerospace Corporation
for their generous support of my graduate studies.

iv

Table of Contents
Dedication ii
Acknowledgements iii
Table of Figures vi
Abstract xiv
Chapter 1: Introduction 1
Chapter 2: Advanced Modulation Formats and Format Transparent Optical
Signal Processing 5
2.1 Motivation for Advanced Modulation Formats 5
2.2 Differential Phase Shift Keying (DPSK) 6
2.3 Differential Quadrature Phase Shift Keying (DQPSK) 15
2.4 Polarization Multiplexing (Pol-MUX) 18
2.5 Coherent Transmission and Reception 21
2.6 Raman Amplification 31
2.7 Wavelength Conversion Using PPLN Waveguides 37
2.8 Four-Wave-Mixing In Highly Nonlinear Fiber 43
Chapter 3: All Optical Multiplexing and Demultiplexing of 100 Gbit/s Pol-
MUX Signals 46
3.1 Introduction 46
3.2 Concept 48
3.3 Experimental Setup 48
3.4 Results and Discussion 52
Chapter 4: λ-Conversion of 160-Gbit/s PDM 16-QAM Using a Single
Periodically-Poled Lithium Niobate Waveguide 55
4.1 Introduction 55
4.2 Concept 57
4.3 Experimental Setup 59
4.4 Results and Discussion 60
Chapter 5: 503 ns Continuously Tunable Delay of 40 Gbit/s OOK and DPSK
with Improved Dispersion Compensation 63

v

5.1 Introduction 63
5.2 Concept 66
5.3 Experimental Setup 68
5.4 Results and Discussion 70
Chapter 6: 1.16 μs Continuously Tunable Delay of 100 Gbit/s DQPSK 73
6.1 Introduction 73
6.2 Experimental Setup 75
6.3 Results and Discussion 77
Chapter 7: Higher-Order Dispersion Compensation to Enable a 3.6-μs
Wavelength-Maintaining Delay of a 100-Gbit/s DQPSK Signal 79
7.1 Introduction 79
7.2 Concept 81
7.3 Experimental Setup 82
7.4 Results and Discussion 85
Chapter 8: Delay Extension to 5-μs for a 10 Gbit/s RZ-DPSK Signal 89
8.1 Experimental Setup 89
8.2 Results and Discussion 92
Chapter 9: Fine Tuning of Optical Delays Using Cascaded Acousto-Optic
Frequency Shifters 95
9.1 Introduction 95
9.2 Concept 97
9.3 Experimental Setup 101
9.4 Results and Discussion 103
Chapter 10: Continuously Tunable All Optical Buffer Using Conversion
Dispersion Based Delay 105
10.1 Introduction 105
10.2 Concept 108
10.3 Experimental Setup 112
10.4 Results and Discussion 117
10.5 Conclusion 130
Conclusion 131
References 134

vi

Table of Figures
Figure 2-1: (a) Illustration of DPSK transmission (b) DPSK constellation 7
Figure 2-2: Theoretical bit error rate curves for coherent and differential
detection versions of binary and quaternary phase modulation
formats. 8
Figure 2-3: Mach-Zehnder modulator configuration for DPSK modulation
[37]. 11
Figure 2-4: Mach-Zehnder modulator configuration for RZ pulse carving
[37]. 12
Figure 2-5: Illustration of a DPSK receiver including delay-line
interferometer for differential-phase to intensity conversion and
balanced detection. 13
Figure 2-6: Frequency domain response of 1-bit interferometer. The input
signal is bandpass filtered in the constructive port and notch
filtered in the destructive port. Spectra are plotted with 10
dB/division and 50 GHz/division. 14
Figure 2-7: Parallel modulator for generation of optical DQPSK, along with
ideal resulting constellation diagram. 16
Figure 2-8: Illustration of a typical DQPSK receiver, along with the
transmission response of each DLI output port. Four ports are
staggered by ¼ of the symbol rate. Constructive and destructive
port of each interferometer staggered by ½ the symbol rate. 17
Figure 2-9: Potential formats for the IEEE 802.3ba 100 Gbit/s Ethernet
Standard include quadrature-PSK, utilizing 4 phase states for 2
bits/symbol and polarization multiplexing (Pol-MUX) to
achieve 4 bits/symbol. 19
Figure 2-10: Coherent transmission system (a) implementation, (b) system
model. 24
Figure 2-11: Single-polarization downconverter employing a (a) heterodyne
and (b) homodyne design. 25

vii

Figure 2-12: Raman gain of fused quartz plotted as a function of frequency
shift from an exciting line at 526 nm. The experimental point is
the gain measured in the amplifier and the error bar represents a
combination of the uncertainties both in the measurement of the
gain and the spontaneous cross section [85]. 32
Figure 2-13: Level diagrams showing (a) stimulated Raman Stokes
scattering; (b) stimulated Raman anti-Stokes scattering; (c)
coherent anti-Stokes four-wave mixing; (d) multiple Stokes and
anti-Stokes scattering; and (e) hyper-Raman scattering [3]. 35
Figure 2-14: Illustration of quasi-phase matching in a periodically poled
Lithium Niobate waveguide. 39
Figure 2-15: Illustration of difference frequency generation in a periodically
poled Lithium Niobate waveguide. 40
Figure 2-16: Illustration of cascaded wavelength conversion schemes in a
PPLN waveguide. Second harmonic generation is followed by
difference frequency generation 41
Figure 2-17: Illustration of cascaded wavelength conversion in a PPLN
waveguide. Sum frequency generation is followed by difference
frequency generation. 42
Figure 2-18: Illustration of four-wave-mixing processes that satisfy the
phase match condition. 45
Figure 3-1: Conceptual diagram of all optical polarization demultiplexing
and polarization multiplexing. 48
Figure 3-2: Experimental setup. A 100 Gbit/s Pol-MUX signal is generated
and combined with two orthogonal pumps for demultiplexing to
two 50 Gbit/s WDM channels. Similarly, two 50 Gbit/s WDM
channels are generated and combined with two orthogonal
pumps for multiplexing into a single 100 Gbit/s Pol-MUX
channel. 50

viii

Figure 3-3: (a) Experimental spectra for demultiplexing (top) and
multiplexing (bottom) between 100 Gbit/s Pol- MUX RZ-OOK
and 2 x 50 Gbit/s RZ-OOK. (b) Experimental spectra for the
demultiplexing (top) and demultiplexing (bottom) when using
RZ-DPSK. (c) Back-to-Back eyes and
demultiplexed/multiplexed eyes for comparison. 52
Figure 3-4: (a) Comparison of RZ-OOK performance after demultiplexing
and multiplexing compared to back-to-back performance. (b)
Comparison of RZ-DPSK performance after demultiplexing and
multiplexing compared to back-to-back. (c) Performance of the
multiplexed 100-Gbit/s Pol-MUX channel after 1km of
uncompensated propagation. 53
Figure 4-1: Conceptual diagram of transparent polarization (a) and phase
and amplitude (b) conversion in a PPLN waveguide. 58
Figure 4-2: Experimental setup. A 20-Gbaud PDM 16-QAM or 40-Gbaud
16-QAM signal is generated and combined with a CW pump in
a bidirectional PPLN based wavelength converter. An EAM is
used for 40-Gbaud to 20-Gbaud down sampling followed by
coherent detection. 59
Figure 4-3: (a) Experimental constellation diagrams for 40-Gbaud single-
polarization 16-QAM back-to-back (top) and after (bottom)
conversion for pump powers of 16.2 dBm (left) and 21 dBm
(right). (b) Experimental constellation diagrams for 20-Gbaud
PDM 16-QAM back-to-back (top) and after (bottom) conversion
for both polarizations. (c) Experimental spectra for 40-Gbaud
single polarization (right) and 20 G-Baud PDM (left) 16-QAM. 61
Figure 4-4: Bit-error-rate (BER) measurements for (a) 40-Gbaud 16-QAM
and (b) 20-Gbaud PDM 16-QAM back-to-back and after
wavelength conversion. (c) Relative power penalty vs. pump
power for 40-Gbaud 16-QAM. 62
Figure 5-1: Conceptual diagram of tunable delay methods. As opposed to
previous methods (a) and (b), following wavelength conversion
(W/C) and delay in dispersion compensating fiber (DCF), the
signal is not returned to the original wavelength for
compensation (c). 66

ix

Figure 5-2: Experimental setup. A 40-Gbit/s signal is wavelength converted
and passed through the Raman pumped DCF. The signal is then
phase conjugated and shifted by ~3.4 nm before passing back
through the DCF for detection. 68
Figure 5-3: Measured fiber dispersion profile for (a) dispersion, (b)
compensation and (c) comparison of residual dispersion for
compensation at a fixed wavelength and using the newly
proposed method. Residual dispersion is reduced by >95%. 70
Figure 5-4: Measured delay versus converted wavelength and experimental
spectra of both the wavelength conversion and phase
conjugation stages. 71
Figure 5-5: Bit-error-rate (BER) curves for (a) 40-Gbit/s RZ-OOK and (b)
RZ-DPSK. Performance after back-to-back, the first wavelength
conversion (Stage 1), and after the full system (Final) is
compared. 40-Gbit/s RZ-OOK (c) and RZ-DPSK (d) back-to-
back performance compared to the minimum, maximum, and
middle delay performances. 72
Figure 6-1: Block diagram. Dispersion compensating fiber (DCF), fiber
Bragg grating (FBG), bandpass filter (BPF), transmitter (TX),
receiver (RX), and highly nonlinear fiber (HNLF). 74
Figure 6-2: (a) Measured delay of 1.16 μs. (b) Received 50 Gbit/s ODB
signal for 10 pm changes in laser wavelength showing ~275 ps
changes in delay. (c) Experimental spectra of first and second
wavelength conversion stages for the maximum delay value. 76
Figure 6-3: (a) 0, 0.5ps, and 1ps delay resolution of a single 40Gbit/s RZ-
OOK bit. (b) RF-spectra showing optical mixing for different
AOM frequency offsets. (c) Bit-error-rate curves for varying
delay values with and without the AOMs. 77
Figure 7-1: (a) Conceptual diagram of pre-dispersion block to enable 100
Gbit/s operation. (b) A 96% reduction in residual 3
rd
-order
dispersion is achieved using fixed fiber-Bragg-gratings (FBGs)
and a tunable spatial light modulator (SLM). 81

x

Figure 7-2: Block diagram. Dispersion compensating fiber (DCF), spatial
light modulator (SLM), band-pass filter (BPF), erbium-doped
fiber amplifier (EDFA), receiver (Rx), and highly nonlinear
fiber (HNLF). 82
Figure 7-3: (a) Measured delay of 3.6 μs for 100-Gbit/s RZ-DQPSK. (b) 7-
Gbit/s packets used to illustrate the full delay tuning range. 85
Figure 7-4: (a) Constellation diagrams showing the DPSK (Top) and
DQPSK (Bottom) signals before (Left) and after (Right) at the
middle delay value, ~1567 nm. (b) Experimental spectra of the
first stage (top) and third stage (bottom) at the minimum delay
value. 86
Figure 7-5: Measured bit-error-rate performance of (a) 100 (Solid), 80
(Dashed), and 20-Gbit/s (Solid) DQPSK and (b) 50 (Solid), 40
(Dashed), and 10-Gbit/s (Solid) DPSK for the minimum (Red),
middle (Blue), and maximum (Green) delay values. 87
Figure 7-6: Power penalty as a function of residual 3
rd
-order dispersion
(ps/nm2) for 100-Gbit/s RZ-DQPSK. 88
Figure 8-1: Improved experimental setup utilizes -48 ns/nm of DCF to
achieve a 5.4 μs relative delay. 89
Figure 8-2: Measured residual dispersion slope after the addition of the extra
DCF. At the minimum wavelength, this corresponds to ~0.8 dB
penalty for a 10 Gbit/s RZ-DPSK signal. 90
Figure 8-3: Experimental measurement of the delay range was performed
using a 300 Mbit/s packet. The full 5.4 μs range is shown (800
ns/Div). 92
Figure 8-4: (a) Experimental spectra for the first and third wavelength
conversion stages. Wavelength maintaining operation is
accomplished. (b) Measured relative delay vs. the converted
wavelength showing the full 5.4 μs delay range. 93
Figure 8-5: Bit-error-rate measurements for the minimum, middle, and
maximum delay values compared to the back-to-back case. 94

xi

Figure 9-1: (a) A tunable laser with 1pm (125MHz) resolution is used to
coarse tune the delay from 0 to 256ns. Cascaded acousto-optic
modulators (AOMs) shift the laser center frequency with 1kHz
resolution; fine tuning the delay from 0 to 25ps. (b) Measured
fine and coarse tuning ranges of our system. 98
Figure 9-2: Acousto-optic frequency shifters for up shifting and down
shifting a CW laser. 100
Figure 9-3: Block diagram. Wavelength Converter (W/C1), dispersion
compensating fiber (DCF), acousto-optic modulator (AOM),
Mach-Zehnder modulator (MZM), receiver (RX), and highly
nonlinear fiber (HNLF). 101
Figure 9-4: (a) Sampling scope trace of 40 Gbit/s RZ-OOK bits with inset
showing 0, 0.5, and 1ps delay shifts. (b) Tuning range of our
cascaded AOMs is shown through optical mixing
measurements. 103
Figure 9-5: Bit-error-rate curves for varying delay values with and without
the AOMs. 104
Figure 10-1: Conceptual block diagram of the demonstrated optical buffer.
Input packet stream is sent to two paths. Upper path induces the
relative delay on the selected packet(s), where the lower path
deletes any desired packet(s). 107
Figure 10-2: Conceptual block diagram of the conversion/dispersion
technique used to generate relative delays in the optical buffer.
The first wavelength conversion controls the amount of delay.
The second wavelength conversion is the phase conjugation
stage. After the delay, the signal is converted back to the
original wavelength to have a wavelength transparent delay. 108
Figure 10-3: Illustration of reconfiguration of the optical buffer. Packets to
be delayed are extracted to the corresponding wavelengths in the
first wavelength conversion stage. Thus, they experience
different amounts of delay in the dispersive element. The
reconfiguration should take place within the guard time between
the packets. Therefore, the minimum guard time without any
data loss is determined by the reconfiguration speed. 110

xii

Figure 10-4: Experimental setup for the optical buffer. Modifications for
demonstration of reconfiguration are shown with dotted lines
and italic titles. MZM: Mach-Zehnder modulator; CLK: clock;
TDL: tunable delay line; BPF: bandpass filter; DCF: dispersion
compensating fiber; SSMF: standard single mode fiber; PPLN:
periodically poled Lithium Niobate waveguide; Rx:
preamplified receiver. 111
Figure 10-5: Packet-1 being buffered from time slot 1 to time slot 11. Eye
diagrams of the signal shown are also given. (a) 40 Gbit/s input
packet stream (424 bits/packet, 1 ns guard time); (b) Packet-1
after extraction to λPKT1 (~1552.5 nm) in PPLN-1. (c) Packet-1
after double passing through the DCF and after SMF, signal is at
λPKT_C (~1556.9 nm) due to phase conjugation; (d) Output
packet stream where delayed Packet-1 is converted to λSig and
original Packet-1 is deleted from time slot 1 in the lower path. 117
Figure 10-6: Experimental spectra of the wavelength conversion processes
in the buffer. (a) Packet extraction in the PPLN-1 with gated
pump λGP1 (1550.6 nm). PPLN-1 QPM wavelength is shown
with a dotted line (~1551.6 nm); (b) Phase conjugation in the
HNLF; (c) Delayed Packet-1 is wavelength converted back to
λSig in PPLN-2. PPLN-2 QPM is shown with the dotted line
and is at ~1552.7 nm. The scale is the same, 8 dB/div and 3
nm/div, for all plots. 119
Figure 10-7: (a) Relative delay achieved for the system for a 40 Gbit/s input
signal; (b) Output packet stream for various buffering scenarios
including zero and maximum delay. 120
Figure 10-8: Experimental spectra of the three wavelength conversion stages
for the cases of: (a) maximum (116 ns), (b) middle, and (c) zero
delay. The first row shows the packet extraction in PPLN-1, the
second row shows the phase conjugation in HNLF, and the third
row shows the wavelength conversion of the delayed Packet-1
to the original wavelength in PPLN-2. For all plots, the center
wavelength is 1552.7 nm and the scale is 3 nm/div for the
horizontal and 8 dB/div for the vertical axis. 122

xiii

Figure 10-9: BER performances for several buffering scenarios. Back-to-
back performance and BER performance of the signal at the
output of lower path (with Packet-1 deleted) is also given for
comparison. 123
Figure 10-10: Packets-2 and -3 being buffered by three and five time slots in
the reconfiguration experiment. (a) The input packet sequence
of 8 packets. The guard time between the Packet-2 and Packet-3
is 25 ps. The inset shows the guard time; (b) An illustration of
the gated pumps generated by the switch and the MZM in the
packet extraction stage; (c) Extracted Packets 2 and 3; (d)
Packets 2 and 3 after the second pass through the DCF; (e)
Output packet sequence where Packets 2 and 3 are inserted at
the corresponding time slots. 124
Figure 10-11: Experimental spectra of the wavelength conversion processes
in the buffer. (a) Packet extraction in the PPLN-1 with gated
pump λGP1 (1550.6 nm). PPLN-1 QPM wavelength is shown
with a dotted line (~1551.6 nm); (b) Phase conjugation in the
HNLF; (c) Delayed Packet-1 is wavelength converted back to
λSig. PPLN-2 QPM is shown with the dotted line and is at
~1552.7 nm. 125
Figure 10-12: Transient response of the 2x2 Lithium Niobate switch used
for toggling between pump lasers in packet extraction. The
rise/fall time is ~25 ps for both output ports. 127
Figure 10-13: Last and first several bits of the Packets 2 and 3, respectively:
(a) Packet-2 and Packet-3 from the input packet stream with a
guard time (i) 25 ps, (ii) 1 ns; (b) Packet-2 and Packet-3 after
extraction in the PPLN-1. For Packet-2, Packet-3 extraction
pump is turned off for demonstration purposes; (c) Packet-2 and
Packet-3 after the delay (before the combination with the lower
arm) when both gated pumps are on. The scale is 100 ps/div for
all the plots except (a)-(ii). 128
Figure 10-14: BER performances of the buffer reconfiguration experiment
for guard times of 1 ns and 25 ps. 130

xiv

Abstract

Increased data traffic demands, along with a continual push to minimize cost per
bit, have recently motivated a paradigm shift away from traditional on-off keying
(OOK) fiber transmission links towards systems utilizing more advanced modulation
formats. In particular, modulation formats that utilize the phase of the optical signal,
including differential phase shift keying (DPSK) and differential quadrature phase
shift keying (DQPSK) along with polarization multiplexing (Pol-MUX), have
recently emerged as the most popular means for transmitting information over long-
haul and ultra-long haul fiber transmission systems. DPSK is motivated by an
increase in receiver sensitivity compared to traditional OOK. DQPSK is motivated
by a doubling of the spectral efficiency, along with increased tolerance to dispersion
and nonlinear distortions. Coherent communications has also emerged as a primary
means of transmitting and receiving optical data due to its support of formats that
utilize both phase and amplitude to further increase the spectral efficiency
(bits/sec/Hz) of the optical channel, including quadrature amplitude modulation
(QAM). Polarization multiplexing of channels is a straight forward method to allow
two channels to share the same wavelength by propagating on orthogonal
polarization axis and is easily supported in coherent systems where the polarization
tracking can be performed in the digital domain. Furthermore, the forthcoming IEEE
100 Gbit/s Ethernet Standard, 802.3ba, provides greater bandwidth, higher data rates,

xv

and supports a mixture of modulation formats. In particular, Pol-MUX (D)QPSK has
grown in interest as the high spectral efficiency allows for 100 Gbit/s transmission
while still occupying the current 50 GHz/channel allocation of current 10 Gbit/s
OOK fiber systems. In this manner, 100 Gbit/s transfer speeds using current fiber
links, amplifiers, and filters may be possible.
In addition to advanced modulation formats, it is expected that optical signal
processing may play a role in the future development of more efficient optical
transmission systems. The hope is that performing signal processing in the optical
domain may reduce optical-to-electronic conversion inefficiencies, eliminate
bottlenecks and take advantage of the ultrahigh bandwidth inherent in optics. While
40 to 50 Gbit/s electronic components are the peak of commercial technology and
100 Gbit/s capable RF components are still in their infancy, optical signal processing
of these high-speed data signals may provide a potential solution. Furthermore, any
optical processing system or sub-system must be capable of handling the wide array
of data formats and data rates that networks may employ. It is also worth noting that
future networks may use a combination of data-rates and formats while it has been
estimated that “we may start seeing the first commercial use of Terabit Ethernets
by 2015”. –Robert Metcalfe.
To this end, the work presented in this Ph.D. dissertation is aimed at addressing
the issue of optical processing for advanced optical modulation formats. All optical

xvi

multiplexing and demultiplexing of Pol-MUX and phase and QAM encoded signals
at the 100 Gbit/s Ethernet standard is addressed. The creation and development of an
extremely large continuously tunable all-optical delay capable of handling a variety
of modulation formats and data rates is presented. As optical delays are viewed as a
critical element to achieve efficient and reconfigurable signal processing, the
presented delay line is also utilized to enable a tunable packet buffer capable of
handling data packets of varying rate, varying size, and multiple modulation formats.

1
Chapter 1:
Introduction

Low-loss optical fiber was introduced in 1970 by Corning. The first national
network based on fiber optic technology did not appear until the mid-1980s. These
links were first operated at an initial rate of merely 51.84 Mb/s, referred to as OC-1
under the Synchronous Optical Networks (SONET) North American standard [82].
Since this time, fiber optic telecommunication has advanced tremendously.
Currently, the majority of long-haul transmission and global networking is made
possible through the use of fiber optic communication links. Deployed links
currently operate up to 10 Gbit/s (OC-192) with 40 Gbit/s links beginning to be
deployed, the 100 Gbit/s standard development nearing completion and
demonstrations of single-channel rates at 1-Tb/s and beyond. In order to meet current
and future projected demands [18], mainly driven by the exponential growth of
internet traffic, efforts are being made to transmit information in a more cost-
effective and efficient manner and at ever-increasingly higher rates.
Initial fiber optic links employed simple on-off keying (OOK) modulation, in
which the laser intensity was directly modulated to encode digital information. Since
then, techniques for transmission and detection of optical information have gone
through many phases. In the late 1970’s there was a significant interest in receivers
employing coherent detection. Coherent detection incorporates a local oscillator
(LO) laser to mix with the incoming signal, thereby generating an electrical beat

2
signal carrying the modulating signal. This allowed for ultimate receiver sensitivity
but required a complex receiver design, including a phase-locked-loop and a narrow
linewidth LO. After the advent of optical amplifiers in the 1980’s, the original
coherent techniques were abandoned, as optical amplification was able to provide
more robust, cost-effective solutions with comparable sensitivity.
A decade or two later, coherent communications began a revival due to the ever
increasing need for optical bandwidth and the decreasing amount of bandwidth
available. Additionally, the ability to compensate for many of the optical
impairments that limit current optical systems using digital signal processing (DSP)
at the receiver presents a cost-effective way of upgrading current links to higher
data-rates while providing more robust operation. Current commercial direction is to
utilize coherent technology to both upgrade existing links and in the deployment of
new ones to provide 100 Gbit Ethernet connectivity.
With the continual demands for increased transmission lengths and data rates,
research has recently focused on more advanced optical transmission and detection
schemes, which provide some key advantages including: enhanced receiver
sensitivity, increased spectral efficiency, increased single-user channel rates, and
reduced complexity. Differential phase shift keying (DPSK) emerged as the first
practically promising scheme for ultra-high bit-rate transmission. While not widely
deployed, DPSK installations are still being utilized. Furthermore, multi-level
versions of DPSK, namely differential quaternary phase shift keying (DQPSK) are

3
being explored as bandwidth efficient methods for meeting future demands. In
addition polarization division multiplexing (PDM) is being used as a means to
further double spectral efficiency, reduce the required bandwidth of electronic
components and reduce sensitivities to fiber impairments. In particular PDM-
(D)QPSK, utilizing coherent receivers, has become the industry favorite for reaching
the 100 Gbit/s mark. The development of the IEEE 100 Gbit/s Ethernet Standard,
802.3ba, has set a goal beyond what is capable with current OOK modulation and
electronic components. The standard includes provisions for the use of these
advanced modulation formats especially multi-level and Pol-MUX formats, due to
their ability to send multiple bits per symbol. This allows for the use of lower rate
electronics at the receiver and transmitter but makes midstream processing and
routing of the data difficult. Formats that utilize both amplitude and phase
modulation have become the most popular means for achieving the 100 Gbit/s
standard and for demonstrations of future 400 Gbit/s systems [53, 95]. Quadrature
amplitude modulation (QAM) has emerged as the dominant candidate for future
modulation formats.
In addition to advanced modulation formats, the use of optical signal processing
may play a key role in future high-speed dense WDM networks. Signal processing is
generally considered an efficient and powerful enabler for a host of communication
functions as well as a system performance enhancer. The hope is that performing
signal processing in the optical domain might reduce any optical-electronic

4
conversion inefficiencies and take advantage of the ultrahigh bandwidth inherent in
optics [94]. It is important that optical signal processing functions be capable
handling and operating on advanced modulation formats. In this manner, the
bottleneck of midstream routing and processing may be alleviated. One of the basic
building blocks to achieve efficient and reconfigurable signal processing is a
continuously tunable optical delay line, and yet, this element has historically been
difficult to realize.
In this dissertation, novel methods for the optical processing of high speed, ≥40
Gbit/s, signals including support for both basic OOK and advanced modulation
formats, are presented. The proposal is structured as follows. The next chapter
introduces potential modulation formats to be used in future 100 Gbit/s systems,
along with some potential non-linear processes that can support these formats. The
subsequent chapters 3-9 present material in support of this dissertation.

5
Chapter 2:
Advanced Modulation Formats and Format
Transparent Optical Signal Processing

This chapter provides a perspective of the progress in the field of optical
modulation formats and an introduction to optical signal processing. To this end, an
overview of advanced modulation formats that are favored for future 100 Gbit/s
systems and nonlinear optical processes capable of supporting these formats.
2.1 Motivation for Advanced Modulation Formats
The choice of transmission and detection of optical information depends on
various system parameters, including link length, number of users, desired bit rate,
desired quality of service and cost. For short distance metro links, OOK may be the
most suitable format. As the link length and the single-channel bit rate increases,
OOK becomes more susceptible to waveform degradations; including chromatic
dispersion (CD), polarization mode dispersion (PMD), self phase modulation (SPM),
cross phase modulation (XPM) and four-wave-mixing (FWM). In such
environments, more advanced modulation formats may be more suitable. DPSK has
recently emerged as the most practically promising format for future high bit rate,
long-haul systems. Other potential formats include, quaternary phase shift keying
(QPSK) and quadrature amplitude modulation (QAM) and both polarization
multiplexing (Pol-MUX) and orthogonal frequency division multiplexing (OFDM)

6
of these signals. In addition, a trade-off exists between non-return-to-zero (NRZ) and
return-to-zero (RZ) formats. NRZ formats tend to be more spectrally efficient, while
RZ formats tend to have higher tolerance to nonlinear effects such as SPM, XPM and
FWM, along with better receiver sensitivity. Since many of these impairments are
exponentially proportionally to the bandwidth of the optical signal, polarization
multiplexing (Pol-MUX) has gained much recent attention due to its inherent ability
to transmit twice the data using the same optical bandwidth. This dissertation
proposal will mainly focus on the nonlinear processing of these advanced formats
modulation formats, in particular DPSK, DQPSK, QAM, and polarization
multiplexing of such formats.
2.2 Differential Phase Shift Keying (DPSK)
Differential phase shift keying has been used extensively in the RF domain for
transmission of digital information. Although (non-differential) binary phase shift
keying (BPSK) has superior tolerance to DPSK, the receiver structure for DPSK is
much simpler and the penalty for differential detection is fairly small. In DPSK,
information is transmitted via the differential phase of the optical carrier:
( ) ( ) ( ) t t t T ϕ φ φ Δ = − + .
The complex electric field of the laser source can be written as:
{ } ( ) cos ( ) E A t t t ω ϕ = + Δ , where A(t) represents the time-varying complex
amplitude, ω is the radial frequency of the optical field and φ(t) is the time-varying
optical phase of the laser. In the typical convention, a 1-bit results in a change in

7
phase of 180° between the current and previous bit, while a 0-bit results in no phase
change (note: the opposite convention has also been used in the past). Because an
optical phase modulator typically encodes the absolute phase, an electronic
differential encoder is employed at the transmitter to encode the bits prior to
modulation onto the optical carrier. The precoding for DPSK follows the
relationship:
1 k k k
d d b
−
= + , where
k
b represents the original bits and
k
d
represents the differentially encoded bits. An illustration of DPSK transmission is
shown below in Figure 2-1(a) and (b).

(a) (b)
Figure 2-1: (a) Illustration of DPSK transmission (b) DPSK constellation
The reason DPSK is an attractive modulation format is the ~3dB reduced OSNR
requirement to achieve a given bit error rate, when compared to OOK. This reduced
OSNR requirement results from the increased Euclidean distance between
constellation points (extra 2 in E-field). The resulting expression for the bit error
rate is ( )
1
exp /
2
b o
BER E N = − when balanced detection is employed [37]. For
example, OOK requires approximately 38.3 photons/bit to reach a BER of 1E
-9
,
compared to 20 photons/bit for DPSK [37]. Note that the E
b
/N
o
can also be expressed
Original Bits
Encoded Bits
1 k k k
d d b
−
= +
Modulated E-field
1 1 1 1
1 1
0 0
0 0 0 0
π 0 0 π 0 0
Bit Phase
In-Phase
(I) Channel
Quadrature
(Q) Channel
Δφ=180°for 1-bit
Δφ=0°for 1-bit

8
as: /( )
rx
P hvR , where P
rx
is the received power, h is Planck’s constant, v is the optical
frequency and R is the bit rate. The increased performance from balanced detection
can be utilized to increase transmission length and/or reduce transmit power. In
addition, DPSK is also more tolerant to narrowband optical filtering [96], chromatic
dispersion [91], and some nonlinear effects [35]. Shown in Figure 2-2 are the
theoretical BER curves for some of the transmission and detection schemes
discussed in this dissertation: differentially detected DPSK, coherently detected
DPSK, coherently detected BPSK, coherently detected DQPSK and differentially
detected DQPSK. Note in general there is a performance penalty associated with
differential detection. For DPSK the penalty is larger at higher BER values (e.g. the
penalty is 1.2 dB at 1E-4 and only 0.4 dB at 1E
-9
).

Figure 2-2: Theoretical bit error rate curves for coherent and differential
detection versions of binary and quaternary phase modulation formats.
DPSK phase modulation can be achieved using a few different schemes. The
earliest reported techniques utilized direct modulation of a DFB laser to achieve a π
0 2 4 6 8 10 12 14 16 18 20
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
Eb/No (dB)
Log10(BER)
DPSK (differential detection)
BPSK and QPSK (coherent detection)
DPSK (coherent detection)
DQPSK (coherent detection)
DQPSK (differential detection)

9
phase shift [90]. This technique has not been widely adopted but has been shown to
provide some benefits for short link systems. The most straightforward method of
generation is to use an electro-optic Lithium Niobate (LiNbO3) phase modulator, in
which an applied voltage on two electrodes induces a change in the index of
refraction of an optical waveguide (situated between the electrodes). The phase of
the output electric field will be proportional to the applied voltage:
) / exp(
π
π V V j E E
i o
= , where V is the applied voltage and Vπ is the voltage required
to impart a phase shift of π radians.
Because the output phase of the phase modulator is linear with voltage, the RF
drive signal must be very accurate in order to avoid excessive degradation. Changes
in amplifier gain and the Vπ of the modulator over life and temperature makes it
difficult to maintain this. Therefore, it is more common to use a Mach-Zehnder
modulator (MZM) to achieve the desired optical phase. In an MZM, light is split and
travels in two separate equidistant paths. Electrodes are applied to one or both of the
paths in order to modulate the optical phase, via a change in index of refraction,
before optically recombining the two paths. In general, the output electric field will
depend on the input field E
i
and the driving voltages, V
1
and V
2
[40]:
















+








=
π π
π π
V
V
j
V
V
j
E
E
i
o
2 1
exp exp
2

When two electrodes are available the MZM is referred to as a dual-drive MZM.
In a single drive MZM the output field takes on the form:

10
2 / 1
cos 1
2 















+ =
π
π
V
V E
E
i
o

The MZM transmission function obeying this equation is shown below in Figure
2-3. Notice that the phase changes between 0 and π when transitioning through the
null point on the transfer function. As illustrated below, to generate DPSK the
modulator is biased at the null position and driven between the two peaks of the
transmission response. This requires a peak-to-peak drive voltage of 2Vπ. Note that
the resulting signal is almost a constant envelope (amplitude) signal, with the
exception of intensity dips located during phase transitions. The resulting dips are a
function of the modulator bandwidth and the peak to peak drive signal. Most
importantly, the phase shift between bits will be almost exactly π, even if the drive
signal is reduced significantly below 2Vπ. Reduction of the peak-to-peak drive does
not impact the optical phase; it only causes a reduction in optical amplitude because
the ones and zeros are not driven to the peak transmission points.

11

Figure 2-3: Mach-Zehnder modulator configuration for DPSK modulation
[37].
Return-to-zero modulation formats in general have higher receiver sensitivity due
to an increase in the peak-to-average optical power [57]. Furthermore, they also tend
to be more resilient to nonlinear effects [35]. The above MZM can also be used to
“pulse carve” the optical NRZ signal into an RZ format. There are three main ways
that this can be performed, as shown in Figure 2-4 below. The first method uses a
sinusoidal clock at the bit rate to generate 50% RZ pulses. This requires a drive
signal with peak-to-peak voltage equal to the Vπ of the modulator, along with a DC
signal to bias the modulator at the quadrature point. The other two methods utilize a
half bit rate clock with peak-to-peak voltage equal to 2Vπ. Biasing at the peak
generates a 33% RZ pulse, while biasing at the null generates a 66% RZ pulse.
Drive Voltage MZM Transmission
Optical Power
Optical Field
π π π π
0
Time
DPSK Drive
Signal
1 0 1 1 1 0
Time
π π π π
0
π π π π π π π π π π π π 0

12

33% RZ Carving 66% RZ Carving 50% RZ Carving
Figure 2-4: Mach-Zehnder modulator configuration for RZ pulse carving [37].
Typically, DPSK detection requires conversion from differential-phase to
intensity. The most common way to accomplish this is by using a delay-line
interferometer, referred to in the literature and this dissertation as a DLI or DI. A
conceptual block diagram of a DLI is shown below in Figure 2-5. It consists of an
optical splitter, a delay and phase shift in one arm, and an optical combiner to
recombine/interfere the two paths. The output of the interferometer is two signals
which represent the constructive and destructive interference between neighboring
bits (i.e. the two outputs will be inverted versions of one another). If adjacent bits
have the same phase they will add together in the constructive port and cancel in the
destructive port. Likewise, when adjacent bits are 180 degrees out of phase they will
cancel in the constructive port and combine in the destructive port. The two inverted
versions at the output of the DLI can then be combined using a balanced
photodetector.
Assuming ideal 3 dB couplers, a 1-bit delay in the delayed path, along with a
phase shift of Δφ the interferometer can be expressed in the following way:
Drive Voltage
MZM Transmission
Half Rate Clock
V
p-p
=2Vπ
Drive Voltage
MZM Transmission
Half Rate Clock
V
p-p
=2Vπ
Drive Voltage
MZM Transmission
Full Rate Clock
V
p-p
=Vπ

13
4 3 4 2 1 4 3 4 2 1 4 4 3 4 4 2 1 4 3 4 2 1
d Input fiel
t j
pler Output cou shift phase and delay Path
T j
Coupler Input
out
out
e t E
j
j
e j
j
E
E
























=






Δ +
0
) (
*
1
1
2
1
*
0
0 1
*
1
1
2
1
) (
) (
2
1
θ
φ

After expansion, the two outputs can be expressed as:
[ ]
ϕ θ θ Δ −
− − =
j T t j t j
out
e e T t E e t E E
) ( ) (
1
) ( ) (
2
1

[ ]
ϕ θ θ Δ −
− + =
j T t j t j
out
e e T t E e t E E
) ( ) (
2
) ( ) (
2
1

where
) (
) (
t j
e t E
θ
represents the input modulated electrical field and ) ( ) ( T t t − −θ θ is
taken from the set {0, π} for DPSK and {0, π/2, π, 3π/2} for DQPSK.

Figure 2-5: Illustration of a DPSK receiver including delay-line
interferometer for differential-phase to intensity conversion and balanced
detection.
Typically, the interferometer requires a method for tuning the optical phase in
one of the two optical paths, in order to obtain the desired phase relationship between
the two paths. This is typically accomplished through a Peltier heating element or
through a piezo-electric device in one path to modify the optical path length, and
typically requires some form of control loop for stabilization over temperature.
Heater
1-bit delay
RF
Output
-
Delay-Line Interferometer
Balanced Detector

Input 40 Gbit/s NRZ
DPSK optical spectra
Figure 2-6: Frequency domain response of 1
signal is bandpass filtered in the constructive port and notch filtered in the
destructive port. Spectra are plott
The operation of a DLI can also be understood in the frequency domain. When
the DLI is properly aligned, the frequency response is proportional to
[ ] ) ( ) ( cos
2
T t t + −ϕ ϕ
destructive port, as shown in
between adjacent peaks or nulls, is equa
Typically, the delay is set equal to the bit rate of the incoming DPSK signal.
However, recent results have shown that improved performance is possible through a
larger delay value, especially in the presence o

NRZ-
DPSK optical spectra

Frequency response of
constructive port ∝
cos
2
(Δφ)
Frequency response of
destructive port
sin
2

Output spectra of
constructive port
Output spectra
destructive port
Frequency domain response of 1-bit interferometer. The input
signal is bandpass filtered in the constructive port and notch filtered in the
destructive port. Spectra are plotted with 10 dB/division and 50
GHz/division.
The operation of a DLI can also be understood in the frequency domain. When
the DLI is properly aligned, the frequency response is proportional to
in the constructive port and [ ( ) ( sin
2
t − ϕ ϕ
destructive port, as shown in Figure 2-6. The free spectral range (FSR), or width
between adjacent peaks or nulls, is equal to the inverse of the interferometer delay.
Typically, the delay is set equal to the bit rate of the incoming DPSK signal.
However, recent results have shown that improved performance is possible through a
larger delay value, especially in the presence of optical filtering or dispersion

14

Frequency response of
destructive port ∝
2
(Δφ)

Output spectra of
destructive port
bit interferometer. The input
signal is bandpass filtered in the constructive port and notch filtered in the
ed with 10 dB/division and 50
The operation of a DLI can also be understood in the frequency domain. When
the DLI is properly aligned, the frequency response is proportional to
] ) ( T t + in the
. The free spectral range (FSR), or width
l to the inverse of the interferometer delay.
Typically, the delay is set equal to the bit rate of the incoming DPSK signal.
However, recent results have shown that improved performance is possible through a
f optical filtering or dispersion [55].

15
The resulting spectrum in the constructive port is bandpass filtered, while the
spectrum in the destructive port is notch filtered.
2.3 Differential Quadrature Phase Shift Keying (DQPSK)
In DQPSK four phase levels are used to encode two bits into one symbol.
Symbols transition at half the aggregate bit rate, resulting in half the optical
bandwidth for a fixed aggregate bit rate when compared to more conventional binary
signaling. This increased spectral efficiency allows for tighter optical filtering,
reduced WDM channel spacing and reduced electronic bandwidth requirements. The
reduction in bandwidth also results in an increased tolerance to chromatic and
polarization mode dispersion.
DQPSK can be generated in a number of ways. The most obvious method is to
drive an optical phase modulator with a multi-level electrical signal with four levels:
{0, +Vπ/2, Vπ, -Vπ/2}. Another way to generate DQPSK, referred to as the “serial
approach”, is to generate a DPSK signal, followed by a phase modulator driven with
a binary electrical signal of level 0 and Vπ/2 [79]. Either of these approaches
requires the following precoding function [79]:
( )( ) ( )( )
1 1 1 1 − − − −
⊕ + ⊕ =
k k k k k k k
I a Q Q I b I

1 −
⊕ ⊕ =
k k k k
Q b a Q

The more common method for DQPSK generation is through the use of a parallel
or nested Mach-Zehnder modulator [38]. A parallel modulator consists of two

16
MZMs in an interferometer configuration with a 90° phase shift in one path, as
shown below in Figure 2-7. Each MZM is driven as if it were a DPSK signal with a
peak-to-peak drive of 2Vπ. The bottom branch is shifted by 90 degrees before
combining with the top branch. The result is a constellation diagram with four
possible phases: {+π/4, +3π/4}.

DQPSK generation using parallel MZM

DQPSK constellation diagram
Figure 2-7: Parallel modulator for generation of optical DQPSK, along with
ideal resulting constellation diagram.
When using a parallel modulator the following precoding function is employed
[38]:
( )( ) ( )( )
1 1 1 1 1 1 − − − − − −
⊕ ⊕ + ⊕ ⊕ =
k k k k k k k k k
I b I Q I a I Q I

( )( ) ( )( )
1 1 1 1 1 1 − − − − − −
⊕ ⊕ + ⊕ ⊕ =
k k k k k k k k k
I b I Q I a I Q Q

where ⊕ denotes the exclusive OR operation, I
k
and Q
k
are the precoded data bits,
and a
k
and b
k
are the original data bits at time k.
DQPSK direct-detection is typically performed using two delay-line
interferometers for differential-phase to intensity conversion. When the appropriate
DPSK
Drive
90°
DPSK
Drive
DC Bias
MZM
MZM
-1.0
-0.5
0.0
0.5
1.0
-1.0 -0.5 0.0 0.5 1.0

17
precoding function is employed the two outputs of the interferometers represent the
differentially-transmitted I and Q components of the constellation. As shown in
Figure 2-8, a phase offset of +π/4 and -π/4 are required in the two interferometers in
order to obtain the appropriate constructive and destructive interference.

Figure 2-8: Illustration of a typical DQPSK receiver, along with the
transmission response of each DLI output port. Four ports are staggered by ¼
of the symbol rate. Constructive and destructive port of each interferometer
staggered by ½ the symbol rate.
Although DQPSK has been shown to be more robust to channel impairments
when compared to DPSK or OOK, it has also been shown to be less tolerant to
imperfections in the transmitter and receiver structures. For example, DQPSK has
been shown to be 6 times more sensitive to phase misalignment in the demodulating
interferometer, when compared to DPSK [48]. Misalignment in the transmitter (in
either MZM or the 90 degree phase shifter) or in the receiver (phase misalignment
with respect to +/- 45 degrees) results in a cross-coupling between the I and Q
components and a resulting decrease in the OSNR.
T
+45°
T
-45°
-
-
I
Q
+0.125 +0.375 -0.375 -0.125
Frequency offset from carrier (normalized to data rate)
Q Destructive I Destructive
Q Constructive
I Constructive

18
2.4 Polarization Multiplexing (Pol-MUX)
Increasing Ethernet data traffic has determined the need for 100 Gbit/s Ethernet
(100GbE) transport solutions [26]. Although multiple-wavelength approaches have
been proposed, they are most likely for short reach deployment of 100 GbE [74]; on
the other hand 100 Gbit/s serial single-wavelength channel solutions avoid
challenges in multiplexing 100 GbE to several optical channels [74, 29] and
represent stronger and more feasible candidates for wide area networks. Electrical
time-division multiplexing systems [24, 97] have already been demonstrated, but the
required high-speed electronics are not yet a mature technology. Moreover, 100
Gbit/s binary serial solutions are dramatically sensitive to fiber impairments in terms
of chromatic dispersion (CD) and polarization mode dispersion (PMD).
Polarization multiplexing (Pol-MUX), in which two data channels are
transmitted on orthogonal polarization axis, serves as a straight forward method for
sending twice the information using the same optical bandwidth. Since the more
significant degrading effects are bandwidth dependent, this allows for the doubling
of the data rate while maintain the same dispersion tolerances. Furthermore, Pol-
MUXing can be used with already spectrally efficient modulation formats to further
increase the data rate without incurring the penalties of added optical bandwidth. To
this end, the scientific community generally acknowledges that the DQPSK format
together with Pol-MUXing may be the best candidate to achieve the 100 Gbit/s bit
rate [29]. DQPSK alone would halve the symbol rate increasing the CD and PMD

19
tolerance [20], yet still requiring 50 GHz electronics. Pol-MUXing of two DQPSK
channels, shown in Figure 2-9, allows the further halving the symbol rate. Thus, by
using 25 Gbaud Pol-MUX DQPSK, 100-Gbit/s capacity is achieved with high CD
and PMD tolerance [9]. Moreover, the narrow spectrum has an additional benefit
beyond the lowered dispersion effects. At 111 Gbit/s, which is a high enough rate to
allow for 100 Gbit/s data with a 7% error correction overhead and the 4% Ethernet
overhead, the signal is still compatible with the existing 50 GHz channel spacing that
is standard in current installed networks, greatly reducing the potential costs of
upgrading existing networks to the 100 Gbit/s rate [28].

Figure 2-9: Potential formats for the IEEE 802.3ba 100 Gbit/s Ethernet Standard
include quadrature-PSK, utilizing 4 phase states for 2 bits/symbol and
polarization multiplexing (Pol-MUX) to achieve 4 bits/symbol.
Because of the requirement for polarization tracking at the receiver, 100 Gbit/s
Pol-MUX DQPSK is generally associated with coherent detection [28], which
however requires ultra-high speed analog-to-digital converters and digital signal
00 01
11 10
00 01
11 10
00 01
11 10
00
01
11
10
Regular (D)QPSK
2 bits per symbol
Pol-MUX (D)QPSK
4 bits per symbol
polarization
axis
polarization
axis

20
processors (DSPs), in order to recover the Pol-MUX data and to compensate for
linear distortions. Real-time solutions are already feasible exploiting direct detection
together with polarization stabilization to demultiplex the Pol-MUX channels [13,
14]. Recently, polarization tracking of a 100 Gbit/s Pol-MUX signal has been
demonstrated by adding a small tone to one of the polarization states [10]. At the
receiver, this polarization state can be preferentially tracked allowing for direct
detection and processing.
Pol-MUXed channels are fairly easy to generate at a transmitter and decompose
at a receiver, since simple polarization-dependent couplers/splitters are available
[10]. However, polarization multiplexing and demultiplexing of high-data-rate
channels at an intermediate routing node in a network would be quite complex but
could bring enhanced throughput and performance. The complexity comes from the
fact that two data channels might originate on two different WDM wavelengths, such
that simple components are insufficient. While several methods for polarization
independent wavelength conversion exist, and wavelength conversion of a Pol-MUX
signal has been demonstrated. Little has been shown on using the inherent
polarization sensitivity of non-linear processes to perform all-optical signal
processing for Pol-MUXed signals. The following chapter presents recent work on
polarization manipulation for WDM-to-Pol-MUX and Pol-MUX-to-WDM
conversion. Furthermore, this conversion process is shown to be phase transparent
supporting advanced formats including DPSK and DQPSK.

21
2.5 Coherent Transmission and Reception
Spectral efficiency limits for various detection and modulation methods have
been studied in the linear [80, 34, 39] and nonlinear regimes [58, 46]. Noncoherent
detection and differentially coherent detection offer good power efficiency only at
low spectral efficiency, because they limit the degrees of freedom available for
encoding of information [46].
The most promising detection technique for achieving high spectral efficiency
while maximizing power (or SNR) efficiency, is coherent detection with polarization
multiplexing, as symbol decisions are made using the in-phase (I) and quadrature-
phase (Q) signals in the two field polarizations, allowing information to be encoded
in all the available degrees of freedom. When the outputs of an optoelectronic
downconverter are sampled at the Nyquist rate, the digitized waveform retains the
full information of the electric field, which enables compensation of transmission
impairments by digital signal processing (DSP) [44]. A DSP-based receiver is highly
advantageous because adaptive algorithms can be used to compensate time-varying
transmission impairments. Advanced forward error-correction coding can also be
implemented. Moreover, digitized signals can be delayed, split and amplified without
degradation in signal quality. DSP-based receivers are ubiquitous in wireless and
digital subscriber line (DSL) systems at lower data rates [44]. In such systems,
computationally intensive techniques have been demonstrated, such as orthogonal
frequency-division multiplexing (OFDM) with multiple-input-multiple-output

22
(MIMO) transmission in a real-time 1 Gbit/s wireless link [45]. Continued hardware
improvements will enable deployment of DSP-based coherent optical systems in the
next few years.
Experimental results in coherent optical communication have been promising.
Kikuchi demonstrated polarization-multiplexed 4-ary quadrature-amplitude
modulation (4-QAM) transmission at 40 Gbit/s with a channel bandwidth of 16 GHz
(2.5 bit/s/Hz) [47]. This experiment used a high-speed sampling oscilloscope to
record the output of a homodyne downconverter. DSP was performed offline
because of the unavailability of sufficiently fast processing hardware. The first
demonstration of real-time coherent detection occurred in 2006, when an 800 Mbit/s
4-QAM signal was coherently detected using a receiver with 5-bit analog-to-digital
converters (ADC) followed by a field programmable gate array [73]. In 2007,
feedforward carrier recovery was demonstrated in real time for 4-QAM at 4.4 Gbit/s
[51]. Savory showed the feasibility of digitally compensating the chromatic
dispersion (CD) in 6,400 km of SMF without inline dispersion compensating fiber
(DCF), with only a 1.2 dB OSNR penalty at 42.8 Gbit/s [77]. Coherent detection of
large QAM constellations has also been demonstrated. For example, 16-ary
transmission at 40 Gbit/s using an amplitude-phase-shift keying (APSK) format was
shown by Sekine et. al. [78]. In 2007, Hongo et. al. demonstrated 64-QAM
transmission over 150 km of dispersion-shifted fiber [41].

23
The most advanced detection method is coherent detection, where the receiver
computes decision variables based on the recovery of the full electric field, which
contains both amplitude and phase information [44]. Coherent detection thus allows
the greatest flexibility in modulation formats, as information can be encoded in
amplitude and phase, or alternatively in both in-phase (I) and quadrature (Q)
components of a carrier. Coherent detection requires the receiver to have knowledge
of the carrier phase, as the received signal is demodulated by a LO that serves as an
absolute phase reference. Traditionally, carrier synchronization has been performed
by a phase-locked loop (PLL). Optical systems can use (i) an optical PLL (OPLL)
that synchronizes the frequency and phase of the LO laser with the TX laser, or (ii)
an electrical PLL where downconversion using a free-running LO laser is followed
by a second stage demodulation by an analog or digital electrical VCO whose
frequency and phase are synchronized. Use of an electrical PLL can be advantageous
in duplex systems, as the transceiver may use one laser as both TX and LO. PLLs are
sensitive to propagation delay in the feedback path, and the delay requirement can be
difficult to satisfy [44]. Feedforward (FF) carrier synchronization overcomes this
problem. In addition, as a FF synchronizer uses both past and future symbols to
estimate the carrier phase, it can achieve better performance than a PLL which, as a
feedback system, can only employ past symbols. Recently, DSP has enabled
polarization alignment and carrier synchronization to be performed in software.

24

Figure 2-10: Coherent transmission system (a) implementation, (b) system model.
A coherent transmission system and its canonical model are shown in Figure
2-10. At the transmitter, Mach-Zehnder (MZ) modulators encode data symbols onto
an optical carrier and perform pulse shaping. If polarization multiplexing is used, the
TX laser output is split into two orthogonal polarization components, which are
modulated separately and combined in a polarization beam splitter (PBS). We can
write the transmitted signal as [44]:
,1 ( ( ))
,2
( )
( ) ( )
( )
s s
tx j t t
tx t k s
k tx
E t
E t P b t kT e
E t
ω φ +
Χ
 
= = −
 
 
∑

where T
S
is the symbol period, P
t
is the average transmitted power, b(t) is the pulse
shape (e.g., non-return-to-zero (NRZ) or return-to-zero (RZ)) with the normalization
2
( )
S
b t dt T =
∫
, ω
S
and ( )
S
t φ are the frequency and phase noise the TX laser, and
1, 2,
,
T
k k k
χ χ Χ   =
 
is a 2x1 complex vector representing the k-th transmitted symbol.

25
We assume that symbols have normalized energy:
2
1
k
E Χ
 
=
 
. For single-
polarization transmission, we can set the unused polarization component x
2,k
to zero.
The E-field at the output of the fiber is E
S
(t)=[E
S,1
(t),E
S,2
(t)]
T
, where:
2
( ( ))
, r ,
1
( ) P ( )
S S
j t t
S l m k lm S
k m
E t c t kT e
ω φ
χ
+
=
= −
∑∑

Under the assumption of Figure 2-10 where inline amplification completely
compensates propagation loss, P
r
= P
t
is the average received power, c
lm
(t)=b(t)⊗h
lm

(t) is a normalized pulse shape, and E
sp,l
(t) is ASE noise in the l-th polarization.
Assuming the NA fiber spans are identical and all inline amplifiers have gain G and
spontaneous emission factor n
sp
, the two-sided power spectral density (psd) of E
sp,l
(t)
is S
Esp
(f) = N
A
n
sp h
ω
s
(G-1)/G W/Hz [43].
The first stage of a coherent receiver is a dual-polarization optoelectronic
downconverter that recovers the baseband modulated signal. In a digital
implementation, the analog outputs are lowpass filtered and sampled at 1/T = M/KT
S
,
where M/K is a rational oversampling ratio. Channel impairments can then be
compensated digitally before symbol detection [44].

Figure 2-11: Single-polarization downconverter employing a (a) heterodyne and (b)
homodyne design.

26
We can consider a single-polarization downconverter, where the LO laser is
aligned in the lth polarization. For a dual polarization downconverter, two single-
polarization converters are used (one aligned to the vertical polarization axis and one
aligned to the horizontal axis), and the LO is aligned 45° linear such that half of its
power enters each single-polarization downconverter. Downconversion from optical
passband to electrical baseband can be achieved in two ways: in a homodyne
receiver, the frequency of the LO laser is tuned to that of the TX laser so the
photoreceiver output is at baseband. In a heterodyne receiver, the LO and TX lasers
differ by an intermediate frequency (IF), and an electrical LO is used to downconvert
the IF signal to baseband. Both implementations are shown in Figure 2-11. Although
we show the optical hybrids as 3-dB fiber couplers, the same networks can be
implemented in free-space optics using 50/50 beam splitters; this was the approach
taken by Tsukamoto [87]. Since a beam splitter has the same transfer function as a
fiber coupler, there is no difference in their performances [44].
In the heterodyne downconverter of Figure 2-11(a), the optical frequency bands
around ω
LO
+ω
IF
and ω
LO
−ω
IF
both map to the same IF. In order to avoid DWDM
crosstalk and to avoid excess ASE from the unwanted image band, optical filtering is
required before the downconverter. The output current of the balanced photodetector
in Figure 2-11 (a) is:
( ) { }
2 2
*
, 1 2 , , ,
( ) ( ) ( ) 2 Im ( ) ( ) ( )
het l S l LO l sh l
I t R E t E t R E t E t I t = − = +

27
where
( ( ))
, ,
( )
LO LO
j t t
LO l LO l
E t P e
ω φ +
= is the E-field of the LO laser, and P
LO,l
· ω
LO
and
( )
LO
t φ are its power, frequency and phase noise. I
sh,l
(t) is the LO shot noise.
Assuming P
LO
>> P
S
, I
sh,l
(t) has a two-sided psd of S
Ish
(f) = qRP
LO
A
2
/Hz.
Substituting our equation for the E-field after the fiber into our equation for the
output current, we get:
( ) ( )
'
, , r , ,
( ) 2 P ( )sin( ) ( ) ( ) ( ) ( )
het l LO l li IF lq IF sp l sh l
I t R P y t t y t cos t E t I t ω ω = + + +

where ω
IF
=ω
s
−ω
LO
is the IF, φ (t) =φ
s
(t)−φ
LO
(t) is the carrier phase, and y
li
(t) and
y
lq
(t) are the real and imaginary parts of:
2
( )
0 ,
1
( ) ( )
j t
l m k lm S
k m
y t c t kT e
φ
χ
=
= −
∑∑

The term
'
, ,
2 ( )
LO r sp l
R P E t is often referred to as the LO-spontaneous beat noise,
and
{ }
( ( )) '
, ,
( ) Im ( )
LO LO
j t t
sp l sp l
E t E t e
ω φ − +
= has a two sided psd of ½S
Esp
(f).
It can similarly be shown that the currents at the outputs of the balanced
photodetectors in the homodyne downconverter (Figure 2-11 (b)) are:
( ) ( )
2 2
'
hom, , 1 2 , r , ,
( ) ( ) ( ) P ( ) ( ) ( )
l i LO l li sp li sh li
I t R E t E t R P y t E t I t = − = + +

( ) ( )
2 2
'
hom, , 3 4 , r , ,
( ) ( ) ( ) P ( ) ( ) ( )
l q LO l lq sp lq sh lq
I t R E t E t R P y t E t I t = − = + +

where E’
sp,li
and E’
sp,lq
are white noises with a two-sided psd ½S
Esp
(f); and I
sh,li
and
I
sh,lq
are white noises with a two-sided psd of ½S
Ish
(f). Since it can be shown that
thermal noise is always negligible compared to shot noise and ASE noise [2], we can
neglect this term. In long-haul systems, the psd of LO-spontaneous beat noise is

28
typically much larger than that of LO shot noise; such systems are thus ASE-limited.
Conversely, unamplified systems do not have ASE, and are therefore LO shot-noise-
limited [44].
If one were to demodulate the heterodyne current by an electrical LO at ω
IF
, as
shown in Figure 2-11(a), the resulting baseband signals I
het,l,i
(t) and I
het,l,q
(t) will be
the same as those derived for the homodyne downconverter in Figure 2-11(b), with
all noises having the same psd’s. Hence, the heterodyne and the two-quadrature
homodyne downconverters have the same performance [8]. A difference between
heterodyne and homodyne downconversion only occurs when the transmitted signal
occupies one quadrature (e.g. 2-PSK) and the system is LO shot-noise limited, as this
enables the use of a single-quadrature homodyne downconverter that has the optical
front-end of Figure 2-11(a), but has ω
s
=ω
LO
. Its output photocurrent is
( )
'
, r , ,
2 P ( ) ( ) ( )
LO l lq sp l sh l
R P y t E t I t + + . Compared to the homodyne equation, the
signal term is doubled (four times the power), while the shot noise power is only
increased by two, thus yielding a sensitivity improvement of 3-dB compared to
heterodyne or two-quadrature homodyne downconversion. This case is not of
practical interest in this dissertation, as long haul systems are ASE-limited, not LO
shot-noise-limited. Also, for good spectral and power efficiencies, modulation
formats that encode information in both I and Q are preferred. Hence, there is no
performance difference between a homodyne and a heterodyne downconverter

29
provided optical filtering is used to reject image-band ASE for the heterodyne
downconverter [44].
The advantages of heterodyne downconversion are that it uses only one balanced
photodetector and has a simpler optical hybrid. However, the photocurrent has a
bandwidth of ω
IF
+ BW, where BW is the signal bandwidth. To avoid signal
distortion caused by overlapping side lobes, ω
IF
needs to be sufficiently large.
Typically, ωIF ≈BW, thus a heterodyne downconverter requires a balanced
photodetector with at least twice the bandwidth of a homodyne downconverter,
whose output photocurrents only have bandwidths of BW. This extra bandwidth
requirement is a major disadvantage [44]. In addition, it is also difficult to obtain
electrical mixers with baseband bandwidth as large as the IF.
Using either the homodyne or heterodyne methods discussed, the optical phase,
polarization, and amplitude can be simultaneously detected. In this manner,
modulation formats beyond the on-off-keying and binary and quadrature phase-shift-
keying discussed in the previous sections can be combined with polarization
multiplexing to deliver very high spectral efficiency. As mentioned, industry has
readily adopted and is in the process of implementing 100 Gbit/s Ethernet systems
based on Pol-MUX QPSK at 25 Gbaud (4 bits per symbol yields 25*4 = 100 Gbit/s)
with coherent (heterodyne) detection. Yet coherent detection allows for much greater
efficiency by moving to both amplitude and phase modulation.

30
The most common research goal is currently 16-ary quadrature amplitude
modulation (QAM) which uses 3 amplitude levels and 12 phase levels to produce 4
bits per symbol (2^4=16). Combined with Pol-MUXing and 50 Gbaud electronics,
single wavelength demonstrations in excess of 400 Gbit/s have been shown [95].
Assuming the use of Gray coding, the BER for a square M-QAM constellation
with coherent detection is approximated by [64]:
3 2 1
( )
2( 1)
QAM b
b
b M
P M erfc
b M M
γ
   
−
≈
   
   
−
   

where M is the number of signal points in the constellation. b = log2 (M) is the
number of bits encoded per symbol,
2 2
/
s k k
E E n γ χ
   
=
   
is the SNR per symbol
in single-polarization transmission,
2 2
/
s k k
E E γ
   
=
   
Χ n is the SNR per symbol
in dual-polarization transmission (e.g. polarization-multiplexed or Pol-SK), and
γ
b
=γ
s
/b is the SNR per bit.
Because QAM uses all four available DOF for encoding information, it has better
SNR efficiency than the other formats, and exhibits a steeper slope at high spectral
efficiency. We can compute the SNR per bit required to achieve a target BER of
10
−3
, which is a typical threshold for receivers employing forward error-correction
coding (FEC) and compare it to the similar cases of phase only modulation (16-PSK)
and amplitude only modulation (16-ASK). 16-QAM requires a signal to noise ration
of ~10.5 dB, almost 4 dB better than the phase only case and almost 15 dB better

31
than the amplitude only case, making it a strong format for future optical systems
[44].
2.6 Raman Amplification
Raman amplification is based on the Stimulated Raman Scattering (SRS)
phenomenon, when a lower frequency 'signal' photon induces the inelastic scattering
of a higher-frequency 'pump' photon in an optical medium in the nonlinear regime.
As a result of this, another 'signal' photon is produced, with the surplus energy
resonantly passed to the vibrational states of the medium [3]. This process, as with
other stimulated emission processes, allows all-optical amplification. The Raman-
active medium is often an optical fiber, although it can also be a bulk crystal, a
waveguide in a photonic integrated circuit, or a cell with a gas or liquid medium. For
telecom purposes, optical fiber is most commonly used as the nonlinear medium for
SRS. In this case, shown in Figure 2-12, it is characterized by a resonance frequency
downshift of ~11 THz (corresponding to a wavelength shift at ~1550 nm of ~90 nm)
[85]. The SRS amplification process can be readily cascaded, thus accessing
essentially any wavelength in the fiber low-loss guiding windows (both 1300 and
1550). In addition to applications in nonlinear and ultrafast optics, Raman
amplification is used in optical telecommunications, allowing all-band wavelength
coverage and in-line distributed signal amplification.

32

Figure 2-12: Raman gain of fused quartz plotted as a function of frequency shift
from an exciting line at 526 nm. The experimental point is the gain measured in
the amplifier and the error bar represents a combination of the uncertainties both
in the measurement of the gain and the spontaneous cross section [85].
In a Raman amplifier, the signal is intensified by Raman amplification. Unlike
the EDFA and SOA the amplification effect is achieved by a nonlinear interaction
between the signal and a pump laser within an optical fiber. Fibers used for Raman
amplifiers do not need to be doped with rare earth ions [3]. In principle, any ordinary
single-mode fiber could be used, and in practice the transmission fibers themselves
are often suitable. This leads to the first of two types of Raman amplifiers, a
distributed amplifier, in which long lengths of fiber are used and the pump lasers are
coupled in with the signal periodically to offset the fiber loss and increase
transmission distances. However, there are special fibers with increased Raman gain,
resulting from certain dopants (e.g. Germania) for enhanced Raman cross sections,

33
or simply from a small effective mode area. Such fibers are used to create the second
type of Raman amplifier, a lumped Raman amplifier, where a shorter piece of fiber is
dedicated to amplification only.
The pump light may be coupled into the transmission fiber in the same direction
as the signal (co-directional pumping), in the opposite direction (counter-directional
pumping) or both. Counter-directional pumping is more common as the transfer of
noise from the pump to the signal is reduced, but typically provides less gain for the
same pump power, type of fiber, and fiber length [3, 85].
The pump power required for Raman amplification is higher than that required
by the EDFA, with in excess of 500 mW being required to achieve useful levels of
gain in a distributed amplifier. Lumped amplifiers, where the pump light can be
safely contained to avoid safety implications of high optical powers, may use over
1W of optical power.
The principal advantage of Raman amplification is its ability to provide
distributed amplification within the transmission fiber, thereby increasing the length
of spans between amplifier and regeneration sites. The amplification bandwidth of
Raman amplifiers is defined by the pump wavelengths utilized and so amplification
can be provided over wider, and different, regions than may be possible with other
amplifier types which rely on dopants and device design to define the amplification
'window'. The gain spectrum can be tailored by using different pump wavelengths
simultaneously. If the pump wavelength is polarized, the Raman gain is polarization-

34
dependent. This effect is often unwanted, but can be suppressed e.g. by using two
polarization-coupled pump diodes or a pump depolarizer [84].
A telecom Raman amplifier is typically pumped with continuous-wave light from
a diode laser. Efficient amplification of ultra-short pulses is also possible using co-
propagating pump pulses. However, the phenomenon of group velocity mismatch
then severely limits the useful interaction length, particularly for pulse durations
below 1 ps [85].
As previously mentioned, the relationship between the polarization and the E
field is nonlinear:
E E E E P ) (
2 3
1 0
3 3
0
1
0
χ χ ε χ ε χ ε + = + =

The quantity in parentheses looks like a susceptibility that changes with light
intensity. In fact, this relation describes a material with a refractive index and
absorption coefficient that are intensity dependent. For Raman amplification, the
high power pump changes the absorption coefficient of the material, making it
negative and producing gain.
Raman scattering can be viewed in the semi-classical model as a two-photon
interaction in which the material makes a real transition from an initial to a final state
and a pump photon is destroyed while a Stokes or anti-Stokes photon is created.
Several types of Raman interactions are possible. These are illustrated in the level
diagrams of Figure 2-13, which show Stokes scattering, anti-Stokes scattering, anti-
Stokes scattering with four-wave mixing, multiple Stokes scattering, and hyper-

35
Raman scattering. Of these, the most common is Stokes scattering (Figure 2-13 (a)),
in which the pump photon at frequency ω
L
is scattered into a longer-wavelength
Stokes photon ω
S
, accompanied by the excitation of an internal mode of the medium
at frequency ω
o
[3].

Figure 2-13: Level diagrams showing (a) stimulated Raman Stokes scattering; (b)
stimulated Raman anti-Stokes scattering; (c) coherent anti-Stokes four-wave
mixing; (d) multiple Stokes and anti-Stokes scattering; and (e) hyper-Raman
scattering [3].
Anti-Stokes scattering (Figure 2-13 (b)), in which the pump photon is scattered
into a shorter-wavelength anti-Stokes photon ω
AS
accompanied by the de-excitation
of an internal mode of the medium, requires initial excitation into upper levels of the
medium. The anti-Stokes interaction illustrated in Figure 2-13 (b) is less common

36
than the Stokes interaction, occurring most often in spontaneous Raman scattering
when the levels are excited thermally. In stimulated processes, this interaction incurs
exponential loss unless a population inversion is created between the initial and final
states [3].
Anti-Stokes Raman scattering involving a four-wave mixing interaction, as
illustrated in Figure 2-13 (c), is much more common in Raman scattering. In this
interaction, two pump photons are scattered into a Stokes and anti-Stokes interaction
with no net excitation or de-excitation of the medium. This interaction requires
perfect or approximate phase matching, depending on the conditions of the scattering
interaction. Multiple Raman scattering (Figure 2-13 (d)) occurs when the Stokes
wave becomes powerful enough to drive its own Raman interaction. This generally
occurs when the pump wave is significantly above the stimulated Raman threshold.
Under these conditions, multiple Stokes waves are generated, each one shifted from
its effective pump wave by the frequency of the internal mode of the medium.
Multiple Stokes and anti-Stokes waves can also be created through four-wave mixing
processes involving one or more of the pump- or frequency-shifted waves. Hyper-
Raman scattering (Figure 2-13 (e)) involves multi-photon interactions in which two
or more pump photons are scattered into a single Stokes photon with excitation of an
appropriate mode of the material [3].

37
In all of these interactions, energy is conserved among the incident and scattered
photons and internal energy of the medium. The appropriate frequency and k-vector
relations are summarized as follows [3]:
Stokes scattering: ω
s
= ω
L
– ω
o
and k
o
= k
L
− k
s

Anti-Stokes scattering ω
AS
= ω
L
+ ω
o
and k
o
= k
AS
− k
L

Coherent anti-Stokes scattering ω
S
+ ω
AS
= 2ω
L
and k
AS
= 2k
L
− k
s

Multiple Stokes scattering ω
ns
= ω
(n−1)S
− ω
o
= ω
L
– nω
o
and k
o
= k
(n−1)S
− k
nS

Hyper-Raman scattering ω
S
= 2ω
L
− ω
o
and k
o
= 2k
L
− k
s

Not all levels in a material can be involved in Raman scattering. In general,
Raman scattering follows the rules for two-photon dipole transitions. In materials
with inversion symmetry, the initial and final states must have the same parity, and
therefore are mutually exclusive with absorptive transitions. In materials without
inversion symmetry, internal levels can be both Raman and optically active [3].
2.7 Wavelength Conversion Using PPLN Waveguides
The total polarization induced by an applied electric field is not linear, but
instead can be expressed in a Taylor expansion as [25]:
( ) ...... : : :
) 3 ( ) 2 ( ) 1 (
+ + + = E E E E E E P
o
χ χ χ ε

The nonlinear response of an atomic system to an applied electric field gives rise
to many interesting nonlinear optical phenomena. The second-order optical
susceptibility, χ
(2)
gives rise to second-harmonic generation, sum frequency
generation, difference frequency generation, parametric amplification and parametric

38
oscillation. The third-order optical susceptibility, χ
(3)
gives rise to stimulated Raman
scattering, stimulated Brillouin scattering, self-phase modulation, cross-phase
modulation, third harmonic generation, and four-wave-mixing. For the purposes of
introducing the techniques used in this dissertation, only second-order processes will
be discussed, in particular second harmonic generation (SHG), sum frequency
generation (SFG) and difference frequency generation (DFG). In the next section,
four-wave mixing in fiber (FWM) will also be introduced.
Second order processes involve the interaction among three different
electromagnetic waves. For example, in sum frequency generation photons at
frequencies υ
1
and υ
2
combine to generate a new photon at frequency υ
3
= υ
1
+ υ
2
. In
order for efficient wavelength conversion to take place, phase matching must be
satisfied. In the case of SFG, phase matching refers to a matching of the propagation
constants such that
0 ) (
2 1 3
= + − = Δ k k k k

where k
j
= n
j
ω/c = 2πn
j
/λ for each electromagnetic wave. For the case of SHG
(frequency doubling), υ
1
= υ
2
and phase matching simplifies to:
0 2
1 3
= − = Δ k k k
.
Likewise, for DFG in which two photons at frequencies υ
1
and υ
2
combine to
generate a new photon at frequency υ
3
= υ
1
- υ
2
, the phase matching criteria becomes:
0 ) (
2 1 3
= − − = Δ k k k k
.

39
Phase matching ensures that the contributions at the new frequency along the
length of the medium add in-phase and therefore do not cancel out, similar to that of
fiber Bragg gratings. Phase mismatch results in a conversion loss of sinc
2
(ΔkL/2),
where L is the length of the medium [100]. Traditional phase matching in an
anisotropic crystal utilizes the presence of a birefringence such that n
(2ω)
= n
(ω)
(for
the case of SHG), thereby requiring signal and pumps to be aligned to the ordinary
and extraordinary axes [100].

Figure 2-14: Illustration of quasi-phase matching in a periodically poled
Lithium Niobate waveguide.
Quasi-phase matching (QPM) allows for efficient phase matching over long
device lengths through a periodic sign reversal of the nonlinear susceptibility at odd
multiples of the coherence length (2π/Δk), shown in Figure 2-14. QPM in LiNbO
3

also allows for the use of the largest nonlinear coefficient, d
33
(which is
approximately seven times that of d
31
), along with co-polarized pumps and signals.
That is, QPM does not require crystal birefringence to obtain phase matching.
Instead of requiring Δk=0, the matching criteria for QPM is instead:
0 ' = Δ − Δ = Δ
G
K k k

Reverse susceptibility once every coherence length
(Grating period, Λ = 2*l
c
)
χ -χ χ -χ χ -χ χ -χ χ -χ χ -χ
0 1 2 3 4 5 6 7 8 9 10
Distance (normalized to coherence length)
Quasi-phase matched
No matching

40
where ΔK
G
=2π/Λ and Λ is the grating period. In the simplest case, the grating period
is set to twice the coherence length.

Figure 2-15: Illustration of difference frequency generation in a periodically
poled Lithium Niobate waveguide.
Wavelength conversion in a PPLN waveguide is typically accomplished using
DFG, as illustrated in Figure 2-15 [50]. An input pump at frequency
p
f mixes with
an input signal at frequency
s
f to generate a spectral copy at
c
f , i.e.
s p c
f f f − = . In
order to satisfy the phase matching criteria it is required that
c
f and
s
f be symmetric
with respect to 2 /
p
f , where 2 /
p
f coincides with the quasi-phase matching
wavelength of the PPLN waveguide. The output field of the converted signal,
∗
∝
s p c
E E E and is therefore proportional to the complex conjugate of the input field.
Despite this process being phase conjugating, the relative phase variations of the
converted signal are directly dependent on the relative phase of the input signal and
the pump. If the pump is a CW source with relatively constant phase, the converted
signal will contain all of the phase information of the original signal making the
Quasi-phase match
wavelength (~1550 nm)
Pump
p
f
Input
Signal
Converted
Signal
s p c
f f f − =
s
f
c
f
2
p
f

41
DFG process phase preserving. This property allows the DFG process to support the
conversion of phase based signals and advanced modulation formats.
To support C-band wavelength conversion, the QPM wavelength of the
waveguide is typically in the 1550 nm region. Note that the QPM wavelength is
temperature sensitive with approximately 10 nm tuning capability over the allowable
range of the device. The pump wavelengths required for DFG within the C-band
therefore lie close to 775 nm. An alternative to using a 775 nm pump is to use a
cascaded wavelength conversion scheme, in which a C-band pump is first converted
down to 775 nm using either SHG or SFG.

Figure 2-16: Illustration of cascaded wavelength conversion schemes in a
PPLN waveguide. Second harmonic generation is followed by difference
frequency generation
Cascaded SHG-DFG is illustrated in Figure 2-16. The input pump is located at
the QPM wavelength of the waveguide. Second harmonic generation leads to a new
pump located at a frequency
p
f 2 . This second harmonic then serves as the pump in
the DFG process between frequencies
p
f 2 and
s
f . The result is the same as that
described above for DFG alone, but only requires pumps located within the C-band.
As SHG generates a pump with a field 2
p s
E E ∝ the phase relationship is not
Quasi-phase match
wavelength (~1550 nm)
Pump 2
nd
harmonic
p
f 2
Input
Signal
Converted
Signal
s p c
f f f − = 2
s
f
c
f
Pump
p
f

42
preserved for the case of the pump. However, the final signal at 2
c p s
f f f = − is still
transparent to the phase of the input signal. Data on at the signal frequency will be
transparently converted preserving phase-based modulation formats.

Figure 2-17: Illustration of cascaded wavelength conversion in a PPLN
waveguide. Sum frequency generation is followed by difference frequency
generation.
Cascaded SFG-DFG can also be utilized as illustrated in Figure 2-17. This
scheme is similar to the SHG-DFG scheme except the first process involves a mix
between two C-band pumps,
1 p
f and
2 p
f located symmetric relative to the QPM
wavelength of the waveguide. Sum frequency generation of the pumps results in a
new pump at
2 1 p p p
f f f + = . The new pump then mixes with the input signal
s
f to
create a phase conjugated copy at frequency
s p c
f f f − = . As the SFG process is
phase maintaining, the generated pump contains the sum of the phases of the input
pumps. This process can be used to add the phases of two binary phase signals into a
quadrature phase signal, or to transparently move a phase based signal to the pump
frequency. Combined with DFG, a phase based signal can be transparently converted
in wavelength without conjugating the spectrum.
Quasi-phase match
wavelength (~1550 nm)
Pump
2 1 p p
f f +
Input
Signal
Converted
Signal
s p p c
f f f f − + = ) (
2 1
s
f
c
f
Pump-2
1 p
f
Pump-1
1 p
f

43
2.8 Four-Wave-Mixing In Highly Nonlinear Fiber
Four-wave-mixing can be understood by considering the third order polarization
term in response to an applied electric field [3]:
E E E P
o
NL :
) 3 (
χ ε =

If we consider four optical waves involved in the nonlinear interaction with
frequencies ω
1
, ω
2
, ω
3
and ω
4
, the total E-field can be written as [3]:
( ) [ ] . . exp ˆ
2
1
4
1
c c t w z k i E x E
j
j j j
+ − =
∑
=

where it has been assumed that all waves are propagating linearly polarized along the
same axis, propagating in the same direction with propagation constants k
j
=n
j
ω/c.
Inserting the total E-field into NL P results in a large number of terms. Following the
work of [3], P
4
can be expanded as:
( )
[ ] [ ]
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 3 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 2 1
4 4 4 4 4 3 4 4 4 4 4 2 1 4 4 3 4 4 2 1
FWM
t z k k k k j o t z k k k k j o
XPM
o
SPM
o
e E E E e E E E
E E E E E E P
.....
4
6
4
6
4
6
4
3
) ( ) ( *
3 2 1
) 3 ( ) ( ) (
3 2 1
) 3 (
4
2
3
2
2
2
1
) 3 (
4
2
4
) 3 (
4
4 3 2 1 4 3 2 1 4 3 2 1 4 3 2 1
+ + +
+ + + + =
− − + − − − + − + + − − + + ω ω ω ω ω ω ω ω
χ
ε
χ
ε
χ
ε
χ
ε

The first term is referred to as self-phase modulation (SPM). SPM results in a
frequency or phase chirp proportional to the instantaneous intensity of the signal
itself (i.e. it does not involve an interaction or mixing with other spectral
components). SPM can lead to un-intentional spectral broadening when launch

44
powers are too high. It is also enables many useful applications such as super-
continuum generation and pulse compression.
Cross phase modulation (XPM) is similar to SPM but instead involves a
frequency chirping of the E-field proportional to the intensity of other spectral
components. XPM is a major concern in WDM multi-channel transmission systems
where phase distortion and spectral broadening can be imposed on the signal of
interest due to the high intensity of other WDM signals propagating within the fiber.
Similarly, XPM also enables many applications, such as wavelength conversion,
switching and optical demultiplexing.
There are many possible mixing terms as a result of the above expansion.
However, only terms satisfying the phase matching condition will lead to the
creation of a photon at a new frequency. Specifically, phase matching requires
conservation of energy and momentum. Energy conservation requires
4 3 2 1
hv hv hv hv = + + or
4 3 2 1
ω ω ω ω = + + in the first FWM term above and
4 3 2 1
hv hv hv hv + = + or
4 3 2 1
ω ω ω ω + = + in the second FWM term above. To satisfy
momentum conservation,
3 2 1 4
k k k k k − − − = Δ for the first term and
2 1 4 3
k k k k k − − + = Δ for the second term be equal to zero as well. The propagation
constants are given by k
j
= n
j
ω/c = 2π n
j
/λ, making efficient FWM highly dependent
on the dispersion profile of the medium as a function of frequency.

45

Figure 2-18: Illustration of four-wave-mixing processes that satisfy the phase
match condition.
For a fixed zero dispersion wavelength (ZDW), phase matching with respect to
that wavelength is possible. Figure 2-18, shows the first order phase matching
conditions for four-wave-mixing. As the phase matching condition (symmetry about
the ZDW) is broken, the efficiency of the FWM process declines. For dispersion
flattened non-linear fibers, the 3 dB decline in FWM efficiency can exceed several
100 nm’s of bandwidth. Furthermore, FWM can be performed using a single pump,
known as degenerate FWM, to satisfy only the Modulation Instability (MI) relation.
With the exception of Modulation Instability all four-wave-mixing processes are
transparent to phase. This makes them ideal for the conversion and manipulation of
phase based formats. In the case of MI, the relationship between the pump
wavelength and the converted wavelength is not phase maintaining. However, the
signal to converted relationship is phase transparent. This makes MI a good choice
for simple wavelength conversion of signals containing advanced modulation
formats.
2 1 2 conv pump pump sig
f f f f = + −
sig
f
1 conv
f
1 pump
f
*
2 conv
f
2 pump
f
*
3 conv
f
Phase Conjugation
1 1
2
conv pump sig
f f f = −
Modulation Instability
3 2 1 conv pump sig pump
f f f f = + −
Bragg Scattering
ZDW

46
Chapter 3:
All Optical Multiplexing and Demultiplexing
of 100 Gbit/s Pol-MUX Signals

This chapter presents the first of two investigations on the all-optical processing
of polarization multiplexed (Pol-MUX) signals. A single 100 Gbit/s Pol-MUX signal
is decomposed into two 50 Gbit/s WDM channels for easy processing. Similarly two
50 Gbit/s WDM channels are combined into a single 100 Gbit/s Pol-MUX channel.
In this manner, WDM channels can be combined into a single Pol-MUX channel to
double their spectral efficiency while a Pol-MUX channel can be demultiplexed for
use in a traditional WDM network or for independent routing or processing of its
sub-channels without the need for detection and electronic processing.
3.1 Introduction
Polarization represents a key domain of the optical wave that can be readily
exploited. A straightforward example is the recent interest in optical transmission
systems that use a polarization-multiplexed (Pol-MUX) data channel [33]. Such Pol-
MUXed transmission doubles the system spectral efficiency (bits/s/Hz) and is more
tolerant to fiber dispersive effects [74]. When combined with advanced modulation
formats, quadrature phase-shift-keying (QPSK) for example, high-bandwidth,
spectrally efficient transmission is possible [33].

47
Pol-MUXed channels are fairly easy to generate at a transmitter and decompose
at a receiver, since simple polarization-dependent couplers/splitters are available
[10]. However, polarization multiplexing and demultiplexing of high-data-rate
channels at an intermediate routing node in a network would be quite complex but
could bring enhanced throughput and performance. The complexity comes from the
fact that two data channels might originate on two different WDM wavelengths, such
that simple components are insufficient. Some type of wavelength conversion and
polarization manipulation may be required to allow for WDM-to-Pol-MUX and Pol-
MUX-to-WDM conversion. Furthermore, this conversion process should be phase
transparent to allow for advanced formats.
High-speed optical (de)multiplexing has been reported for the wavelength, time,
and phase domains [60, 98, 54, 103]. However, there has been little reported in terms
of optically (de)multiplexing two channels (from)/into a single channel in the
polarization domain. In this chapter, we discuss the experimental demonstration of
all-optical polarization multiplexing and polarization demultiplexing between two
50-Gbit/s channels and a single 100-Gbit/s channel. Orthogonal pumps in a
bidirectional highly nonlinear fiber (HNLF) loop are used to both demultiplex a
single 100-Gbit/s Pol-MUX channel into two 50-Gbit/s WDM channels and
multiplex two 50-Gbit/s WDM channels into a single 100-Gbit/s Pol-MUX channel.
Both On-Off-Keying (OOK) and Differential Phase-Shift-Keying (DPSK) are shown
with no measurable penalty for demultiplexing and <2 dB penalty for multiplexing.

48
3.2 Concept
Shown in Figure 3-1 is a conceptual diagram of our technique. Two orthogonally
polarized pumps, P
1
, P
2
, are separated in wavelength and used to independently
operate on each of the polarization channels of a Pol-MUX signal. Using four-wave-
mixing (FWM) in HNLF, the X- and Y-polarized channels, S
1
and S
2
, are
demultiplexed and converted to two new wavelengths resulting in the WDM
channels S’
1
and S’
2
. The reverse process is also available where the two WDM
signals, S’
1
, S’
2
, are converted and multiplexed to a single wavelength except on
orthogonal polarizations. In this manner, WDM channels can be combined into a
single Pol-MUX channel to double their spectral efficiency while a Pol-MUX
channel can be demultiplexed for use in a traditional WDM network or for
independent routing or processing of its sub-channels.

Figure 3-1: Conceptual diagram of all optical polarization demultiplexing and
polarization multiplexing.
3.3 Experimental Setup
An experimental block diagram of our setup is shown in Figure 3-2. A Mach-
Zehnder modulator (MZM) driven by a 50-Gbit/s PRBS 2
31
-1 data stream is used to
generate both On-Off-Keying (OOK) and differential Phase-Shift-Keying (DPSK)
λ λ λ λ
x
y
S
1 S’
1
S
2
S’
2 P
2
P
1
Pol-Mux
WDM
All-Optical
Pol-DeMux
All-Optical
Pol-Mux

49
signals. Full-rate pulse carving is used to generate both 50% return-to-zero (RZ)
waveforms generating RZ-OOK and RZ-DPSK. Polarization multiplexing is
achieved by splitting the signal into two paths, de-correlating with a fiber delay, and
recombining the signals using a polarization beam combiner/splitter (PBC/PBS) to
ensure orthogonality. For the case of WDM channels, two CW sources are combined
and modulated before being split and independently amplified, filtered, and de-
correlated. Four-wave-mixing is achieved in a bidirectional fiber loop consisting of
330 meters of highly nonlinear fiber (HNLF) with a zero dispersion wavelength of
1562 nm, a PBS, a polarization controller set to rotate 90°, and a circulator to
separate the data moving into and out of the loop. The incoming channels and pumps
are aligned to the PBS such that the X and Y polarization travel in opposite
directions around the loop. One channel and one pump travel in each direction. After
one pass around the loop, the pumps, signals, and converted channels are recombined
by the PBS and routed by the circulator to the receiver. The appropriate wavelength
is filtered off and sent to a pre-amplified receiver for bit-error-rate (BER)
measurements. A PBS is used to separate the Pol-MUX data channels for detection
while a delay-line interferometer with a 20 ps delay and balanced photo-receiver are
used for DPSK detection.

50

Figure 3-2: Experimental setup. A 100 Gbit/s Pol-MUX signal is generated and
combined with two orthogonal pumps for demultiplexing to two 50 Gbit/s WDM
channels. Similarly, two 50 Gbit/s WDM channels are generated and combined
with two orthogonal pumps for multiplexing into a single 100 Gbit/s Pol-MUX
channel.
For the case of demultiplexing 100-Gbit/s Pol- MUX to two 50-Gbit/s WDM
channels, the signal, λ
Sig
≈ 1555 nm, is combined with two pumps, λ
P1
≈ 1560 nm
and λ
P2
≈ 1562 nm. The signal and pumps are independently amplified and filtered
with 2 nm bandpass filters to suppress the ASE noise from the amplifiers. The
polarization of the Pol-MUX signal is adjusted so that the X-polarized and Y-
Polarized channels are separated and sent in opposite directions around the bi-
directional FWM setup. The pumps are also polarized such that λ
P1
travels in the
same direction as the X-polarization while λ
P2
travels with the Y-polarization. Two
converted signals are generated via degenerate FWM at λ
X
= 2×λ
P1
– λ
Sig
≈ 1565 nm
and λ
y
= 2×λ
P2
– λ
Sig
≈ 1569 nm respectively. This degenerate FWM process is phase
conjugating, but it is also inherently phase maintaining allowing for both OOK and
DPSK data to be demultiplexed to two different WDM channels, top of Figure 3-3(a)
and (b).
λ
Sig2 Pol-Mux Transmitter
TDL
Att
EDFA
HNLF
PBS
λ
P1
90°
Rotation
λ
P2
EDFA
λ
Sig1
WDM Only
50 Gb/s RZ-OOK
and RZ-DPSK Tx
WDM Only
T
-
DPSK Only
OOK Only
Pol-Mux Only

51
For the reverse case of multiplexing two 50-Gbit/s WDM channels to a single
100-Gbit/s Pol-MUX channel, two WDM signals, λ
Sig1
≈ 1565 nm and λ
Sig2
≈ 1569
nm are combined with the two pumps at λ
P1
and λ
P2
respectively. Again the
polarization of the signals and pumps are aligned so that only one signal and one
pump travel in the same direction around the loop. This time, degenerate four-wave-
mixing produces two converted signals at the same wavelength, λ
MUX
= 2×λ
P1
– λ
Sig1

= 2×λ
P2
– λ
Sig2
≈ 1555 nm. The two converted signals are also orthogonally polarized
resulting in a single 100-Gbit/s Pol-MUX channel at λ
MUX
. Again the phase
transparent nature of the FWM process allows for both OOK and DPSK data to be
multiplexed into a single Pol-MUX channel, bottom of Figure 3-3(a) and (b).

(a) (b)
10dB/D
2nm/D
λSig
λP1 λP2
λx λy
10dB/D
2nm/D
λP1
λP2
1551.15 nm 1561.15 nm 1571.15 nm
λMUX
λSig1 λSig2
10dB/D
2nm/D
λSig
λP1 λP2
λx λy
λP1
λP2
1551.15 nm 1561.15 nm 1571.15 nm
λMUX
λSig1 λSig2
10dB/D
2nm/D

52

(c)
Figure 3-3: (a) Experimental spectra for demultiplexing (top) and multiplexing
(bottom) between 100 Gbit/s Pol- MUX RZ-OOK and 2 x 50 Gbit/s RZ-OOK. (b)
Experimental spectra for the demultiplexing (top) and demultiplexing (bottom)
when using RZ-DPSK. (c) Back-to-Back eyes and demultiplexed/multiplexed eyes
for comparison.
3.4 Results and Discussion
Figure 3-3(c) shows a comparison of the transmitted or back-to-back eye
diagrams with the eye diagrams after demultiplexing and multiplexing when using
RZ-OOK and RZ-DPSK. Demultiplexing of RZ-OOK and RZ-DPSK are shown in
the first two columns respectively. Multiplexing of RZ-OOK and RZ-DPSK are to a
single channel are shown in column three. Note that the Pol-MUX sub-channels are
offset by half of a bit-time for better comparison of the multiplexed eye-diagrams.
B2B - 50G RZ-OOK
DeMUXed – Pol X
B2B 100G PolMux
50G RZ-DPSK
100 G PolMux to WDM 2x50 G WDM to 100 G PolMux to WDM
B2B - 50G RZ-DPSK
DeMUXed – Pol Y
DeMUXed – Pol X
DeMUXed – Pol Y
PolMUXed – RZ-OOK
PolMUXed – RZ-DPSK (AMI)
RZ-OOK RZ-DPSK 100 G PolMux

53

(a) (b)

(c)
Figure 3-4: (a) Comparison of RZ-OOK performance after demultiplexing and
multiplexing compared to back-to-back performance. (b) Comparison of RZ-
DPSK performance after demultiplexing and multiplexing compared to back-to-
back. (c) Performance of the multiplexed 100-Gbit/s Pol-MUX channel after 1km
of uncompensated propagation.
Bit-error-rate (BER) curves were used to assess the performance of both the
demultiplexing and multiplexing functions. Figure 3-4(a) shows the performance of
RZ-OOK compared to the back-to-back case. For demultiplexing, of the 100-Gbit/s
Pol-MUX channel, <0.5 dB penalty is observed for either converted channel.
Multiplexing of the two WDM channels into a single 100-Gbit/s Pol-MUX resulted
in a penalty of <2 dB. Similarly, Figure 3-4(b) shows the performance of RZ-DPSK.
-45 -40 -35 -30 -25 -20
2
3
4
5
6
7
8
9
Received Power (dBm)
-Log10(BER)
B2B
PolMux
To WDM
WDM To
PolMux
RZ-OOK Performance
-45 -40 -35 -30 -25 -20
2
3
4
5
6
7
8
9
Received Power (dBm)
-Log10(BER)
B2B
PolMux
To WDM
WDM To
PolMux
RZ-DPSK Performance
-45 -40 -35 -30 -25 -20
2
3
4
5
6
7
8
9
Received Power (dBm)
-Log10(BER)
Transmission Performance

54
As expected, RZ-DPSK performed ~3 dB better than RZ-OOK due to the use of
balanced detection. Pol-MUX to WDM demultiplexing showed penalties of 0.4 and
0.9 dB for the X and Y polarization respectively. Again, an additional ~ 2 dB penalty
was observed for the case of WDM to Pol-MUX multiplexing. We believe this
penalty is due to the finite extinction ratio of the polarizing beam combiners which
may result in a small amount of power from the orthogonal data channel being
converted onto the other polarization.
To further assess the quality the multiplexed 100-Gbit/s Pol-MUX channel, it
was propagated over 1km of uncompensated SMF-28 fiber. For comparison, the
back-to-back signal was also propagated over the same distance. Both the back-to-
back and the multiplexed signal experienced an additional ~1.5 dB penalty following
propagation as shown in Figure 3-4(c).

55
Chapter 4:
λ-Conversion of 160-Gbit/s PDM 16-QAM Using a
Single Periodically-Poled Lithium Niobate Waveguide

This chapter presents the second investigation on the all-optical processing of
polarization multiplexed (Pol-MUX) signals. We experimentally demonstrate
wavelength conversion of both single polarization 40-Gbuad 16-QAM and 20-Gbuad
polarization-division-multiplexed (PDM) 16-QAM in a PPLN waveguide. A
polarization insensitive scheme utilizing bi-directional operation is employed. The
power penalty as a function of pump power is also investigated and a minimum
conversion penalty of ~0.5 dB is obtained.
4.1 Introduction
The recent emergence of higher-order modulation formats to enable spectrally
efficient, high data-rate transmission has placed an increasing emphasis on the
complexity of transmitter and receiver electronics [74]. In particular, quadrature
amplitude modulation (QAM) has shown great potential for increasing the spectral
efficiency and enabling high data-rate transmission [95]. Optical signal processing,
which can operate on an entire data channel without “touching” each individual bit,
may help reduce this electronic requirement by performing basic signal processing
functions in the time domain [99]. One of the basic optical signal processing
functions is nonlinear wavelength conversion with possible applications for routing,

56
switching, and general data grooming in optical networks. An ideal wavelength
converter may include the ability to i) be data-rate independent, ii) transparent to
both the amplitude and phase of the incoming signal, and iii) transparent to the
polarization of the incoming signal.
The first attribute of data-rate transparency can be accomplished in a variety of
optical materials that support ultra-fast nonlinear interactions, including periodically-
poled Lithium Niobate (PPLN) and highly nonlinear fiber (HNLF). The second
attribute of phase and amplitude transparency can also be accomplished by operating
the nonlinear element in a linear conversion region, however, as the number of phase
and amplitude levels increases, the linearity required also increases and the region of
operation may shrink [99, 12]. The final attribute of polarization transparency is of
increasing importance with the use of polarization division multiplexing (PDM) as a
straight forward method to double the spectral efficiency [1].
Recent demonstrations of higher-order modulation format conversion and
polarization insensitive wavelength conversion have included: i) wavelength
conversion of OFDM at 10 Gbit/s in a PPLN waveguide [99], ii) polarization
insensitive wavelength conversion of 100-Gbit/s quadrature phase-shift-keying
(QPSK) in a PPLN waveguide [42], iii) wavelength conversion of polarization
multiplexed 100-Gbit/s on-off-keying (OOK) and differential phase-shift-keying
(DPSK) in HNLF[65], iv) wavelength conversion of 20-Gbit/s ASK-DPSK in HNLF
[30], and v) polarization insensitive wavelength conversion of 114-Gbit/s 8-ary PSK

57
in HNLF [106]. Each of these demonstrations utilized a relatively small number of
amplitude and phase levels while many are not polarization transparent. A laudable
goal would be the demonstration of PDM 16-ary QAM (PDM 16-QAM) in which 16
unique phase and amplitude levels exist imposing a tight requirement on the linearity
of the wavelength conversion process.
In this chapter, the wavelength conversion of 40-Gbaud single polarization and
20-Gbaud PDM 16-QAM, both with an aggregate data-rate of 160 Gbit/s, is
experimentally shown. A single PPLN waveguide in a bidirectional configuration is
used to achieve polarization diverse operation [56]. A power penalty of ~0.5 dB is
obtained for optimal bias conditions of the PPLN waveguide. Power penalty
measurements for varying pump powers show the window of linear operation for
which transparent phase and amplitude conversion is achieved.
4.2 Concept
The conceptual block diagram of this technique is shown in Figure 4-1. A
polarization multiplexed signal and continuous wave (CW) pump aligned at 45°
linear relative to the signal are inserted into the bi-directional wavelength converter
through a circulator. A polarizing beam splitter/combiner (PBS/PBC) separates the
signal into the X and Y polarizations while the CW pump is split between the two
polarizations. The X-polarization traverses the loop counterclockwise following the
path of the blue arrows.

58
(a)
(b)
Figure 4-1: Conceptual diagram of transparent polarization (a) and phase and
amplitude (b) conversion in a PPLN waveguide.
It first undergoes a 90° rotation so that the pump and the signal are aligned to the
Y-polarization axis before passing through the PPLN. After passing through the
PPLN, a copy of the data amplitude and phase is generated according to the cascaded
second harmonic generation followed by difference frequency generation
(SHG/DFG), f
S’1
= 2*f
P
– f
S1
, Figure 4-1(b). Similarly, the Y-Polarization travels in
the opposite direction (clockwise) as shown by the purple arrows. It first passes
through the PPLN to generate the wavelength converted copy of its data before being
rotated by 90° so that it can pass backwards through the PBC. After passing back
through the polarizing beam combiner, the recombined signal is directed out by the
circulator. This output signal now contains the wavelength converted copy of the
original Pol-MUX signal. In general, the data signal need not be aligned in
λ λ λ λ
x
y
λ λ λ λ
x
y
PPLN
PBS
90°
Rotation
λ λ λ λ
x
y
S 1
S 2
P 1
λ λ λ λ
x
y
λ λ λ λ
x
y
λ λ λ λ
x
y
λ λ λ λ
x
y
λ λ λ λ
x
y
S 2
S’ 2
S 1 S’ 1
λ λ λ λ
S
1
S’
1
SHG
DFG
QPM
f
S’1
= 2*f
P
– f
S1

P

59
polarization and the setup yields polarization insensitive wavelength conversion due
to its inherit polarization diversity [56].

Figure 4-2: Experimental setup. A 20-Gbaud PDM 16-QAM or 40-Gbaud 16-
QAM signal is generated and combined with a CW pump in a bidirectional PPLN
based wavelength converter. An EAM is used for 40-Gbaud to 20-Gbaud down
sampling followed by coherent detection.
4.3 Experimental Setup
An experimental block diagram of our setup is shown in Figure 4-2. A laser at
λ
Sig
≈1544.9 nm is first modulated with a QPSK waveform at the desired rate, 40 or
20 Gbaud. An integrated free-space multiplexer consisting of i) an optical split, ii) a
1-bit delay and 6-dB attenuation in one arm, and iii) recombining of the two arms, is
used to convert the QPSK signal to 16-QAM [59]. For 20 Gbaud PDM 16-QAM,
polarization multiplexing is achieved by splitting the signal into two paths, de-
correlating with a fiber delay, and recombining the signals using a polarization beam
combiner/splitter (PBC/PBS) to ensure orthogonality. Wavelength conversion is
achieved in a bidirectional loop consisting of a PPLN waveguide with a quasi-phase-
matching (QPM) wavelength of 1549.9 nm, a PBS, a polarization controller set to
rotate 90°, and a circulator to separate the data moving into and out of the loop. The
For Pol-Mux Only
TDL
Att
EDFA PPLN
PBS
λ
P
90°
Rotation
EDFA
λ
Sig
20 – 40 Gbaud
16-QAM Tx
45° linearly polarized
20 GHz
Coherent Rx
EAM
For 40 Gbaud Only

60
incoming CW pump at λ≈1549.9 is aligned 45° linearly polarized relative to the PBS
such that the half of its power travels in each direction around the loop. Second
harmonic generation (SHG) followed by difference frequency generation (DFG) are
used to wavelength convert the signal ~10 nm longer in wavelength with an optical
efficiency of -16 dB. After one pass around the loop, the pump, signal, and converted
channels are recombined by the PBS and routed by the circulator to the receiver. The
converted wavelength at λ≈1554.9 nm is filtered off and sent to a pre-amplified
coherent Rx for signal processing and bit-error-rate (BER) measurements. Since our
analog-to-digital converter was limited to ~20 GHz, an electro-absorption modulator
(EAM) was used to down sample the 40-Gbaud 16-QAM signal to 20 Gbaud, prior
to detection.
4.4 Results and Discussion
The experimental constellation diagrams and spectra are shown in Figure 4-3.
For 40-Gbaud single polarization 16-QAM, Figure 4-3(a), optimal conversion was
obtained for a pump power of 16.2 dBm in both directions around the loop. By
increasing the pump power the wavelength converter begins to move into the
depletion regime of operation and we start to see amplitude saturation of the
converted signal, Figure 4-3(a) (bottom right). The highest amplitude level,
corresponding to the four outside corners of the constellation, is compressed inward
toward the second amplitude level, giving the constellation a rounded look [4].
Figure 4-3(b) shows the experimental constellations for 20-Gbaud PDM 16-QAM

61
where optimal performance was obtained for pump powers of ~17 dBm in each
direction. Figure 4-3(c) shows the experimental spectra for the 40-Gbaud single
polarization (right) and 20-Gbaud dual polarization (left).

(a) (b)

(c)
Figure 4-3: (a) Experimental constellation diagrams for 40-Gbaud single-
polarization 16-QAM back-to-back (top) and after (bottom) conversion for pump
powers of 16.2 dBm (left) and 21 dBm (right). (b) Experimental constellation
diagrams for 20-Gbaud PDM 16-QAM back-to-back (top) and after (bottom)
conversion for both polarizations. (c) Experimental spectra for 40-Gbaud single
polarization (right) and 20 G-Baud PDM (left) 16-QAM.
Bit-error-rate (BER) measurements were performed on the captured waveforms
to assess the system performance. Figure 4-4(a) and (b) show the performance of 40-
Gbaud single polarization 16-QAM and 20-Gbaud PDM 16-QAM respectively. Less
than a 0.5 dB penalty for 40-Gbaud and ~0.6 dB penalty for 20-Gbaud exist for the
wavelength converted signal at a BER of 1E
-3
. Figure 4-4(c) shows the measured
B2B
21 dBm 16.2 dBm
X-B2B Y-B2B
X-17 dbm Y-17 dBm
10dB/D
2.5nm/D
λSig
λP
1537.4 nm 1549.9 nm 1562.4nm
λCon
10dB/D
2.5nm/D
λSig
λP
1537.4 nm 1549.9 nm 1562.4nm
λCon

62
relative power penalty for 40-Gbaud 16-QAM as a function of the pump power. The
power penalty increases rapidly as signal depletion begins to occur and the
constellation amplitude levels become compressed. Increasing the signal power may
prevent depletion and allow for the use of a higher power pump to improve
conversion efficiency, but a higher power EDFA will be required and may reduce the
signal-to-noise ratio of the system.

(a) (b)

(c)
Figure 4-4: Bit-error-rate (BER) measurements for (a) 40-Gbaud 16-QAM and (b)
20-Gbaud PDM 16-QAM back-to-back and after wavelength conversion. (c)
Relative power penalty vs. pump power for 40-Gbaud 16-QAM.
1E-5
1E-4
1E-3
1E-2
1E-1
Bit-Error-Rate
Power Received (dBm)
-40 -35 -30 -25 -20
Back-to-Back (1544.9)
Converted (1554.9)
1E-5
1E-4
1E-3
1E-2
1E-1
Bit-Error-Rate
Power Received (dBm)
-40 -35 -30 -25 -20
Back-to-Back (1544.9)
Converted (1554.9)
0
1
2
3
4
5
6
Relative Power Penalty (dB)
Pump Power (dBm)
12 14 16 18 20 22

63
Chapter 5:
503 ns Continuously Tunable Delay of 40 Gbit/s OOK
and DPSK with Improved Dispersion Compensation

In this chapter, a method for achieving a variable optical delay element using
wavelength conversion in HNLF, dispersion compensating fiber, and intra-channel
dispersion compensation is introduced. A delay of up to 503 ns is demonstrated
using 40 Gbit/s RZ-DPSK and 40 Gbit/s RZ-OOK modulation formats. The impact
of the delay element is quantified through bit error rate measurements. In the
following chapters, two new approaches for residual dispersion compensation in the
delay element, fine tuning of the delay element and an application of the delay are
introduced.
5.1 Introduction
Signal processing is a powerful enabler and performance enhancer for a host of
communications functions, and performing processing in the optical domain reduces
any optical-electronic conversion inefficiencies. For data rates ≥40-Gbit/s in which
the length/time scale of a single bit is fairly short, variable optical timing can be
critical in time multiplexing, synchronization, switching, equalization, correlation,
and time-slot interchange [76, 102, 23]. One of the most basic elements needed to
achieve accurate and high-throughput signal processing is a continuously tunable
optical delay line, and yet this element has historically been difficult to realize. Such

64
tunable delays may allow for compensation of delay variations caused by
temperature changes in the fiber and laser stability. Desirable features of such
tunable optical delays include the ability to handle high data-rates, be transparent to
data modulation formats, and accommodate a large range of delays to enable the
delay of large data blocks. Furthermore, given the present importance of highly-
sensitive differential-phase-shift-keying (DPSK) and spectrally-efficient differential-
quadrature-phase-shift-keying (DQPSK), modulation format transparency may be a
desirable feature for an optical delay element.
Various techniques have been used to demonstrate tunable optical delays,
including: (i) selecting among a discrete set of optical propagation paths, which
produces only a finite set of delays[81], (ii) using slow-light-based photonic
resonances, in which delays tend to be < 1 ns for Gbit/s signals [88], and (iii)
wavelength conversion coupled with chromatic dispersion [92, 17, 27, 6]. This last
technique uses tunable wavelength conversion combined with inter-channel
chromatic dispersion and intra-channel dispersion compensation. Recently published
results for conversion/dispersion include 1.8 μs for a 10 Gbit/s on-off keying (OOK)
signal and 1.5 μs for a 40 Gbit/s DPSK signal using parametric wavelength
conversion [5], 4.2 ns for a 3.5 ps pulse-train using self-phase modulation [71], 800
ps for a 10 ps pulse-train using parametric conversion [81], and 105 ns for a 40
Gbit/s OOK 40 Gbit/s DPSK 80 Gbit/s DQPSK signals using periodically-poled
Lithium Niobate (PPLN) waveguides [17].

65
A significant challenge in all these methods is the natural data degradation due to
intra-data-channel chromatic dispersion. Such dispersion can be: (i) 2
nd
order, which
is the dispersion value at a specific wavelength, and (ii) 3
rd
order, which is the slope
of the dispersion profile where the dispersion value varies with wavelength. As the
wavelength of the signal is varied to tune delay, it experiences a different amount of
dispersion at each wavelength. As delay values and data rates increase beyond ~100
ns and 40-Gbit/s, this residual dispersion may become a limiting factor in system
performance.
Various techniques have been described which attempt to partially address these
problems, including: (a) using a dispersion compensator with a value that can be
tuned [102, 17], and (b) phase conjugation in which one pass through the dispersive
element is performed at a tunable wavelength to control the delay while the other is
at a fixed wavelength for compensation [75, 81, 6, 70]. The first method requires a
dispersion compensator which must be widely tunable and may be difficult to
achieve for large dispersion values and optical bandwidths. The second method
requires the delay element to have a dispersion profile which is flat over the
bandwidth of interest, which may be difficult to achieve over large bandwidths.
In this chapter, early results of a new method for delay compensation are
presented. A 503 ns, tunable delay of 40-Gbit/s return-to-zero on-off-keying (RZ-
OOK) and return-to-zero differential phase-shift-keying (RZ-DPSK) signals using
optical phase conjugation in which the conjugated signal only sees a small

66
wavelength shift in-comparison to previous methods is achieved. This relatively
small wavelength shift allows for a reduction in residual dispersion of ~95% over
previous methods while doubling the relative delay for the same wavelength range
and amount of dispersion. This method was adopted and has been used to achieve the
more recent record results [5, 67].

Figure 5-1: Conceptual diagram of tunable delay methods. As opposed to
previous methods (a) and (b), following wavelength conversion (W/C) and
delay in dispersion compensating fiber (DCF), the signal is not returned to the
original wavelength for compensation (c).
5.2 Concept
Figure 5-1 shows a conceptual comparison between the previously mentioned
compensation methods (a, b) and this technique (c). In all cases, dispersion
compensating fiber (DCF) is used to generate a wavelength-dependent group delay
(inter-wavelength dispersion), in addition to generating unwanted intra-channel
dispersion. By varying the wavelength through the DCF, the signal experiences a
variable delay. In methods (a) and (b), a wavelength converter after the DCF is used
(a)
(b)
(c)
W/C
1
t
Dispersive
Element
1
Δτ
Δτ
Δτ
Tunable
Dispersion
Compensator
W/C
2
λ
in
λ
1
λ
2
λ
1
λ
2
λ
in
λ
in λ
in
λ
in
Optical
Phase
Conj.
1
W/C
1
t
Dispersive
Element
1
Dispersive
Element
1
Δτ Δτ
λ
in
λ
1
λ
2
λ
1
λ
2
λ
in
λ
in
λ
in
λ
in
Δτ
Optical
Phase
Conj.
1
W/C
1
t
Dispersive
Element
1
Dispersive
Element
1
Δτ
W/C
2
Δτ 2×Δτ 2×Δτ
λ
in
λ
1
λ
2
λ
1
λ
2
~λ
1
~λ
2
~λ
1
~λ
2
λ
in
λ
in

67
to convert the signal back to its original wavelength for compensation, either with
the use of a tunable dispersion compensator (a) or by phase conjugating the signal,
before it is passed through the DCF a second time (b). In (c), the signal is still
wavelength converted and phase-conjugated before being passed through the DCF a
second time. However, instead of converting the signal back to its original
wavelength as done in method (b), which can be a shift of several 10’s of nano-
meters or more, it is shifted by only a few nano-meters allowing the signal to retrace
its path at nearly the same variable wavelength. This relatively small wavelength
shift allows the signal to see almost the identical dispersion after being conjugated,
thus minimizing the residual dispersion. In comparison, previous method (b)
performed compensation at a fixed wavelength, resulting in a variable amount
residual dispersion as the delay value is changed. By allowing the compensation
wavelength to vary as the delay wavelength varies, our new method allows the
dispersion and compensation to be matched. This matching of the dispersion
compensation greatly reduces requirements on the dispersion profile of the
dispersive element.
Additionally, this method allows the return pass through the dispersive element
to also occur at a varying wavelength. The signal experiences a second wavelength-
dependent group delay, effectively doubling the amount of delay achieved. The total
delay (Δτ) through the system can now be determined by the product of the
wavelength conversion bandwidth (Δλ) and the dispersion (D): Δτ ≈ 2×Δλ×D, where

68
the factor of two comes from taking advantage of the relative group velocity
variation on both passes through the DCF.

Figure 5-2: Experimental setup. A 40-Gbit/s signal is wavelength converted
and passed through the Raman pumped DCF. The signal is then phase
conjugated and shifted by ~3.4 nm before passing back through the DCF for
detection.
5.3 Experimental Setup
An experimental block diagram of our setup is shown in Figure 5-2. RZ-OOK
and RZ-DPSK are generated using a Mach-Zehnder modulator (MZM) driven by a
40-Gbit/s PRBS 2
15
-1 data stream. Full-rate pulse carving is used to generate a 50%
return-to-zero (RZ) waveform. Tunable wavelength conversion is achieved using a
~100 m piece of dispersion-flattened highly nonlinear fiber (HNLF) with a ZDW of
~1550 nm and γ = 25 W
-1
·km
-1
. Four-wave-mixing (FWM) is used where in a high
power pump (14.6 dBm) is swept in wavelength to control the location of the
converted signal. The signal (λ
0
≈ 1532 nm, 14 dBm) is up-converted from 1535.7 to
1563.4 nm. A minimum spacing of 1.7 nm is used to allow for filter roll off. The
signal and pump powers are kept low to prevent stimulated Brillouin scattering, SBS.
MZM
λ
0
Pulse
Carver
CLK
Filter
1.2nm
EDFA
λ
P1
DCF
HNLF ≈
400m
Raman
Pumps
Raman
Pumps -9047
ps/nm
Rx
40 Gb/s
Data
HNLF ≈
400m
λ
P2
Filters
1.2nm EDFA
Dispersive
Element
1
Optical Phase
Conj.
1
W/C
1

69
As such, no parametric gain is observed and conversion efficiency varied with pump
spacing from -9 dB to -18 dB.
The converted signal is then filtered out and sent through the DCF (-9047 ps/nm
of dispersion) to impose a wavelength-dependent delay. Raman amplification in the
DCF is used to compensate for the 22 dB of fiber loss. Two 150 mW pumps at
~1450 nm and ~1460 nm are used to co- and counter-pump the fiber [3]. Following
the DCF, the delayed signal is phase conjugated using a second piece of ~300 m
HNLF with a zero-dispersion wavelength (ZDW) of ~1560 nm and γ = 25 W
-1
·km
-1

[3]. Again, FWM is used to convert the signal, ~1.5 dBm, across the pump, 14.6
dBm. Conversion efficiency is approximately -9 dB and no parametric gain is
obtained. This allows the signal to be phase conjugated and converted a relatively
small, ~3.4 nm in wavelength. In comparison, previous methods converted the signal
back to the original wavelength, a shift of several 10’s of nanometers or more. To
compensate for the intra-channel dispersion, the signal retraces its path through the
DCF. Since this return trip is at nearly the same wavelength, the conjugated signal
experiences both a similar group velocity and a similar dispersion profile, effectively
doubling the relative delay and minimizing residual dispersion. The third wavelength
conversion stage was not implemented here, but would be needed to make the
scheme wavelength transparent. Furthermore, this final wavelength conversion stage
can now operate on a non-dispersed signal, allowing for optical processing
techniques, including regeneration, to be employed.

70
(a) (b) (c)
Figure 5-3: Measured fiber dispersion profile for (a) dispersion, (b)
compensation and (c) comparison of residual dispersion for compensation at a
fixed wavelength and using the newly proposed method. Residual dispersion is
reduced by >95%.
5.4 Results and Discussion
Figure 5-3 shows the reduction in residual dispersion between the different
techniques. The measured fiber slope over the wavelength range has a variation of
nearly 900 ps/nm. Since this method compensates the signal at nearly the same
wavelength it was dispersed at, as opposed to a fixed wavelength, the residual
dispersion after compensation is greatly reduced. In our case, the residual dispersion
of the fixed wavelength compensation scheme exceeded ±440 ps/nm while our new
approach reduced the residual dispersion from +30 ps/nm to -20 ps/nm.
Polarization of the signal is adjusted for optimal wavelength conversion
efficiency; polarization-independent conversion schemes do exist and could be used
to alleviate this issue. The total polarization mode dispersion, PMD, of our fiber
(DCF + HNLF) is ~4.2 ps. To first order, PMD in general does not affect the
accumulated dispersion but can impact the wavelength conversion efficiency.
-8500
-8300
-8100
-7900
-7700
-7500
1530 1540 1550 1560 1570
Tuning
Range
Dispersion (ps/nm)
Wavelength (nm)
Dispersion
-8500
-8300
-8100
-7900
-7700
-7500
1530 1540 1550 1560 1570
Wavelength (nm)
Previous
New
Compensation
-500
-250
0
250
500
1530 1540 1550 1560 1570
Residual
Dispersion (ps/nm)
Previous
New
Wavelength (nm)

71

Figure 5-4: Measured delay versus converted wavelength and experimental
spectra of both the wavelength conversion and phase conjugation stages.
Shown in Figure 5-4 is the relative delay of the system as a function of the
converted wavelength measured with a 40-Gbit/s RZ-OOK signal. The total delay
achieved is 503 ns covering 27.7 nm from 1535.7 to 1563.4 nm. Since the phase
conjugated signal cannot be placed arbitrarily close to the pump due to filter
restrictions, an extra ~3.4 nm of bandwidth is required that does not contribute to the
delay. Figure 5-4 also shows the experimental spectra for the maximum delay
condition. The delay can be tuned by varying the wavelengths of the two pumps
while the amount of residual dispersion can be tuned by changing the distance of
phase conjugation.
Shown in Figure 5-5(a) and (b) is a comparison of the back-to-back performance
to the performance following the first wavelength conversion (Stage 1) and after the
full delay system (Final) for 40-Gbit/s RZ-OOK and RZ-DPSK. Figure 5-5(c) and
(d) show the performance for minimum, middle, and maximum delay values. At
BERs of 10
-9
, OOK and DPSK have maximum power penalties of 3.7 and 2.9 dB,
Converted Wavelength (nm)
D elay (n s)
1535 1565
0
100
200
300
400
500
600
503 ns
P h a se
C o n ju g atio n
λcon
1563.4 nm
λpump
1565.3 nm
λconj
1567.2 nm
λ-C o n v ersio n
λcon
1563.4 nm
λsig
1534.6 nm
λpump
1550 nm
10dB/D
3nm/D
8dB/D
1nm/D
1550

72
respectively. DPSK performed only 2 dB better than OOK, rather than the expected
3 dB, due to transmitter and receiver limitations. We believe the “flaring” of the
BER curves is due to the use of cascaded filters throughout the delay module and
limited amplifier gain bandwidths.

(a) (b)

(c) (d)
Figure 5-5: Bit-error-rate (BER) curves for (a) 40-Gbit/s RZ-OOK and (b) RZ-
DPSK. Performance after back-to-back, the first wavelength conversion (Stage
1), and after the full system (Final) is compared. 40-Gbit/s RZ-OOK (c) and
RZ-DPSK (d) back-to-back performance compared to the minimum,
maximum, and middle delay performances.

-40 -38 -36 -34 -32 -30 -28 -26 -24 -22
2
3
4
5
6
7
8
9
Back-to-Back
Stage1 - 1563.4
Final - 1563.4
Received Power (dB)
-Log10(BER)
B2B
Stage 1
Final
40 Gb/s RZ-OOK
-40 -38 -36 -34 -32 -30 -28 -26 -24 -22
2
3
4
5
6
7
8
9
Back-to-Back
Stage1 - 1557.0
Final - 1557.0
Received Power (dB)
-Log10(BER)
B2B
Stage 1
Final
40 Gb/s RZ-DPSK
Received Power (dB)
-Log10(BER)
40 Gb/s RZ-OOK
6
7
8
9
-40 -38 -36 -34 -32 -30 -28 -26 -24 -22
2
3
4
5
Back-to-Back
1535.7 - Min.
1542.1 - Middle
1563.4 - Max.
1535.7
nm
1542.1
nm
1563.4
nm
-40 -38 -36 -34 -32 -30 -28 -26 -24 -22
2
3
4
5
6
7
8
9
Received Power (dB)
-Log10(BER)
1535.7
nm
1557.0
nm
1560.4
nm
40 Gb/s RZ-DPSK
Back-to-Back
1535.7 - Min.
1557.0 - Middle
1563.4 - Max.

73
Chapter 6:
1.16 μs Continuously Tunable
Delay of 100 Gbit/s DQPSK

In this chapter, we demonstrate a tunable optical delay element using wavelength
conversion in a highly nonlinear fiber, dispersion compensating fiber, and optical
phase conjugation. A continuous delay of up to 1.16-μs equaling >55,000 symbols at
50 Gbit/s, for 100 Gbit/s NRZ-DQPSK and 50 Gbit/s NRZ-DPSK modulation
formats, is demonstrated.
6.1 Introduction
Continuously tunable optical delays have the potential to enable highly efficient
and high-speed optical data signal processing. Furthermore, given the present
importance of spectrally-efficient differential-quadrature-phase-shift-keying
(DQPSK) and the forthcoming IEEE 100 Gbit/s Ethernet Standard, 802.3ba, which
includes provisions for many potential formats [74], desirable characteristics of a
tunable optical delay line may include: (a) transparency to the data modulation
format, (b) high-speed operation, and (c) the ability to accommodate a large range of
delays.
One promising method of creating optical delays is conversion-dispersion. An
incoming data signal: (i) is wavelength converted, (ii) the converted signal is passed
through a high-chromatic-dispersion element which produces a wavelength-

74
dependent time delay, and (iii) the output can be wavelength-converted back to its
original wavelength. Recently, delays >1 μs at data rates of 10 to 40 Gbit/s for data
modulation formats of On-Off-Keying (OOK) and Differential Phase-Shift-Keying
(DPSK) have been demonstrated [19, 61, 4]. A report also demonstrated the delay of
an 80 Gbit/s DQPSK channel by 105 ns using PPLN wavelength converters [17]. A
laudable goal would be to demonstrate the DQPSK format at the 100-Gbit/s Ethernet
standard rate and for a much longer delay time.
In this chapter, we demonstrate an optically-controlled tunable delay element
using wavelength conversion in a highly nonlinear fiber, dispersion compensating
fiber, and optical phase conjugation, enabling a continuous delay of up to 1.16-μs
using 100 Gbit/s NRZ-DQPSK and 50 Gbit/s NRZ-DPSK modulation formats. This
corresponds to a delay-to-bit-rate product exceeding 55,000 for 50 Gbit/s DPSK and
110,000 for 100-Gbit/s DQPSK. A BER of 10
-9
was obtained for all formats, over
the extent of the delay tuning range.

Figure 6-1: Block diagram. Dispersion compensating fiber (DCF), fiber Bragg
grating (FBG), bandpass filter (BPF), transmitter (TX), receiver (RX), and
highly nonlinear fiber (HNLF).
BPF
2nm
EDFA
λ
P1
DCF
HNLF ≈
300m
Raman
Pumps
-13,300
ps/nm
λ
P2
BPF
2nm
HNLF ≈
300m
Rx
BPF
2nm
λ
IN
50 Gb/s NRZ-
DPSK or 100 Gb/s
NRZ-DQPSK Tx
FBG
+95 ps/nm
or 0 ps/nm

75
6.2 Experimental Setup
An experimental block diagram of our setup is shown in Figure 6-1. 100 Gbit/s
NRZ-DQPSK is generated by driving two parallel integrated Mach-Zehnder
modulators (MZM) with 50 Gbit/s, 2
31
-1, PRBS data. The data is shifted by 123 bits
to de-correlate the two data streams. One data stream is removed to generate the 50
Gbit/s NRZ-DPSK signal. The optical signal is then wavelength converted using a
300m piece of highly nonlinear fiber (HNLF) with a zero-dispersion wavelength
(ZDW) of ~1560 nm. A 1-pump degenerate four-wave-mixing (FWM) approach is
used to convert the signal (λ
IN
≈ 1536 nm) from 1540 to 1583 nm. The converted
signal is then filtered out and sent through the DCF (D ≈ -13.3 ns/nm at 1550 nm) to
impose a wavelength-dependent delay. Counter propagating Raman amplification in
the DCF is used to compensate for the ~80 dB of loss. Following the DCF, the
delayed signal is phase conjugated using a second piece of HNLF (ZDW ≈ 1558).
Again, a 1-pump FWM approach is used to convert the signal ~4 to 14 nm up in
wavelength. The signal is passed through the DCF a second time to compensate the
intra-channel dispersion and to undergo a second wavelength dependent delay. A
third wavelength conversion stage would be needed to return the output to its
original wavelength, but is not implemented. The signal is received using a pre-
amplified receiver with a 50 GHz delay-line interferometer (DLI) and balanced
photo-receiver.

76

(a) (b)

(c)
Figure 6-2: (a) Measured delay of 1.16 μs. (b) Received 50 Gbit/s ODB signal for
10 pm changes in laser wavelength showing ~275 ps changes in delay. (c)
Experimental spectra of first and second wavelength conversion stages for the
maximum delay value.
As the dispersion profile is not linear with wavelength, several different types of
fiber are combined to create a relatively flat profile. The residual dispersion is
removed using a switchable FBG based dispersion compensator to pre-disperse the
input signal with either +95 ps/nm, for converted wavelengths of 1540 to 1571 nm,
or by 0 ps/nm, for wavelengths of 1568 to 1583 nm, and adjusting the phase
conjugation distance[63]. The total measured delay of 1.16 μs is shown in Figure
6-2(a) above. Figure 6-2(b) shows the received optical duobinary (ODB) signal after
the DLI as the first stage pump is varied in 10 pm steps. The expected ~275 ps
0
0.2
0.4
0.6
0.8
1
1.2
1540 1555 1570 1585
Relative Delay (us)
Converted Wavelength (nm)
10 pm/step : 275 ps/step
100 ps/div
λ=1561.80
λ=1561.82
λ=1561.81 275 ps
275 ps
λ-Conversion
Phase
Conjugation
λcon
1583 nm
λcon
1583 nm
λsig
1536 nm
λpump
1559.2 nm
λpump
1585.1 nm
λconj
1587.2 nm
10dB/D
1nm/D
10dB/D
7nm/D

77
changes in delay are easily seen. Figure 6-2(c) shows the experimental spectra of
both wavelength conversion stages for the case of minimum delay.

(a) (b)

(c)
Figure 6-3: (a) 0, 0.5ps, and 1ps delay resolution of a single 40Gbit/s RZ-OOK
bit. (b) RF-spectra showing optical mixing for different AOM frequency offsets.
(c) Bit-error-rate curves for varying delay values with and without the AOMs.
6.3 Results and Discussion
The performance of our delay system is assessed through bit-error-rate (BER)
measurements and is shown in Figure 6-3(a) and (b). Power penalties from 2.3 to 5.4
dB and 1.6 to 3.9 dB are observed for 100 Gbit/s DQPSK and 50 Gbit/s DPSK
respectively. Figure 6-3(c) shows the calculated second-order residual dispersion
(ps/nm
2
) which is not compensated by the phase conjugation process. At the
-40 -35 -30 -25 -20 -15
2
3
4
5
6
7
8
9
Received Power (dB)
-Log10(BER)
100 Gb/s NRZ-DQPSK
1540
nm
1561.8
nm
1583
nm
-45 -40 -35 -30 -25 -20
2
3
4
5
6
7
8
9
Received Power (dB)
-Log10(BER)
50 Gb/s NRZ-DPSK
1540 nm
1561.8 nm
1583 nm
-80
-60
-40
-20
0
20
1530 1550 1570 1590
Wavelength (nm)
Residual Dispersion Slope
2
nd
Order Dispersion (ps/nm
2
)

78
maximum delay value, ~68 ps/nm
2
of residual second order dispersion contributes ~1
dB of the observed penalty. This second order dispersion may prove to be a limiting
factor for achieving larger delays and higher data rates.

79
Chapter 7:
Higher-Order Dispersion Compensation to
Enable a 3.6-μs Wavelength-Maintaining
Delay of a 100-Gbit/s DQPSK Signal

In this chapter, we demonstrate a method for dispersion slope compensation of a
conversion/dispersion based optical delay to enable 100-Gbit/s operation based on a
spatial light modulator and fiber-Bragg-gratings. A continuous delay of up to 3.6 μs
for 100, 80, and 20-Gbit/s differential quadrature phase-shift-keyed (DQSPK) and
50, 40 and 10-Gbit/s differential phase-shift-keyed (DPSK) waveforms is shown. A
time-delay bit-rate product of ~360,000 for 100-Gbit/s DQPSK with wavelength-
maintaining operation is achieved.
7.1 Introduction
Due to the demand for increased capacity and higher bandwidth,
telecommunications networks are pressed to improve their efficiency, both in terms
of their spectral efficiency (bits/sec/Hz) and their ability to efficiently process data.
Modulation formats that encode data using both amplitude and phase have shown
great promise in providing spectral efficiency [74]. Similarly, continuously tunable
optical delays may enable efficient and reconfigurable signal processing [88, 104].
One approach is conversion/dispersion based delay, in which: (i) the incoming signal
is wavelength converted, (ii) the converted signal is passed through a chromatic-
dispersive element that produces a wavelength-dependent group velocity (i.e.,

80
differential delay), and (iii) the output can be wavelength-converted back to the
original wavelength. Note that the intra-channel dispersion accumulation may
require compensation to prevent any degrading effects. Desirable characteristics of a
tunable optical delay may include: (a) transparency to the modulation format, (b)
high data-rate operation, and (c) the ability to accommodate a large range of delay
values.
Recent results on tunable optical delays include: (i) 10 Gbit/s non-return-to-zero
on-off-keyed (NRZ-OOK) data over >7 μs that combines conversion/dispersion with
discrete delays [21], (ii) 10-Gbit/s NRZ-OOK data over 1.8 μs and 40-Gbit/s
differential-phase-shift-keyed (NRZ-DPSK) data over 1.5 μs [5], and (iii) 100-Gbit/s
differential-quadrature PSK (NRZ-DQPSK) and 50-Gbit/s NRZ-DPSK over 1.1 μs
[68]. One significant challenge towards the continued performance increases of
tunable delays for high-bit-rate signals is the compensation of higher-order
chromatic dispersion [68, 63, 49, 62]. Several techniques have been demonstrated in
an effort to flatten the overall dispersion profile, including concatenating multiple
types of dispersive elements [68] and using a pre-dispersion element [5, 63, 49, 62].
The pre-dispersion method has shown controllable compensation of both 2
nd
- and
3
rd
-order dispersion and has been used to demonstrate 22 ns of delay, capable of
achieving a THz passband [49]. However, when incorporating large dispersion
values over wide wavelength ranges, accurate dispersion profile matching may be
difficult.

81
In this chapter, we demonstrate 3.6 μs of continuously tunable optical delay for
100, 80, and 20-Gbit/s RZ-DQPSK and 50, 40, 10-Gbit/s RZ-DPSK signals using
conversion/dispersion and 2
nd
- and 3
rd
-order dispersion compensation. We use a
switchable array of fixed dispersion value fiber-Bragg-gratings (FBGs) for coarse
compensation and a tunable liquid-crystal-on-Silicon (LCoS) spatial-light-modulator
(SLM) for fine compensation. Residual 3
rd
-order dispersion is reduced from -230 to
<±10 ps/nm
2
. At a 10
-9
bit-error-rate for a 3.6-μs delay of a 100-Gbit/s RZ-DQPSK
signal, a time-delay bit-rate product of ~360,000 is achieved.

Figure 7-1: (a) Conceptual diagram of pre-dispersion block to enable 100 Gbit/s
operation. (b) A 96% reduction in residual 3
rd
-order dispersion is achieved using
fixed fiber-Bragg-gratings (FBGs) and a tunable spatial light modulator (SLM).
7.2 Concept
A conceptual diagram of our pre-compensation block is shown in Figure 7-1(a).
A commercially available SLM with a 1x4 switch is combined with fixed FBGs to
provide dispersion matching. The maximum dispersion-bandwidth product of our
SLM is ±40 ps. For a 1-nm bandwidth, this translates into a tuning range of ±40
ps/nm of 2
nd
-order dispersion and ±40 ps/nm
2
for 3
rd
-order dispersion, each of which
Residual Dispersion Slope (ps/nm
2
)
1600 nm 1520 nm
300
100
-100
-300
Uncompensated
Compensated
FBG1
FBG2
FBG3
FBG4
Wavelength (nm)

82
can be controlled independently. The four FBGs provide: (FBG1) +50 ps/nm and
+50 ps/nm
2
, (FBG2) +140 ps/nm and +100 ps/nm
2
, (FBG3) +280 ps/nm and +150
ps/nm
2
, and (FBG4) +420 ps/nm and +200 ps/nm
2
, respectively. In this manner, the
delay range is separated into four overlapping bands that can be dispersion matched
to lower the 2
nd
- and the 3
rd
-order residual dispersion and enable high-data-rate
operation over a large delay. Shown in Figure 7-1(b) is the uncompensated residual
3
rd
-order dispersion, the overlapping effective 3
rd
-order dispersion profiles of the
four FBGs, and the final compensated 3
rd
-order dispersion profile.

Figure 7-2: Block diagram. Dispersion compensating fiber (DCF), spatial light
modulator (SLM), band-pass filter (BPF), erbium-doped fiber amplifier (EDFA),
receiver (Rx), and highly nonlinear fiber (HNLF).
7.3 Experimental Setup
The experimental block diagram for our technique is shown in Figure 7-2. Both
RZ-DPSK at 50, 40, and 10 Gbit/s and RZ-DQPSK at 100, 80, and 20 Gbit/s (λSig ≈
BPF 2nm
EDFA
λ
P1
HNLF ≈
460m
λ
Sig
DCF
Raman
Pumps
-33
ps/nm
λ
P2
BPF
2nm
HNLF ≈
330m
Rx
BPF
2nm
Raman
Pumps
BPF
2nm
HNLF ≈
520m
λ
P3
BPF
2nm
λ
P3
λ
P1
λ
P2
λ
1
2 3
4
Pre-compensation
Block
λ
Sig

83
1535.3 nm) are first sent to the pre-compensation block. The appropriate FBG is
chosen depending on the delay value selected; FBG4 for ~1539-1562 nm, FBG3 for
~1560-1575 nm, FBG2 for ~1573-1586 nm, and FBG1 for ~1584-1593 nm,
respectively. The optical path length of each of the 4 FBG branches is temporally
matched. A 2-nm wavelength overlap with <2 ps/nm and <4 ps/nm
2
variation is
maintained to allow for spectral continuity. In this manner, a continuous range of
delays can be maintained when switching internally between different FBGs.
Additionally, the SLM can be adjusted to minimize the 2
nd
and 3
rd
-order residual
dispersion at each delay value. This coarse (FBGs) and fine (SLM) tunable pre-
compensation allows the residual 2
nd
- and 3
rd
-order dispersion to be kept below ±14
ps/nm and ±10 ps/nm
2
, respectively, over a ~54-nm bandwidth.
Following the pre-dispersion block, the signal then undergoes its first wavelength
conversion in a 460-m piece of highly nonlinear fiber (HNLF) with a zero-dispersion
wavelength (ZDW) of ~1556 nm. A degenerate four-wave-mixing (FWM) approach
is used to convert the signal wavelength over the range of ~1539.3 to ~1593.3 nm
[3]. The signal and pump powers vary from 15-20 dBm. The converted signal is then
filtered out and sent to the DCF (i.e., port 1 of Figure 7-2) to impose a wavelength
dependent delay. The total dispersion value is measured to be ~ -33 ns/nm at 1550
nm. Co- and counter-propagating Raman amplification in the DCF is used to
compensate for the ~11 dB of loss in each of the 8 fiber sections. Launch powers into
each section of fiber are < -5 dBm for each signal.

84
Following the DCF (i.e., port 2 in Figure 7-2), the delayed signal is phase
conjugated using a second piece of 330-m HNLF (ZDW ≈ 1558 nm). Again, a
degenerate FWM approach phase conjugates the signal ~4 to 14 nm up in
wavelength. The pump and signal powers are each kept low (10-15 dBm) to avoid
excess distortion on the dispersed signal. The conjugated signal is separated from the
high power pump using a filter. It is then passed through the DCF a second time (i.e.,
port 3 in Figure 7-2) to compensate for the intra-channel dispersion and to undergo a
second wavelength dependent delay.
Following the second pass through the DCF (i.e., port 4 in Figure 7-2), the signal
is selected using a filter and returned to the original wavelength using a third
degenerate FWM setup with ~520-m of HNLF (ZDW ≈ 1558 nm). The pump and
signal powers varied from ~17 to 20.5 dBm to return the conjugated signal in the
range of ~1543.5 to 1597.1 nm back to the original wavelength of 1535.3 nm. The
signal is received using a pre-amplified receiver with 50, 40, and 10-GHz delay-line
interferometers and a 32-GHz balanced photo-receiver.

85

Figure 7-3: (a) Measured delay of 3.6 μs for 100-Gbit/s RZ-DQPSK. (b) 7-Gbit/s
packets used to illustrate the full delay tuning range.
7.4 Results and Discussion
The measured continuous delay for the input 100-Gbit/s RZ-DQPSK signal at
λSig≈1535.3 nm is shown in Figure 7-3(a) as a function of the converted
wavelength. In Figure 7-3(b), a 500-bit packet at 7 Gbit/s is used only to illustrate the
delay over the ~54-nm range. The phase conjugation wavelength separation is
adjusted at each point to match the dispersion of both passes through the DCF as
determined by the pre-dispersion and dispersion slope of the DCF [62, 63, 49]. The
delay resolution, ~64 ps, was limited by the 1-pm wavelength resolution of our pump
laser [69].
Figure 7-4(a) shows the constellation diagrams of 50-Gbit/s RZ-DPSK (top) and
100-Gbit/s RZ-DQPSK (bottom) before (Inset) and after the delay element captured
by a complex optical spectrum analyzer. Added distortions following the multiple
wavelength conversions and Raman amplified DCF spools are noticeable.
0
0.5
1
1.5
2
2.5
3
3.5
4
1535 1556 1577 1598
Converted Wavelength (nm)
Measured Optical Delay
(a) (b)
1.4 μs
500 ns/D
3.6 μs

86
Experimental spectra for the first and third λ-conversion stages at the minimum delay
value for 100-Gbit/s RZ-DQPSK are shown in Figure 7-4(b).

Figure 7-4: (a) Constellation diagrams showing the DPSK (Top) and DQPSK
(Bottom) signals before (Left) and after (Right) at the middle delay value, ~1567
nm. (b) Experimental spectra of the first stage (top) and third stage (bottom) at the
minimum delay value.
The measured BER performance curves for RZ-DQPSK and RZ-DPSK are
shown in Figure 7-5(a) and (b) respectively. The maximum (1539.3 nm), middle
(1567 nm), and minimum (1593.3 nm) delay values represent the best and worst case
performances for the entire range of delays values. For DQPSK, only the in-phase (I)
channel is shown; additionally, the quadrature-channel (Q) is measured in all cases
and performs within ~0.2 dB of the I channel. 10-Gbaud/s signals have penalties of
<1dB for DPSK and ~1 dB for DQPSK. 40-Gbaud/s and 50-Gbaud/s signals have
respective maximum penalties of 4 and 4.4 dB for DPSK and 5.8 and 7.1 dB for
DQPSK.
Before After
(a)
(b)
λcon
1593.3 nm
λsig
1535.3 nm
λpump
1564.7 nm
λconj
1597.1 nm
10dB/D
7nm/D
10dB/D
7nm/D
λsig
1535.3 nm
λpump
1566.2 nm

87

(a) (b)
Figure 7-5: Measured bit-error-rate performance of (a) 100 (Solid), 80 (Dashed),
and 20-Gbit/s (Solid) DQPSK and (b) 50 (Solid), 40 (Dashed), and 10-Gbit/s
(Solid) DPSK for the minimum (Red), middle (Blue), and maximum (Green)
delay values.
The dispersion slope compensation is varied to assess system impact. For -230
ps/nm
2
of residual dispersion slope, there is <1 dB of improvement when adding the
correct amount of higher-order compensation at 10 Gbaud/s. However, at 40 and 50
Gbaud/s, the signals are not recoverable without proper slope compensation. Figure
7-6 shows the power penalty for different values of residual 3
rd
-order dispersion on
the 100-Gbit/s DQPSK signal. To maintain <1 dB penalty, the 3
rd
-order
compensation needs to be accurate to better than ~45 ps/nm
2
.
20G
Max
80G
Max
100G
Max
Received Power (dBm)
10G
Max
40G
Max
50G
Max
Received Power (dBm)
Figure 7-6: Power penalty as a function of residual 3

: Power penalty as a function of residual 3
rd
-order dispersion (ps/nm2)
for 100-Gbit/s RZ-DQPSK.

88
order dispersion (ps/nm2)

89
Chapter 8:
Delay Extension to 5-μs for a
10 Gbit/s RZ-DPSK Signal

In this chapter we experimentally demonstrate the extension of our delay system
for 5.4 μs. By lowering the symbol-rate from 50 Gbaud to 10 Gbaud we can increase
our dispersion tolerance by a factor of ~25. This allows us to add 50% more DCF to
our system and still achieve a lower system penalty.

Figure 8-1: Improved experimental setup utilizes -48 ns/nm of DCF to achieve a
5.4 μs relative delay.
8.1 Experimental Setup
The experimental block diagram for our technique is shown in Figure 8-1. It
remains nearly identical to the optimized setup in the previous chapter. 10 Gbit/s RZ-
DPSK is generated using a Mach-Zehnder modulator followed by a pulse carver. The
waveform is first sent to the pre-compensation block. The appropriate FBG is chosen
depending on the delay value selected; FBG4 for ~1539-1562 nm, FBG3 for ~1560-
BPF 2nm
EDFA
λ
P1
HNLF ≈
460m
λ
Sig
DCF
Raman
Pumps
-48
ns/nm
λ
P2
BPF
2nm
HNLF ≈
330m
Rx
BPF
2nm
Raman
Pumps
BPF
2nm
HNLF ≈
520m
λ
P3
BPF
2nm
λ
P3
λ
P1
λ
P2
λ
1
2 3
4
Pre-compensation
Block
λ
Sig

90
1575 nm, FBG2 for ~1573-1586 nm, and FBG1 for ~1584-1593 nm, respectively.
The optical path length of each of the 4 FBG branches is temporally matched. A 2-
nm wavelength overlap with <2 ps/nm and <4 ps/nm
2
variation is maintained to
allow for spectral continuity. In this manner, a continuous range of delays can be
maintained when switching internally between different FBGs. Additionally, the
SLM can be adjusted to minimize the 2
nd
and 3
rd
-order residual dispersion at each
delay value. Since the pre-compensation setup was no longer matched to our fiber
(we are using 50% more fiber), the 2
nd
-order dispersion was minimized using phase
conjugation [63] while the 3
rd
-order dispersion varied from 0 to -127 ps/nm
2
as
shown in Figure 8-2.

Figure 8-2: Measured residual dispersion slope after the addition of the extra
DCF. At the minimum wavelength, this corresponds to ~0.8 dB penalty for a 10
Gbit/s RZ-DPSK signal.
Following the pre-dispersion block, the signal then undergoes its first wavelength
conversion in a 460-m piece of highly nonlinear fiber (HNLF) with a zero-dispersion
-150
-100
-50
0
1530 1545 1560 1575 1590 1605
Wavelength (nm)
Residual Dispersion Slope

91
wavelength (ZDW) of ~1556 nm. A degenerate four-wave-mixing (FWM) approach
is used to convert the signal wavelength over the range of ~1539.3 to ~1593.3 nm
[3]. The signal and pump powers vary from 12-21 dBm. The converted signal is then
filtered out and sent to the DCF (i.e., port 1 of Figure 8-1) to impose a wavelength
dependent delay. The total dispersion value is measured to be ~ -48 ns/nm at 1550
nm. Co- and counter-propagating Raman amplification in the DCF is used to
compensate for the ~11 dB of loss in each of the 8 fiber sections. Launch powers into
each section of fiber are < -5 dBm for each signal.
Following the DCF (i.e., port 2 in Figure 8-1), the delayed signal is phase
conjugated using a second piece of 330-m HNLF (ZDW ≈ 1558 nm). Again, a
degenerate FWM approach phase conjugates the signal ~3 to 18 nm up in
wavelength. The pump and signal powers are each kept low (11-15 dBm) to avoid
excess distortion on the dispersed signal. The conjugated signal is separated from the
high power pump using a filter. It is then passed through the DCF a second time (i.e.,
port 3 in Figure 8-1) to compensate for the intra-channel dispersion and to undergo a
second wavelength dependent delay.
Following the second pass through the DCF (i.e., port 4 in Figure 8-1), the signal
is selected using a filter and returned to the original wavelength using a third
degenerate FWM setup with ~520-m of HNLF (ZDW ≈ 1558 nm). The pump and
signal powers varied from ~14 to 19 dBm to return the conjugated signal in the range
of ~1543.5 to 1597.1 nm back to the original wavelength of 1535.3 nm. The signal is

92
received using a pre-amplified receiver with 50, 40, and 10-GHz delay-line
interferometers and a 32-GHz balanced photo-receiver.

Figure 8-3: Experimental measurement of the delay range was performed using a
300 Mbit/s packet. The full 5.4 μs range is shown (800 ns/Div).
8.2 Results and Discussion
The measured continuous delay for the input 10-Gbit/s RZ-DPSK signal at λSig
≈ 1535.3 nm is depicted in Figure 8-3 and shown in Figure 8-4(b) as a function of
the converted wavelength. In Figure 8-3, a packet at 300 Mb/s is used only to
illustrate the delay over the ~54-nm range. The phase conjugation wavelength
separation is adjusted at each point to match the dispersion of both passes through
the DCF as determined by the pre-dispersion and dispersion slope of the DCF[62, 63,
49]. The delay resolution, ~97 ps, was limited by the 1-pm wavelength resolution of
our pump laser [69]. Experimental spectra for the first and third λ-conversion stages
at the minimum delay value for 100-Gbit/s RZ-DQPSK are shown in Figure 8-4(a).

93
Note that the use of degenerate four-wave-mixing allows the final stage to be not
only wavelength maintaining, but can be independently tuned to return the delayed
signal to any output wavelength in the ~54 nm range.

(a) (b)
Figure 8-4: (a) Experimental spectra for the first and third wavelength conversion
stages. Wavelength maintaining operation is accomplished. (b) Measured relative
delay vs. the converted wavelength showing the full 5.4 μs delay range.
The measured BER performance curves for 10 Gbit/s RZ-DPSK is shown in
Figure 8-5. The maximum (1539.3 nm), middle (1567 nm), and minimum (1593.3
nm) delay values represent the best and worst case performances for the entire range
of delays values. Power penalties ranged from the 3 to 5 dB, with the worst
performance at the edge wavelengths. These wavelengths are on the edge of the
Raman gain profile, the filter tuning ranges, and the EDFA gain profiles. This causes
a decreased signal-to-noise ratio and reduces the overall performance for those delay
values.
λcon
1593.3 nm
λsig
1534.3 nm
λpump
1564.7 nm
λconj
1597.1 nm
10dB/D
7nm/D
10dB/D
7nm/D
λsig
1534.3 nm
λpump
1566.2 nm
0
1
2
3
4
5
6
1535 1556 1577 1598
Converted Wavelength (nm)

94

Figure 8-5: Bit-error-rate measurements for the minimum, middle, and maximum
delay values compared to the back-to-back case.

1
10G RZ-DPSK B2B 10G Max 1539.3
10G Min 1594.3 10G Mid 1567
2
3
4
5
6
7
8
9
-Log
10
(BER)
Received Power (dBm)
-48 -42 -36 -30 -24
Min
Mid
Max

95
Chapter 9:
Fine Tuning of Optical Delays Using Cascaded
Acousto-Optic Frequency Shifters

We demonstrate a technique for fine tuning of optical delays using cascaded
acousto-optic modulators to improve delay resolution by five orders of magnitude
compared to a 1-pm tunable laser. A 256-ns delay with <0.5-ps resolution is shown
for 40-Gbit/s RZ-OOK with no added penalty.
9.1 Introduction
As is true in both the electronic and optical domains, time delays form the
backbone in many types of signal processing functions. The ability to controllably
tune the delays adds significantly to both: (i) optimizing system performance by
accurate delays, and (ii) dynamic reconfigurability to different system parameters.
Recently, there has been increased interest in the ability to produce tunable optical
delays such that fine granularity can be achieved to improve the performance of
many types of optical signal processing functions [88]. For data rates ≥40 Gbit/s in
which the length/time scale of a single bit is fairly short, accurate optical timing can
be critical in time multiplexing, synchronization, switching, equalization, correlation,
and time-slot interchange [88, 104]. Of course, accuracy and fine-tuning granularity
become critical as the bit rate and timing sensitivity increase.

96
There have been several reports of tunable optical delays that exceed 1 ns for >1
Gbit/s data signals [104, 69, 63, 49, 66, 19, 61, 4]. One promising technique is called
conversion/dispersion, in which: (i) a data signal is wavelength converted, (ii) it
passes through a high-chromatic-dispersion element which produces a wavelength-
dependent time delay, and (iii) the output can be wavelength-converted back to its
original wavelength. This technique produces a time delay that is a product of the
wavelength conversion range (nm) multiplied by the total dispersion value (ps/nm).
Depending on the non-linear material used for wavelength conversion, this
conversion range can be very broad, 10’s of nano-meters in an SOA, ~50 to 100 nm
in PPLN waveguides, >150 nm in Silicon waveguides, and 100’s of nano-meters in
fiber [89, 16, 32, 11]. However, the resolution and tuning range of a wavelength
conversion process is typically limited by the resolution and tuning range of a pump
laser, which can also limit the tuning range and resolution of the delay.
The ability to accurately control the delay is critical not only for stable operation
of the delay, but for many potential delay applications. Drift of the pump lasers and
of the dispersive element can cause drift in the delay system. Accurate control of the
delay, combined with feedback, may be necessary to help provide stable system
operation. As bit rates increase from 40 to 100’s of Gbit/s, this fine tuning becomes
critical. For example, a 10% accuracy in bit interleaving of a 100 Gbit/s data signal
requires a temporal resolution of 1 ps. However, the typical approach for delay
tuning has been to tune the wavelength converter’s pump laser, and even good lasers

97
tend to have wavelength tuning steps of ~1 pm. A 300 ns delay using a 20 nm
wavelength range with this 1 pm laser step translates into a delay step of 15 ps, far
beyond what would be required for good system operation. With recent delay results
from 0.5 μs to 1.8 μs for both 10 and 40 Gbit/s data, a desirable feature would be to
enable very fine tuning of the optical delays down to and below the 1 ps regime.
In this chapter, we demonstrate a method for improving the delay resolution
using cascaded acousto-optic mixers and further discuss the impact of the cascaded
acousto-optic frequency shift on a highly dispersed signal prior to compensation and
the resolution limitations of our proposed setup. Coarse tuning (>15ps) is performed
using the tunable laser source to reach a maximum delay of 256 ns for a 40 Gbit/s
return-to-zero on-off-keying (RZ-OOK) signal. Cascaded acousto-optic modulators
(AOMs) are used to precisely control the laser center wavelength and achieve a delay
resolution of <0.5 ps.
9.2 Concept
The conceptual diagram of our technique is shown in Figure 9-1(a). Inside the
delay module, an input signal is wavelength converted before passing through
dispersion compensating fiber (DCF). The DCF is used to generate a wavelength-
dependent group delay. A wavelength converter after the DCF is used for near-
wavelength maintaining phase conjugation of the signal, before it is passed through
the DCF a second time. This double-pass allows for compensation of the residual
dispersion and a second relative group delay. The maximum amount of delay (Δτ)

98
can be determined by the product of the wavelength conversion bandwidth (Δλ) and
the dispersion (D): Δτ ≈ 2*D*Δλ, where the factor of two comes from the relative
group velocity variation on both passes through the DCF. Both the wavelength of the
first wavelength conversion stage and of the phase conjugation stage affects the
delay as well as the dispersion through the system.
(a)
(b)
Figure 9-1: (a) A tunable laser with 1pm (125MHz) resolution is used to coarse
tune the delay from 0 to 256ns. Cascaded acousto-optic modulators (AOMs) shift
the laser center frequency with 1kHz resolution; fine tuning the delay from 0 to
25ps. (b) Measured fine and coarse tuning ranges of our system.
For accurate delay and accurate compensation, fine tuning of both pumps may be
necessary. Typical tunable laser sources can be tuned in steps of 1 pm (~125 MHz) at
1550 nm. With a DCF of -7,556 ps/nm (measured at 1550nm), this translates into a
delay resolution of ~15 ps for our system. To demonstrate accurate control of
wavelength and delay, one of the pump lasers is passed through a pair of cascaded
AOM1
(+ Δf
1
)
AOM2
(-Δf
2
)
Tunable
Laser
f f
f
0
+ Δf
1
f
f
0
+ Δf
1
-Δf
2
Small Net Shift Δf = Δf
1
-Δf
2
f
0
+Δf
1
-Δf
2
Shift the laser
freq. by Δ f
1
shift the laser
freq. by Δ f
2
0 200 1541 1559
Coarse Tuning (ns)
300
0
25
0
Fine Tuning (ps)
Δf (MHz)
Wavelength (nm)

99
AOMs to precisely tune the laser’s center wavelength using acousto-optic mixing.
By cascading the AOMs, we assembled a setup capable of a frequency resolution of
1 kHz, less than the 125 MHz resolution of the laser.
Figure 9-2 depicts the operation of our acousto-optic modulators. An RF tone, f
a
,
is applied to each modulator such that the output laser frequency is given by f
out
=
f
in
+f
a
for the up shifted case, and f
out
= f
in
–f
a
for the down shifted case. By cascading
one up shifting and one down shifting modulator in series and applying separate RF
tones, f
1
and f
2
, the total shift in laser wavelength is given by f
out
= f
in
+Δf, where Δf
= f
1
–f
2
[100]. This method allows for a much broader tuning range of the output laser
frequency while also allowing for very fine control of the total shift in wavelength.
Individually, each modulator had a minimum shift in wavelength, but by cascading
two modulators, the wavelength shift depends directly on the difference in applied
RF tones. The stability and resolution of the RF source is one of the fundamental
limitations of this approach. Our RF tone controller had a 1 kHz resolution which
allowed the output laser wavelength to be adjusted in 1 kHz steps, exceeding the
~125 MHz (1 pm @ 1550nm) step resolution of a conventional tunable laser.

100

Figure 9-2: Acousto-optic frequency shifters for up shifting and down shifting a
CW laser.
The two AO modulators, AOM1 and AOM2, are made from Indium Phosphide
with Bragg angles of 90 mrad and 81 mrad and separation angles of 180 mrad and
162 mrad respectively. The modulators are designed for minimal loss when a 595
MHz and 530 MHz RF tone is applied, respectively. As the RF tone frequency is
shifted the loss increases from 2.3 to 6 dB and from 2.2 to 5 dB for shifts of ±70
MHz and ±65 MHz respectively. This allows tuning of the total frequency shift, Δf,
from 0 to 200 MHz with a total loss of 5.5 to 11 dB. As the optical output is passed
through a high-power amplifier, 11 dB of total loss is acceptable for our application,
although a narrower RF tuning range would provide less loss. In general, the tuning
range only needs to be as large as the minimum resolution of the tunable laser, ~125
MHz (1 pm) in our case.
AOM 1: Up Shift AOM 2: Down Shift
f
out
=f
in
+f
a
f
out
=f
in
-f
a
Transmitted
f
in
Reflected
f
in
+f
a
Input
f
in
Input
f
in
Transmitted
f
in
Reflected
f
in
-f
a
Electric
Input f
a
Electric
Input f
a

101

Figure 9-3: Block diagram. Wavelength Converter (W/C1), dispersion
compensating fiber (DCF), acousto-optic modulator (AOM), Mach-Zehnder
modulator (MZM), receiver (RX), and highly nonlinear fiber (HNLF).
9.3 Experimental Setup
An experimental block diagram of our setup is shown in Figure 9-3. A 40 Gbit/s
RZ-OOK signal is wavelength converted using highly nonlinear fiber (HNLF). A
degenerate four-wave-mixing approach is used where the high power pump is swept
in wavelength to control the location of the converted signal. The signal (λ
0
≈ 1537
nm) is up-converted from 1541 to 1558 nm. The converted signal is then filtered out
and sent through the DCF (D = -7,556 ps/nm at 1550nm) to impose a wavelength-
dependent delay. Raman amplification in the DCF is used to compensate for the 22
dB of loss in the fiber. The delayed signal is then phase conjugated using a second
piece of HNLF. Again, a degenerate FWM approach is used to convert the signal ~4
nm up in wavelength. The signal is passed through the DCF a second time to
MZM
λ
0
Pulse
Carver
CLK
Filter
1.2nm
EDFA
λ
P1
DCF
HNLF ≈
300m
Raman
Pumps
Raman
Pumps -9047
ps/nm
Rx
40 Gb/s
Data
HNLF ≈
100m
λ
P2
Filters
1.2nm
EDFA
Dispersive
Element
1
Optical Phase
Conj.
1
W/C
1
AOM1
AOM2

102
compensate residual dispersion and to undergo a second wavelength dependent
delay. After the second pass through the DCF, the delayed signal is sent to a pre-
amplified receiver for bit-error-rate (BER) measurements. A third wavelength
conversion stage could be used to make the system wavelength maintaining without
impacting the fine tuning resolution.
The wavelength of the second pump laser (λ
P2
) is precisely controlled using two
AOMs. Control of the first pump laser (λ
P1
) is also possible; however, fine tuning of
the second pump was chosen to investigate any impact the AOMs might impart
during the phase conjugation process. At this point in the system, the signal is
heavily dispersed and any added distortion will impair the signal after it is
compensated. The first modulator performs a +595 MHz ± 70 MHz while the second
performs a -530 MHz ± 65 MHz frequency shift and are controlled to a 1 kHz
resolution. This gives a total frequency shift of Δf = f
1
-f
2
= 0 to 200 MHz. Adjusting
the laser wavelength allows for coarse tuning of the delay from 0 to 256 ns with ~15
ps resolution while adjusting the AOMs allows fine tuning from 0 to 25 ps with <0.5
ps resolution.

103

(a) (b)
Figure 9-4: (a) Sampling scope trace of 40 Gbit/s RZ-OOK bits with inset
showing 0, 0.5, and 1ps delay shifts. (b) Tuning range of our cascaded AOMs is
shown through optical mixing measurements.
9.4 Results and Discussion
Figure 9-4(a) shows part of the 40 Gbit/s RZ-OOK bit stream, captured by a
sampling scope with < 200 fs resolution. The inset shows a zoom of a single bit for
frequency shifts of 0, 4.13, and 8.26 MHz corresponding to delays of 0, 0.5, and 1
ps, respectively. Optical mixing measurements, shown in Figure 9-4(b), where used
to show the fine tuning range of the laser wavelength. The maximum shift of 200
MHz shows that this method can adequately cover the ~125 MHz step size of our
tunable laser.
BER measurements are used to evaluate the impact of the fine tuning on the
delay system. The pump amplifier is adjusted to keep the pump power fixed when
the AOMs are bypassed. Figure 9-5 shows BER curves with and without the AOMs
for converted wavelengths of 1541, 1549, 1558 nm corresponding to the maximum,
middle, and minimum delay values. For the displayed curves, the AOMs were set to
12.5
ps
Div
2
0.5 ps Delay
1 ps Delay
Δf=
4.13 MHz
Δf=
8.26 MHz
Δf=
0 MHz
ps
Div
0.2
μW
Div
1kHz 10kHz 100kHz 1MHz 10MHz 100MHz
RF Power (dB)
-90
-10
1GHz
Difference in AOM Frequencies (Δf)

104
provide a fixed shift of 60 MHz shift, although shifts of 0 to 200 MHz were
measured with no change in performance. A 1.2 dB penalty is noticed for the
minimum and maximum delay values. The addition of the AOMs for fine tuning did
not contribute an observable system penalty. The proposed fine tuning method could
be used with a feedback loop to increase system stability including compensation of
fiber length changes due to temperature variations.

Figure 9-5: Bit-error-rate curves for varying delay values with and without the
AOMs.

Back-To-Back
Stage 1 - 1549 - w/ AOMs
Delay - 1549 - w/o AOMs
Delay - 1549 - w/ AOMs
Delay - 1541 - w/o AOMs
Delay - 1558 - w/o AOMs
Delay - 1558 - w/ AOMs
Delay - 1558 - w/ AOMs
Delay - 1549 - w/o AOMs
Delay - 1541 - w/ AOMs
-40 -38 -36 -34 -32 -30 -28
2
3
4
5
6
7
8
9
Received Power (dBm)
-Log
10
(BER)

105
Chapter 10:
Continuously Tunable All Optical Buffer
Using Conversion Dispersion Based Delay

We experimentally demonstrate a continuously tunable, all-optical packet buffer
based on conversion dispersion delays. 40 Gbit/s Asynchronous Transfer Mode
(ATM) packets with return-to-zero on-off keying (RZ-OOK) data are buffered up to
10-packet length (116 ns). The packet buffer performance is characterized for several
delay values. Power penalties of 3 dB, 2 dB, and 0.5 dB for 0-packet, 5-packet, and
10-packet delays are achieved, respectively, at a bit error rate (BER) of 10
-9
.
Reconfiguration of the packet buffer is also investigated and reconfiguration times as
fast as 25 ps are shown by using a high speed optical switch to toggle between
wavelength conversion pumps. The reconfiguration is also demonstrated for a 1 ns
guard time between the packets. It is observed that reconfiguration with this method
results in ~1.1 dB and ~2 dB extra power penalty at 10
-9
BER for 1 ns and 25 ps
guard times, respectively.
10.1 Introduction
As network data-rates and capacity grow the limited scaling and performance of
electrical routers may become a bottleneck for system operation. One potential
technology for highly efficient and high-capacity networking is optical packet
switching [25]. A critical challenge for any optical switch is the need to implement
the rapid resolution of contention and congestion within the core routers [72].

106
Traditionally, this has been difficult to realize in the optical domain due to the lack
of optical memory. Typically, optical contention resolution can take the form of
packet dropping, wavelength conversion, deflection routing, and buffering. While the
first methods may be inefficient to implement on a large scale, optical buffering has
shown the potential for high-capacity operation.
Several techniques have been published that demonstrate optical buffers,
including: (i) using material resonances or coupled resonant structures to decrease
group velocity [88, 31], and (ii) feed-back or feed-forward fiber delay line buffers
[101, 15, 86, 1, 83]. These buffers tend to switch between a discrete set of fixed
delay times producing a tradeoff between the amount of delay and the delay
granularity, or provide small continuous delays (up to several bits) limited by the
system bandwidth or distortions. Optical buffering can also be achieved by relative
delays based on the conversion/dispersion techniques, which are shown to offer large
delays [19, 4, 66], with fine granularity [69]. Conversion/dispersion delays tunably
wavelength convert the optical signal before passing it through a chromatic-
dispersive element and wavelength converting back to the original wavelength. The
amount of delay can be tuned by changing the converted wavelength. Recently,
conversion dispersion has been used to generate a tunable delay (up to 13.7 ns)
capable of buffering a 40 Gbit/s variable-length packets by 2 packet lengths [102].
However, residual chromatic dispersion limits the buffer from accommodating larger
packet sizes and larger delays while reconfiguration is not shown.

107
In this chapter, we demonstrate and characterize a 10-packet-depth, 40 Gbit/s
optical buffer with a <0.5 ns reconfiguration time using a 116 ns, continuously
tunable conversion/dispersion delay. The use of a continuously tunable optical delay
allows for adjustment to changes in packet size and data-rate of the incoming optical
signal. The buffer is both amplitude and phase maintaining, making it transparent to
a host of modulation formats. Optical phase conjugation is used to minimize the
residual dispersion limitation in the delay [93, 36]. Rapid reconfiguration by
switching between wavelength converting pumps is shown for switching windows as
fast as 25 ps (1 bit-time).

Figure 10-1: Conceptual block diagram of the demonstrated optical buffer. Input
packet stream is sent to two paths. Upper path induces the relative delay on the
selected packet(s), where the lower path deletes any desired packet(s).
λ λ λ λ
s
(t) λ λ λ λ
s
(t)
Extract
Packets
Wavelength
Dependent Delay
Convert
Back to λ λ λ λ
s
Optical
Modulator
RF Signal
2x2 Fast Optical
Selector Switch
λ λ λ λ
P1
λ λ λ λ
P2 t
Δτ Δτ Δτ Δτ
1 1 1 1
t
3 2 10 1
t
3 2 10 1
t
3
2
10
1
t
1 2
t
3
2
10
1
2 1
Δτ Δτ Δτ Δτ
2 2 2 2
λ λ λ λ
C
Δτ Δτ Δτ Δτ
2 2 2 2
Δτ Δτ Δτ Δτ
1 1 1 1
λ λ λ λ
C
(t)
Fast-Switched
Pumps
Extracted Packets
t
t
2 1
λ λ λ λ
P1
λ λ λ λ
P2

108
10.2 Concept
A conceptual block diagram of the buffer is shown in Figure 10-1. The buffer
consists of two functional paths in parallel. The incoming packet stream is split into
two and sent to these two paths for processing. The upper path induces the relative
delay on the desired packet(s) by utilizing the conversion/dispersion based delay,
while the lower path is used for deletion of packet(s) from their original time slot(s).
Additionally, this gives the functionality of emptying any desired time slot if a data
packet is not needed.

Figure 10-2: Conceptual block diagram of the conversion/dispersion technique
used to generate relative delays in the optical buffer. The first wavelength
conversion controls the amount of delay. The second wavelength conversion is
the phase conjugation stage. After the delay, the signal is converted back to the
original wavelength to have a wavelength transparent delay.
In the upper path, three wavelength conversion stages are used to complete the
relative delay for the selected packet(s). A highly dispersive medium is used for the
wavelength dependant delay which utilizes the group velocity variation due to the
chromatic dispersion [19, 4, 66, 104, 62, 63, 49]. A conceptual block diagram of
delays based on the conversion/dispersion technique is shown in Figure 10-2. The
λ λ λ λ
Sig
(t) λ λ λ λ
Sig
(t)
Dispersive
Element (D)
λ λ λ λ
C
(t) λ λ λ λ
C
’
(t)
‘Delayed’
Output
Signal
Input
Signal
Phase
Conjugation
λ λ λ λ
C
(t) λ λ λ λ
C
’
(t)
Tunable
Wavelength
Conversion
Tunable
Wavelength
Conversion
λ λ λ λ
C
(t) : {λ λ λ λ
C_MIN
, λ λ λ λ
C_MAX
}
Δλ = Δλ = Δλ = Δλ = λ λ λ λ
C_MAX
- λ λ λ λ
C_MIN
Δτ Δτ Δτ Δτ ≅ ≅ ≅ ≅ 2 2 2 2 Δ Δ Δ Δλ λ λ λ D

109
input signal is wavelength converted to a desired wavelength (λ
C
(t)) that will
determine the relative delay in the dispersive medium. Raman amplification is used
in the dispersive medium to overcome the losses [70]. After the dispersive medium, a
second wavelength conversion is utilized to phase conjugate the signal to a nearby
wavelength (λ
C
’(t)). The phase conjugated signal is then sent back through the same
dispersive element to make use of the same group velocity dependence. Therefore,
the induced relative delay is almost doubled while dispersion compensation is also
achieved. Hence, the maximum relative delay (Δτ) induced by the system is
approximately equal to 2×Δλ×D, where Δλ is the difference between the minimum
and the maximum wavelengths that the signal is converted to, and D is the dispersion
of the medium. In our scheme, the first wavelength conversion stage uses gated
pump(s) for the wavelength conversion. The pump(s) is/are turned on and
synchronized only for the duration of the desired packet(s). This results in the
wavelength conversion of only the packet(s) that is (are) going to be delayed.
Extracted packets are then sent to the dispersive medium to induce the relative delay.
Phase conjugation, as described, is used to double the delay and compensate the
dispersion. After the second pass through the dispersive element, a third wavelength
conversion stage is used to convert the delayed packet(s) back to the original input
wavelength. Some additional fiber is used in the lower path to emulate flight time
matching/synchronization of the two paths prior to combining. The faster wavelength
for the selected packet(s) is used as the reference: The time alignments of the two

110
paths are realized such that this wavelength results in no relative delay with respect
to the original packet time slot. Hence, the output packet stream is identical to the
input packet stream for this wavelength. This feature allows the buffer to be used
with tunable packet sizes and variable bit rates as the delay is continuous between
zero and the maximum value.

Figure 10-3: Illustration of reconfiguration of the optical buffer. Packets to be
delayed are extracted to the corresponding wavelengths in the first wavelength
conversion stage. Thus, they experience different amounts of delay in the
dispersive element. The reconfiguration should take place within the guard time
between the packets. Therefore, the minimum guard time without any data loss is
determined by the reconfiguration speed.
Reconfiguration of the packet buffer requires tunability of the delay within a
guard time between the packets as illustrated in Figure 10-3. The most rapid
reconfiguration is needed when two consecutive packets are to be buffered by
different delay values. This requires wavelength conversion of the consecutive
packets to two different wavelengths. As demonstrated in [52] a single gated pump
can enable bit- and packet-level wavelength conversion. For the reconfiguration two
separate gated pumps are required to realize the wavelength conversion of two
consecutive packets to two different wavelengths. A fast 2x2 optical switch is
utilized to realize the switching between packet extraction pumps for this purpose.
Two consecutive packets entering the dispersive element at different wavelengths
Delay Module
Delay
Reconfiguration
Delay-2
Delay-1
3 1 2
3 1 2
(switching between
gated pumps within
the guard time)
t
t

111
will have different delays. However, a phase conjugation scheme that maintains the
relative packet wavelength difference is required for the phase conjugation process.
We have used two wavelength conversion stages in parallel that phase conjugates the
extracted packets independently, where other methods that can allow realization of
this process may also be possible. If the buffer is to process more than two packets
independently at once, a fast tuning laser or a laser bank with a large scale optical
switch is necessary for the extraction process. This case would also require more
packet signals to be phase conjugated, and hence, the number of phase conjugation
stages should be increased accordingly.

Figure 10-4: Experimental setup for the optical buffer. Modifications for
demonstration of reconfiguration are shown with dotted lines and italic titles.
MZM: Mach-Zehnder modulator; CLK: clock; TDL: tunable delay line; BPF:
bandpass filter; DCF: dispersion compensating fiber; SSMF: standard single
mode fiber; PPLN: periodically poled Lithium Niobate waveguide; Rx:
preamplified receiver.
40 Gb/s
Packetized
Data
MZM
CLK
EDFA
BPF
λ λ λ λ
Sig
Raman
Pumps
x 2
Packet
Erasure
Signal
λ λ λ λ
P1
1.5 nm
(3.5 nm)
Packet
Selection
Signal
MZM
PPLN-1
MZM
-7500
ps/nm
HNLF
`
PPLN-2
BPF
1.2nm
λ λ λ λ
P3
λ λ λ λ
P4
Rx
~70 km
SSMF + DCF
BPF
1.2nm
8.2 km SSMF
(4 km)
BPFs
2nm
BPF
1.2nm
(4 nm)
Phase
Conjugation 1
Phase Conjugation 2
(Reconfiguration Only)
λ λ λ λ
GP1
λ λ λ λ
GP2
Delay Selection Signal
(Reconfiguration Only)
2x2 Switch
EXTRACT PACKETS OPTICAL DELAY CONVERT BACK
PACKET DELETION
λ λ λ λ
P2
(Buffer Only)
TDL
TDL
(-3250)
DCF
TDL
TDL

112
10.3 Experimental Setup
Buffer
The experimental block diagram of the demonstrated buffer is shown in Figure
10-4. A 40 Gbit/s packetized input signal is generated by programming a pulse-
pattern generator (PPG). The packet stream is composed of 11, 424-bit-long (53
byte), ATM packets where packet slots 2, 6 and 11 are left open to allow buffered
packet to be reinserted. Initially, a 1 ns guard time is used by programming 40 zero
bits between the consecutive packets. A Mach-Zehnder modulator (MZM) is used
for data modulation of the optical carrier (λ
Sig
~ 1540.82 nm) with the packetized
data. Another MZM driven with a 40 GHz RF clock is used for full-rate pulse
carving to achieve a 50% RZ-OOK waveform. The input packet stream is first split
into two copies, one for the delay of the selected packet and the other for the deletion
of the selected packet as described in the previous section. In the first copy (upper
path) a packet (Packet-1) is selected to be delayed. This packet is extracted to the
required wavelength λ
PKT1
that will induce the desired delay. This is achieved by
using a two-pump wavelength conversion scheme in a periodically-poled Lithium-
Niobate (PPLN) waveguide, PPLN-1 [16]. The χ
(2)
:χ
(2)
processes of sum frequency
generation (SFG) between two pumps symmetrically located with respect to the
quasi-phase matching (QPM) wavelength of the PPLN waveguide followed by
difference frequency generation (DFG) between the sum signal and a continuous
wave (CW) pump in the PPLN waveguide results in the idler generated at the

113
frequency f
idler
= f
Pump1
+ f
Pump2
– f
CW
. We have modulated the DFG CW pump with a
MZM to generate the gated pump (λ
GP1
) for the packet extraction process. A PPG
synchronized with the input data is used to generate the gating signal which is on
only for the duration of Packet-1. The gated pump is then synchronized with Packet-
1 in PPLN-1 using an optical tunable delay line (TDL), leading to the extraction of
Packet-1 to λ
PKT1
by the f
PKT1
= f
Sig
+ f
P1
– f
GP1
relation. As the input signal is used as
a pump, the extracted packet is not phase conjugated. The PPLN-1 waveguide is 4
cm long and the QPM wavelength is tuned to 1551.6 nm. By tuning the gated pump
wavelength λ
GP1
, the wavelength Packet-1 is extracted to is tuned in the 1552.5 to
1560.2 nm range, corresponding to the maximum (116 ns: 10 packets) and minimum
(0 ns: 0 packet) delays, respectively. Packet-1 is then filtered by an optical bandpass
filter (OBPF) with a bandwidth of 1.5 nm and passed through -7500 ps/nm of
dispersion compensating fiber (DCF) (-180 ps/nm/km) which is Raman pumped with
two co-propagating pumps at 1450 nm and two counter-propagating pumps at 1460
nm (150 mW each) to mitigate the ~18 dB loss of the DCF. The dispersed packet is
then λ-converted to λ
PKT1
_
C
using degenerate four-wave mixing (FWM) in a highly
non-linear fiber (HNLF) for phase conjugation [105]. A CW pump and Packet-1
signal is amplified with EDFAs and filtered with 2 nm filters prior to launching into
the HNLF. A phase modulation method for Stimulated Brillouin Scattering (SBS)
suppression is not employed. A ~2.2 nm offset is kept between the phase conjugating
pump wavelength (λ
P2
) and λ
PKT1
for all different cases of λ
PKT1
corresponding to

114
different delay values. Hence, the phase conjugated signal wavelength, λ
PKT1
_
C
, is
ranging from ~1556.9 to ~1564.6 nm. The HNLF used for phase conjugation is ~330
m long with a nonlinear coefficient (γ) of 25 W
-1
·km
-1
, a zero dispersion wavelength
(ZDW) of 1562.2 nm, and a dispersion slope ~0.026 ps/nm
2
/km. The phase
conjugated Packet-1 is filtered by an OBPF with a 1.2 nm bandwidth, amplified, and
sent back through the DCF to complete the delay and compensate for the intra-
channel dispersion. An 8.2 km long SMF, corresponding to a total dispersion of
~140 ps/nm, is used to compensate for the residual dispersion offset caused by the
nonzero dispersion slope of the DCF. Following the delay and dispersion
compensation, Packet-1 is converted back to the original wavelength (λ
Sig
) using a
second PPLN waveguide, PPLN-2, with a QPM wavelength of 1552.8 nm. A two
pump conversion scheme similar to the packet extraction stage is used for
wavelength conversion in PPLN-2. Two pump lasers (λ
P3
, λ
P4
) are tuned according
to λ
PKT1_C
such that the converted output signal is always at λ
Sig
. This signal, delayed
Packet-1 at λ
Sig
, is then filtered with a 1.2 nm filter and finally combined with the
signal from the lower path by a 3 dB coupler. We used two different mediums for
wavelength conversions in the buffer, where ideally similar medium could be used in
all wavelength conversion stages to realize the buffer. The wavelength conversion
techniques are polarization dependent and we used polarization controllers to
maximize the conversion efficiency for each stage.

115
The second copy of the input packet stream is sent to the lower path for deletion
of packets. It might be desirable to delete the packet(s) being delayed from the
original time slots; however, any unnecessary packet(s) can be deleted from the
stream to create open time slots. An MZM modulator is used for the deletion process
driven by an RF signal generated by a PPG synchronized with the data. The packet
to be deleted is time aligned with the PPG signal using a TDL as shown in Figure
10-4. After the MZM, the packet stream is sent through 60 km of SMF, followed by
10 km DCF (-1005 ps/nm) to emulate the time synchronization with the upper path.
An EDFA is used to amplify the signal to match the power of the signal combined
from the upper path. A TDL is used to match the two path lengths exactly (within the
repeating 11 packet data stream) for the faster wavelength of the selected packet(s).
The time alignments of the two paths are adjusted such that this wavelength results
in no relative delay with respect to the original packet time slot. Thus, a reference
(zero delay) is achieved for the buffer. The resulting output packet stream is then
sent to a pre-amplified receiver for bit error rate (BER) measurements. A 1.2 nm
filter is used in the pre-amplified receiver following the low noise EDFA.
Reconfiguration
For the demonstration of reconfigurability, the experimental setup is slightly
modified. The reconfiguration experiment is conducted for two different guard times:
1 ns (40 bits) and 25 ps (1 bit) by programming the input packet stream in the PPG.
In the first wavelength conversion stage, a 2x2 20 GHz Lithium Niobate switch is

116
utilized to switch between the dummy pump lasers. The switch is driven with a
programmed PPG synchronized with the input signal clock. The output of the switch
is sent to an MZM to gate the pumps such that they only exist during the duration of
the two selected packets. The transition between the pump lasers occurs within the
guard time between the packets. Hence, two consecutive packets in the data stream
will be wavelength converted by two different DFG pumps, with the f
PKT2
= f
Sig
+ f
P1

– f
GP2,
and f
PKT3
= f
Sig
+ f
P1
– f
GP3
. An OBPF (bandwidth of 3.5 nm) is used to filter
the extracted packets. Packets are then sent through the DCF (-3250 ps/nm). The
QPM of the PPLN-1 is tuned to 1551.6 nm. The total power lunched to the PPLN-1
is kept under 27 dBm. The signal after the DCF is split into two copies to realize the
phase conjugation on each packet with two independent phase conjugation stages as
shown in Figure 10-4 with dotted lines. The same phase conjugation setup mentioned
in the previous section is used for one of the packets. For the second phase
conjugation a 100 m HNLF with γ = 25 W
-1
·km
-1
, ZDW = 1558.2 nm, and s = 0.026
ps/nm
2
/km is used. A TDL is used to match the path lengths of the two phase
conjugation stages. After the phase conjugation, the signals are combined and sent
back through the DCF as discussed before. After the DCF, a 4 km spool of SSMF is
used for the residual dispersion compensation. A third wavelength conversion stage
is not employed for the reconfiguration experiments. However, multiplexing of these
packets to a single wavelength in a PPLN is possible as shown in [102]. The
dispersion compensated packets are then combined with the packet stream from the

117
lower path and sent to the receiver. Delayed packets are not deleted from the original
packet stream in the lower path. In the receiver no filters are used as the multiplexed
signal is present on multiple wavelengths.

Figure 10-5: Packet-1 being buffered from time slot 1 to time slot 11. Eye
diagrams of the signal shown are also given. (a) 40 Gbit/s input packet stream
(424 bits/packet, 1 ns guard time); (b) Packet-1 after extraction to λPKT1
(~1552.5 nm) in PPLN-1. (c) Packet-1 after double passing through the DCF and
after SMF, signal is at λPKT_C (~1556.9 nm) due to phase conjugation; (d)
Output packet stream where delayed Packet-1 is converted to λSig and original
Packet-1 is deleted from time slot 1 in the lower path.
10.4 Results and Discussion
Buffer Results
Figure 10-5 shows the evolution of a packet (Packet-1) in the buffer for the
longest delay value along with the corresponding eye diagrams. The corresponding
spectra for each wavelength conversion stage are shown in Figure 10-6. Figure
10-5(a) is the 11 packet-long input stream to the buffer where the guard time
(a)
(b)
(c)
(d)
Packet-1
Extracted
Packet-1
Delayed
Input Packets
Output Packets
6 11
15 ns/div
1
1
2
1
1
Δ Δ Δ Δt = 116 ns
1
6 2
λ λ λ λ
PKT1
λ λ λ λ
PKT1_C
λ λ λ λ
Sig
λ λ λ λ
Sig

118
between the packets is 1 ns. Packet-1 is extracted to λ
PKT1
= ~1552.5 nm in PPLN-1
as shown in Figure 10-5(b). The spectrum for the extraction process is shown in
Figure 10-6(a). The average power of the Packet-1 signal is ~ -13 dBm. As this
signal is off for 10/11 of the duration, the peak power of the signal is ~10 dB higher
than the average power. Packet-1 at λ
PKT1
is then sent through the DCF and then
phase conjugated to λ
PKT1_C
as shown in Figure 10-6(b). Conversion efficiency for
the degenerate FWM process is ~ -16 dB. After the second pass through the DCF the
packet is delayed by 116 ns. The delayed and dispersion compensated Packet-1 after
the 8.2 km SMF is shown in Figure 10-5(c). Following this step, Packet-1 is
wavelength converted from λ
PKT1_C
back to the input signal wavelength λ
Sig
in
PPLN-2 as shown in Figure 10-6(c). In the lower path Packet-1 is deleted from the
input stream by the MZM. Both paths are then coupled together forming the output
packet stream as shown in Figure 10-5(d). As shown, Packet-1 is deleted from time
slot 1 and buffered to the 11
th
time slot.

119

Figure 10-6: Experimental spectra of the wavelength conversion processes in the
buffer. (a) Packet extraction in the PPLN-1 with gated pump λGP1 (1550.6 nm).
PPLN-1 QPM wavelength is shown with a dotted line (~1551.6 nm); (b) Phase
conjugation in the HNLF; (c) Delayed Packet-1 is wavelength converted back to
λSig in PPLN-2. PPLN-2 QPM is shown with the dotted line and is at ~1552.7
nm. The scale is the same, 8 dB/div and 3 nm/div, for all plots.
The relative delay achieved by the wavelength conversion/dispersion process is
shown in Figure 10-7(a). The maximum relative delay achieved is 116 ns which
corresponds to a buffer depth of 10 packets including the 1 ns buffer time between
packets (1 packet = 424 data bits = 10.6 ns). The granularity of the delay is limited
by the granularity of the pump lasers which is 1 pm. This results in a delay resolution
of 15 ps with the amount of dispersion used. The buffer is tested for various delay
values. The set of delays with λ
PKT1
= 1560.23, 1559.46, 1556.36, and 1552.5 nm,
1537.7nm 1567.7nm 1552.7nm
(a) PPLN–1:
(b) HNLF:
λ λ λ λ
Sig
λ λ λ λ
P1
λ λ λ λ
GP1
λ λ λ λ
PKT1
λ λ λ λ
PKT
1
λ λ λ λ
PKT1_C
λ λ λ λ
P3 λ λ λ λ
P4
λ λ λ λ
S
λ λ λ λ
P2
8 dB/div
3 nm/div
λ λ λ λ
PKT1_C
(c) PPLN–2:

120
corresponding to delays of 0, 1, 5 and 10 packets, respectively, are shown in Figure
10-7(b).

Figure 10-7: (a) Relative delay achieved for the system for a 40 Gbit/s input
signal; (b) Output packet stream for various buffering scenarios including zero
and maximum delay.
In all cases, delayed Packet-1 is converted back to the original wavelength (λ
Sig
)
in PPLN-2 by tuning the pumps λ
P3
and λ
P4
. The complete set of wavelength
conversion spectra for all three stages for minimum, middle, and maximum delays
are shown in Figure 10-8(a), (b) and (c), respectively. The wavelength of extracted
Packet-1 is controlled by the wavelength of λ
GP1
as shown. A ~3.3 nm wavelength
separation between the λ
PKT1
and λ
PKT1_C
is kept for all delay cases as shown in the
second row. The third wavelength conversion process converts the delayed packet
back to the input wavelength (λ
Sig
) for each delay value, as shown in the third row of
5-Packet Delay (Δ Δ Δ Δt = 58 ns)
1-Packet Delay (Δ Δ Δ Δt = 11.6 ns)
(b) (a)
10-Packet Delay (Δ Δ Δ Δt = 116 ns)
6 11 2
1
1
1
1
11
6
2 11
1
1
2 1 6
0-Packet Delay (Δ Δ Δ Δt = 0 ns)
0
20
40
60
80
100
120
1550 1555 1560 1565
Relative Delay (ns)
Converted Wavelength (nm)

121
Figure 10-8. In the cases of maximum (a) and middle (b) delay, the third stage
wavelength conversion uses the Packet-1 signal as one of the pumps for SFG
process, which is a not phase conjugating process. For the zero delay case (c) the
packet signal is used for the DFG process in a different two pump configuration,
since a dummy laser as used in (a) and (b) would overlap with the signal itself.
Instead, the signal is used as the dummy signal for the DFG process. This is a phase
conjugation process as opposed to (a) and (b), however it does not change the buffer
application. In general for the buffer, only the second stage wavelength conversion
needs to be phase conjugating for compensation of the dispersion. The delayed
Packet-1 at λ
Sig
is then combined with the lower path signal where Packet-1 is
deleted from the original time slot to generate the output packet stream, with Packet-
1 buffered to the desired time slot, as given in Figure 10-7(b).

122

Figure 10-8: Experimental spectra of the three wavelength conversion stages for
the cases of: (a) maximum (116 ns), (b) middle, and (c) zero delay. The first row
shows the packet extraction in PPLN-1, the second row shows the phase
conjugation in HNLF, and the third row shows the wavelength conversion of the
delayed Packet-1 to the original wavelength in PPLN-2. For all plots, the center
wavelength is 1552.7 nm and the scale is 3 nm/div for the horizontal and 8 dB/div
for the vertical axis.
BER measurements are performed to characterize system performance for
various delay values as shown in Figure 10-9. The BER test-set is programmed for
the new packet stream generated at the buffer output. The BER performance of the
buffer for delay values of 0 ns, 58 ns, and 116 ns are shown. We observe ~3 dB, ~2
and ~0.5dB power penalties at BERs of 10
-9
for 0-packet, 5-packet, and 10-packet
delays, respectively. We believe proximity to the high-power pump induces a higher
penalty for the 0-packet delay due to the non-ideal roll-off profile of the filter used in
the packet extraction stage. This causes some pump power to leak as the pump is not
Packet-1
Extraction
(PPLN-1)
Phase
Conjugation
(HNLF)
Conversion
Back to λ λ λ λ
Sig
(PPLN-2)
(a)
(b) (c)
λ λ λ λ
Sig
λ λ λ λ
P1
λ λ λ λ
GP1
λ λ λ λ
PKT1
λ λ λ λ
Sig
λ λ λ λ
P1 λ λ λ λ
GP1
λ λ λ λ
PKT1
λ λ λ λ
Sig
λ λ λ λ
P1
λ λ λ λ
GP1
λ λ λ λ
PKT1
λ λ λ λ
PKT1
λ λ λ λ
PKT1
λ λ λ λ
PKT1
λ λ λ λ
PKT1_C
λ λ λ λ
PKT1_C
λ λ λ λ
PKT1_C
λ λ λ λ
PKT1_C λ λ λ λ
PKT1_C
λ λ λ λ
PKT1_C
λ λ λ λ
P4
λ λ λ λ
P4
λ λ λ λ
P4
λ λ λ λ
Sig
λ λ λ λ
Sig
λ λ λ λ
Sig
λ λ λ λ
P3
λ λ λ λ
P3 λ λ λ λ
P3

123
perfectly suppressed. In addition to this, residual dispersion varies for different delay
values as the SMF used is fixed length and will perfectly compensate only for a
particular wavelength. Hence, the uncompensated residual dispersion gives an
additional power penalty for the 0-packet delay case.

Figure 10-9: BER performances for several buffering scenarios. Back-to-back
performance and BER performance of the signal at the output of lower path (with
Packet-1 deleted) is also given for comparison.
-LOG
10
(BER)
Received Power (dBm)
-42 -40 -38 -36 -34 -32 -30 -28 -28
2
3
4
5
6
7
8
9

After Deletion
10 Packet Delay
5 Packet Delay
0 Packet Delay
Back-To-Back
After Deletion

124

Figure 10-10: Packets-2 and -3 being buffered by three and five time slots in the
reconfiguration experiment. (a) The input packet sequence of 8 packets. The
guard time between the Packet-2 and Packet-3 is 25 ps. The inset shows the guard
time; (b) An illustration of the gated pumps generated by the switch and the MZM
in the packet extraction stage; (c) Extracted Packets 2 and 3; (d) Packets 2 and 3
after the second pass through the DCF; (e) Output packet sequence where Packets
2 and 3 are inserted at the corresponding time slots.
Reconfiguration Results
Rapid reconfiguration times from 1 ns down to 25 ps are also demonstrated using
a 20 GHz 2x2 Lithium Niobate switch. The input packet stream of 8 packets is
shown in Figure 10-10(a) where Packets 2 and 3 are going to be buffered to time
slots #5 and #7, respectively. The inset shows the buffer time of 25 ps between
Packets 2 and 3. As only Packets 2 and 3 are going to be delayed, the guard time
between the rest of the packets is kept at 1ns. A PPG toggles the switch between two
pump lasers (λ
GP1
~ 1556.0 nm, λ
GP2
~ 1557.84 nm) in the packet extraction stage
(PPLN-1) as illustrated in Figure 10-10(b). The pump switching signal and the
6
5 7
2 3
3
Input Packets
10 ns/div
Gated
Pumps
λ
GP2
Packets 2&3
Extracted
λ
PKT2
Delayed Packets
Output
8
4 3 1
3 6 8 2 3 1 4 2
(a)
(b)
(c)
(d)
(e)
2
λ
GP1
λ
PKT3
2
2 3
Δ Δ Δ Δt
PKT2
= 34.8 ns
Δ Δ Δ Δt
PKT3
= 46.4 ns

125
clocking signal are arranged such that only Packet-2 and Packet-3 are extracted to
two different wavelengths, λ
PKT2
(~1547.16 nm) and λ
PKT3
(~1545.36 nm),
respectively. The extracted packet stream is shown in Figure 10-10(c). Note that the
two extracted packets are at different wavelengths.

Figure 10-11: Experimental spectra of the wavelength conversion processes in the
buffer. (a) Packet extraction in the PPLN-1 with gated pump λGP1 (1550.6 nm).
PPLN-1 QPM wavelength is shown with a dotted line (~1551.6 nm); (b) Phase
conjugation in the HNLF; (c) Delayed Packet-1 is wavelength converted back to
λSig. PPLN-2 QPM is shown with the dotted line and is at ~1552.7 nm.
The experimental spectra for the reconfiguration experiment with 25 ps guard
time are shown in Figure 10-11. Figure 10-11(a) shows the spectra of the cascaded
SFG-DFG process with the two gated pumps in PPLN-1. Packet-2 and Packet-3
experience different amounts of group delay (λt
PKT2
= 17.4 ns, λt
PKT3
=23.2 ns) in the
λ λ λ λ
S
λ λ λ λ
P1
λ λ λ λ
GP1
λ λ λ λ
PKT2
λ λ λ λ
GP2 λ λ λ λ
PKT3
1537.5nm 1567.5nm 1552.5nm
λ λ λ λ
PKT2
λ λ λ λ
PKT2_C
λ λ λ λ
PKT3
λ λ λ λ
PKT3_C
8 dB/div
3 nm/div
(a) PPLN–1:
(b) HNLF-1:
4 nm
Filter
(c) HNLF-2:

126
DCF due to the difference in wavelength (Δλ= λ
PKT2
- λ
PKT3
= 1.8 nm). Packet-3,
which is at a shorter wavelength, experiences a larger group delay. By using two
conjugation stages, the two packets are separately conjugated to λ
PKT2
_
C
(~1553.9
nm) and λ
PKT3
_
C
(~1552.1 nm) such that the wavelength difference between the two
packets is preserved (Δλ= λ
PKT2_C
- λ
PKT3_C
= 1.8 nm). The experimental spectra of
the phase conjugation process for both packets are shown in Figure 10-11(b) and (c).
As Packet-3 is still at a shorter wavelength, the relative delay between the packets
accumulates (λt
PKT2
= 34.8 ns, λt
PKT2
= 46.4 ns) when the two signals are sent back
through the DCF for dispersion compensation. If a single phase conjugation stage
with a single pump were used for both signals, Packet 3 would be at a longer
wavelength after phase conjugation (Δλ= -1.8 nm), hence cancelling the relative
delay during the second pass through the DCF. The signals at the output of the phase
conjugation stages are combined after timing of the both arms are adjusted by a TDL
on one of the phase conjugation stages. This is required as the equipment used in the
phase conjugation stages is not identical. The time alignment is checked by tuning
the gated pumps to identical wavelengths such that Packet-2 and Packet-3 experience
the same relative delay. After combining two signals, phase conjugated Packet-2 and
Packet-3 are filtered off by a 4 nm filter. Then, they are sent back through the DCF.
After this second pass through the DCF, the delayed packets are coupled with the
signal from the lower path. Figure 10-11(e) shows the output packet sequence and
eye diagram at the output. As the guard time between the other packets is 1 ns, the

127
guard time between Packet-2 and Packet-6 (as well as between packet 6 and packet
3) is 500 ps at the output packet stream. The same experiment is repeated for a guard
time of 1 ns between Packets 2 and 3. In the reconfiguration experiments, Packet-2
and Packet-3 are not deleted from their original time slots in the lower path. Thus,
the original copies along with the delayed copies coexist in the output packet stream
as in Figure 10-11(e).

Figure 10-12: Transient response of the 2x2 Lithium Niobate switch used for
toggling between pump lasers in packet extraction. The rise/fall time is ~25 ps for
both output ports.
The transient response of the 2x2 Lithium Niobate switch is shown in Figure
10-12. The rise (10%) and fall (90%) times for the switch are measured to be ~25 ps
for both output ports. This, therefore, limits the guard time to be 25 ps, or one bit at
40 Gbit/s. A zero guard time allows one bit to leak into both packets, in the packet
extraction process, due to the poor extinction ratio. However, this suggests that zero
guard time may be achievable by using a larger bandwidth switch.
20 ps/div
Output Port 1 Output Port 2

128

Figure 10-13: Last and first several bits of the Packets 2 and 3, respectively: (a)
Packet-2 and Packet-3 from the input packet stream with a guard time (i) 25 ps,
(ii) 1 ns; (b) Packet-2 and Packet-3 after extraction in the PPLN-1. For Packet-2,
Packet-3 extraction pump is turned off for demonstration purposes; (c) Packet-2
and Packet-3 after the delay (before the combination with the lower arm) when
both gated pumps are on. The scale is 100 ps/div for all the plots except (a)-(ii).
Figure 10-13(a) shows the guard time between the Packet-2 and Packet-3 and the
first and last several bits for 25 ps (i) and 1 ns (ii) cases. The edges of the two
packets after the extraction are also shown in Figure 10-13(b). For demonstration of
the gating effect of the switch, the corresponding pump of a packet is turned off in
each case. Therefore, Figure 10-13(b) plots only show one of the extracted packets
while the other one is not extracted due to the gating effect of the switch. Figure
10-13(c) shows corresponding edges of the packets after the delay before combining
with the lower arm of the packet buffer. However, for this case both of the extracting
(a)
(b)
(c)
Packet-2 Packet-3
Packet-2
Packet-2
Packet-3
Packet-3
100 ps/div
Packet-3 Packet-2
Turned Off Turned Off
200 ps/div
Packet-2 Packet-3
(i) (ii)

129
pumps are on, and packets are simultaneously delayed by λt
PKT2
= 34.8 ns, and λt
PKT2
= 46.4 ns. Due to the delay the neighboring time slot appears empty in Figure
10-13(c).
The effect of the reconfiguration setup is characterized with BER measurements
as well. The output bit pattern is programmed to the error detector for each
experiment with different guard times. As described before no filters were used in
the receiver for this experiment. Therefore, the receiver is characterized with a new
back-to-back BER measurement. BER measurements for the reconfiguration
experiments with guard times of 1 ns and 25 ps are given in Figure 10-14.
Reconfiguration of the packet buffer shows power penalties of ~4.1 dB and ~5 dB
for guard times of 1 ns and 25 ps, respectively. Compared to the buffer only case, 1
ns reconfiguration introduces an extra 1.1 dB penalty, which we believe is due to the
lowered wavelength conversion efficiency of the packet extraction stage in the multi-
packet scheme as the gated pump powers are decreased due to the insertion of the
2x2 switch prior to the amplification by the fixed gain EDFA. The additional 0.9 dB
penalty for the 25 ps reconfiguration case may be attributed finite extinction ratio
(~25 dB) and the limited transient response of the 2x2 optical switch.

130

Figure 10-14: BER performances of the buffer reconfiguration experiment for
guard times of 1 ns and 25 ps.
10.5 Conclusion
We have demonstrated an optical buffer that is capable of buffering ATM
packets up to 10 packet length with a tunable 116 ns delay achieved by
conversion/dispersion technique. The relative delay used for the buffer is
continuously tunable with 15 ps granularity. Thus, the packet buffer can be tuned to
support different bit rates and packet lengths. As the wavelength conversion stages
used in the buffer preserve the phase information, this buffer has the potential to be
used for phase encoded packets as well [66, 7, 22]. Additionally, the
reconfigurability of the buffer is tested. Using a fast optical switch to toggle between
the converted wavelengths that define the delay, we showed that a reconfiguration
time as fast as 25 ps are potentially achievable for buffers with conversion/dispersion
based delays.
-LOG
10
(BER)
Received Power (dBm)
-35 -32 -29 -26 -23 -20
2
3
4
5
6
7
8
9

With Reconfiguration - 25 ps
Back-To-Back
With Reconfiguration - 1 ns

131
Conclusion

Optical amplitude and phase based modulation formats, more specifically QPSK
and QAM combined with polarization multiplexing, have emerged as the most
popular methods for transmitting information across long-haul and ultra-long haul
optical transmission systems. In addition to advanced modulation formats, optical
signal processing is expected to play a key role in future optical systems. In
particular, the ability to accommodate amplitude and phase based formats is of great
interest. The most basic building block for optical signal processing is a tunable
optical delay line.
In this dissertation, we present a number of optical signal processing
functionalities aimed at accommodating Pol-MUX signals and a variety of
modulation formats. In particular, the development of a large tunable optical delay is
presented.
We introduce and experimentally demonstrate all optical WDM to Pol-MUX
multiplexing and Pol-MUX to WDM demultiplexing. A single 100 Gbit/s Pol-MUX
signal is decomposed into two 50 Gbit/s WDM channels for easy processing.
Similarly two 50 Gbit/s WDM channels are combined into a single 100 Gbit/s Pol-
MUX channel. In this manner, WDM channels can be combined into a single Pol-
MUX channel to double their spectral efficiency while a Pol-MUX channel can be
demultiplexed for use in a traditional WDM network or for independent routing or

132
processing of its sub-channels without the need for detection and electronic
processing.
We introduce and experimentally wavelength conversion of both single
polarization 40-Gbuad 16-QAM and 20-Gbuad polarization-division-multiplexed
(PDM) 16-QAM in a PPLN waveguide. A polarization insensitive scheme utilizing
bi-directional operation is employed. The effects of pump-depletion on the converted
signal are measured.
We introduce and experimentally demonstrate a method for achieving a variable
optical delay element using wavelength conversion in HNLF, dispersion
compensating fiber, and intra-channel dispersion compensation. A delay of up to 503
ns is demonstrated using 40 Gbit/s RZ-DPSK and 40 Gbit/s RZ-OOK modulation
formats.
We introduce and experimentally demonstrate a novel method for extending the
range of our tunable delay. Utilizing mid-span optical phase conjugation, dispersion
compensation over a much wider wavelength range is achieved. A continuous delay
of up to 1.16-μs equaling >55,000 symbols at 50 Gbit/s, for 100 Gbit/s NRZ-DQPSK
and 50 Gbit/s NRZ-DPSK modulation formats, is demonstrated.
We introduce and experimentally demonstrate a method for further extending the
range of our delay using a spatial light modulator (SLM) and fiber-Bragg-gratings
for dispersion slope compensation to enable 100-Gbit/s operation. A continuous
delay of up to 3.6 μs for 100, 80, and 20-Gbit/s differential quadrature phase-shift-

133
keyed (DQSPK) and 50, 40 and 10-Gbit/s differential phase-shift-keyed (DPSK)
waveforms is shown. A time-delay bit-rate product of ~360,000 for 100-Gbit/s
DQPSK with wavelength-maintaining operation is achieved.
We introduce and experimentally demonstrate a method for fine tuning of optical
delays. Using cascaded acousto-optic modulators to improve wavelength control, and
thus delay resolution, by five orders of magnitude compared to a 1-pm tunable laser.
A 256-ns delay with <0.5-ps resolution is shown for 40-Gbit/s RZ-OOK with no
added penalty.
Lastly, an application of an optical delay as continuously tunable, all-optical
packet buffer based on conversion dispersion delays is presented. 40 Gbit/s
Asynchronous Transfer Mode (ATM) packets with return-to-zero on-off keying (RZ-
OOK) data are buffered up to 10-packet length (116 ns). The packet buffer
performance is characterized for several delay values. Reconfiguration of the packet
buffer is also investigated and reconfiguration times as fast as 25 ps are shown by
using a high speed optical switch to toggle between wavelength conversion pumps.

134
References

[1] A. Agarwal, et. al., "All-optical loadable and erasable storage buffer based
on parametric nonlinearity in fiber," Journal of Lightwave Technology,
vol. 23, pp. 2229-2237, July 2005.
[2] G. P. Agrawal, Fiber-Optic Communication Systems, 3rd ed. New York,
USA: Wiley, 2002.
[3] G. P. Agrawal, Nonlinear Fiber Optics.: Springer Berlin / Heidelberg, 2000.
[4] N. Alic, et. al., "1.83-μs Wavelength-Transparent All-Optical Delay," in
Conference on Optical Fiber Communications, 2009, p. PDPA1.
[5] N. Alic, et. al., "Microsecond Parametric Optical Delays," Journal of
Lightwave Technology, vol. 28, pp. 448-455, 2010.
[6] N. Alic, et. al., "105-ns Continuously Tunable Delay of 10-Gbit/s Optical
Signal," Photonics Technology Letters, vol. 20, pp. 1187-1189, 2008.
[7] P. A. Andersen, et. al., "Wavelength conversion of a 40-Gbit/s RZ-DPSK
signal using four-wave mixing in a dispersion-flattened highly nonlinear
photonic crystal fiber," Photon. Technol. Lett, vol. 17, no. 9, pp. 1908-
1910, September 2005.
[8] J. R. Barry, et. al., "Carrier synchronization for homodyne and heterodyne
detection of optical quadriphase-shift keying," Journal of Lightwave
Technology, vol. 10, pp. 1939–1951, 1992.
[9] P. Boffi, et. al., "Measurement of PMD tolerance in 40-Gbit/s polarization-
multiplexed RZ-DQPSK," Optics Express, vol. 16, pp. 13398-13404,
August 2008.
[10] P. Boffi, et. al., "Stable 100-Gbit/s POLMUX-DQPSK Transmission with
Automatic Polarization Stabilization," Photonics Technology Letters, vol.
21, no. 11, pp. 745-747, 2008.
[11] J. M. Chavez Boggio, et. al., "730-nm optical parametric conversion from
near- to short-wave infrared band," Optics Express, vol. 16, pp. 5435-
5443, 2008.
[12] A. Bogoni, et. al., "640Gbit/s All-Optical Regeneration in a PPLN
Waveguide," in Photonics Society Annual Meeting, 2010, p. TuM3.

135
[13] D. van den Borne, et. al., "1.6-b/s/Hz spectrally efficient transmission over
1700 km of SSMF using 40 x 85.6 Gbit/s POLMUX-RZ-DQPSK,"
Journal of Lightwave Technologies, vol. 25, pp. 222-232, January 2007.
[14] S. Chandrasekhar, et. al., "Hybrid 107-Gbit/s polarization-multiplexed
DQPSK and 42.7-Gbit/s DQPSK transmission at 1.4-bits/s/Hz spectral
efficiency over 1280 km of SSMF and 4 bandwidth-managed ROADMs,"
in Proceedings of European Conference on Optical Communication,
Berlin, 2007, p. PD1.92.
[15] N. Chi, et. al., "Experimental characteristics of optical crosspoint switch
matrix and its applications in optical packet switching," Journal
Lightwave Technology, vol. 24, pp. 3646-3653, October 2006.
[16] M. H. Chou, et. al., "1.5- m-Band Wavelength Conversion Based on
Cascaded Second-Order Nonlinearity in LiNbO Waveguides," Photonics
Technology Letters, vol. 11, pp. 653-655, 1999.
[17] L. Christen, et. al., "Tunable 105-ns Optical Delay for 80-Gbit/s RZ-
DQPSK, 40-Gbit/s RZ-DPSK, and 40-Gbit/s RZ-OOK Signals using
Wavelength Conversion and Chromatic Dispersion," in Optical Fiber
Communications Conference, 2008, p. OTuD1.
[18] R. Cowper, et. al., "Perspective on the Evolution of Next Generation High-
Capacity Systems," Nortel Networks Inc., 1999.
[19] Y. Dai, et. al., "1 μs tunable delay using parametric mixing and optical phase
conjugation in Si waveguides," Optics Express, vol. 17, pp. 7004-7010,
2009.
[20] M. Daikoku, et. al., "100-Gbit/s DQPSK Transmission Experiment Without
OTDM for 100G Ethernet Transport," Journal of Lightwave
Technologies, vol. 25, pp. 139-145, January 2007.
[21] Y. Dai, et. al., "Ultralong continuously tunable parametric delay via a
cascading of discrete stages," Optics Express, vol. 18, pp. 333-339, 2010.
[22] P. Devgan, et. al., "Highly Efficient Multichannel Wavelength Conversion
of DPSK Signals," J. Lightw. Technol, vol. 24, pp. 3677-3682, 2006.
[23] M. K. Dhodhi, et. al., "Bottlenecks in next generation DWDM-based optical
networks," Journal on Computer Communications, vol. 24, pp. 1726–
1733, 2001.
[24] C. R. Doerr, et. al., "A single-chip optical equalizer enabling 107-Gbit/s
optical non-return-to-zero signal generation," in Proceedings of European
Conference on Optical Communication, Glasgow, 2005, pp. 13-14.

136
[25] H. J. S. Dorren, et. al., "Optical packet switching and buffering by using all-
optical signal processing methods," Journal of Lightwave Technology,
vol. 21, pp. 2-12, January 2003.
[26] M. Duelk, "Next-generation 100G Ethernet," in Proceedings of European
Conference on Optical Communication, Glasgow, M. Duelk, “Next-
generation 100G Ethernet,” in Proc. of European Conference on Optical
Communication, Glasgow, Sept. 2005, Technical Digest CD (2005),
paper Tu3.1.2. 2005, p. paper Tu3.1.2.
[27] I. Fazal, et. al., "Optical data packet synchronization and multiplexing using
a tunable optical delay based on wavelength conversion and inter-channel
chromatic dispersion," Opt. Express, vol. 15, pp. 10492-10497, 2007.
[28] C. R. S. Fludger, et. al., "Coherent Equalization and POLMUX-RZ-DQPSK
for Robust 100-GE Transmission," Journal of Lightwave Technologies,
vol. 26, pp. 64-72, January 2008.
[29] C. R. S. Fludger, et. al., "Towards Robust 100G Ethernet Transmission," in
LEOS Summer Topical Meetings, 2007, pp. 224-225.
[30] M. Fok, et. al., "Performance Investigation of One-to-Six Wavelength
Multicasting of ASK–DPSK Signal in a Highly Nonlinear Bismuth Oxide
Fiber," Journal of Lightwave Technology, vol. 27, no. 15, 2009.
[31] N. K. Fontaine, et. al., "Continuously tunable optical buffering at 40 Gbit/s
for optical packet switching networks," Journal of Lightwave
Technology, vol. 26, no. 23, pp. 3776-3783, December 2008.
[32] M. A. Foster, et. al., "Broad-band continuous-wave parametric wavelength
conversion in silicon nanowaveguides," Optics Express, vol. 15, pp.
12949–12958, 2007.
[33] G. Gavioli, et. al., "100Gbit/s WDM NRZ-PM-QPSK Long-Haul
Transmission Experiment over Installed Fiber Probing Non-Linear Reach
With and Without DCUs," in European Conference on Optical
Communications, 2009.
[34] J. M. Geist, "Capacity and cutoff rate for dense M-ary PSK constellations,"
in MILCOM, Monterey, 1990, pp. 768-770.
[35] A. H. Gnauck, "Linear and Nonlinear Performance of 42.7-Gbit/s Single-
Polarization RZ-DQPSK Format," Photonic Technology Letters, vol. 18,
2006.
[36] A. H. Gnauck, et. al., "10-Gbit/s 360-km transmission over dispersive fiber
using midsystem spectral inversion," Photonic Technology Letters, vol. 5,
no. 6, pp. 663-666, June 1993.

137
[37] A. H. Gnauck, et. al., "Optical Phase-Shift-Keyed Transmission," Journal of
Lightwave Technology, vol. 23, no. 1, 2005.
[38] R. A. Griffin, "10 Gbit/s optical differential quadrature phase shift keying
(DQPSK) Transmission using GaAs/AlGaAs Integration," in Proceedings
of Optical Fiber Communications Conference 2002, 2002, pp. Paper
FD6-1.
[39] K.-P. Ho, "Exact evaluation of the capacity for intensity-modulated direct-
detection channels with optical amplifier noises," Photonic Technology
Letters, vol. 17, pp. 858-860, 2005.
[40] K.-P. Ho, "Generation of Arbitrary Quadrature Signals Using One Dual-
Drive Modulator," Journal of Lightwave Technology, vol. 23, no. 2, 2005.
[41] J. Hongo, et. al., "1-Gsymbol/s 64-QAM coherent optical transmission over
150 km," Photonic Technology Letters, vol. 19, pp. 638-640, 2007.
[42] H. Hu, et. al., "Polarization Insensitive All-Optical Wavelength Conversion
of 320 Gbit/s RZ-DQPSK Data Signals," in Optical Fiber
Communications Conference, 2009.
[43] E. Ip, et. al., "Carrier synchronization for 3- and 4-bit-per-symbol optical
transmission," Journal of Lightwave Technologies, vol. 23, pp. 4110–
4124, 2005.
[44] E. Ip, et. al., "Coherent detection in optical fiber systems," Optics Express,
vol. 16, pp. 753-791, 2008.
[45] V. Jungnickel, et. al., "1 Gbit/s MIMO-OFDM transmission experiments,"
in IEEE Conference on Vehicular Technology, Dallas, 2005, pp. 861–
866.
[46] J. M. Kahn, et. al., "Spectral efficiency limits and modulation/detection
techniques for DWDM Systems," Journal of Selected Topics in Quantum
Electronics, vol. 10, pp. 259–271, 2004.
[47] K. Kikuchi, "Coherent detection of phase-shift keying signals using digital
carrier-phase estimation," in Conference on Optical Fiber
Communications, Anaheim, 2006, p. Paper OTuI4.
[48] H. Kim, et. al., "Robustness to laser frequency offset in direct-detection
DPSK and DQPSK systems," Journal of Lightwave Technologies, vol.
21, pp. 1887-1891, 2003.
[49] T. Kurosu, et. al., "Continuously tunable 22 ns delay for wideband optical
signals using a parametric delay-dispersion tuner," Optics Letters, vol. 34,
pp. 1441-1443, 2009.

138
[50] C. Langrock, et. al., "All-Optical Signal Processing Using χ(2)
Nonlinearities in Guided-Wave Devices," J. Lightwave Technol., vol. 24,
pp. 2579- 2581, 2006.
[51] A. Leven, et. al., "Coherent receivers for practical optical communication
systems," in Conference on Optical Fiber Communications, Anaheim,
2007, p. OThK4.
[52] Q. Lin, et. al., "40-Gbit/s optical switching and wavelength multicasting in a
two-pump parametric device," Photonic Technology Letters, vol. 17, no.
11, pp. 2376-2378, November 2005.
[53] X. Liu, "Single Coherent Detection of a 606-Gb/s CO-FDM Signal with 32-
QAM Subcarrier Modulation Using 4x 80-Gsamples/s ADCs," in ECOC,
2010, p. paper PD2.6.
[54] G.-W. Lu, et. al., "Optical Phase Add/Drop for Format Conversion between
DQPSK and DPSK," in Optical Fiber Communications Conference,
2009.
[55] C. Malouin, et. al., "DPSK Receiver Design - Optical Filtering
Considerations," in Proceedings of Optical Fiber Communications
Conference 2007, C. Malouin, J. Bennike, and T. Schmidt, "DPSK
Receiver Design - Optical Filtering Considerations," in Proceedings of
Optical Fiber Communications Conference 2007, p. 3. 2007, pp. 3-5.
[56] P. Martelli, et. al., "All-Optical Wavelength Conversion of a 100-Gbit/s
Polarization-Multiplexed Signal," Optics Express, vol. 17, no. 20, 2009.
[57] T. Matsuda, "Comparison Between NRZ and RZ Signal Formats for In-Line
Amplifier Transmission in the Zero Dispersion Regime," Journal of
Lightwave Technology, 1998.
[58] P. P. Mitra, et. al., "Nonlinear limits to the information capacity of optical
fiber communications," Nature, vol. 411, pp. 1027-1030, 2001.
[59] I. Morohashi, et. al., "16 QAM Synthesis by Angular Superposition of
Polarization using Dual-Polarization QPSK Modulator," in European
Conference on Optical Communications, 2010, p. P3.14.
[60] H. C. H. Mulvad, et. al., "1.28 TBIT/s single-polarisation serial OOK optical
data generation and demultiplexing," Electronics Letters, vol. 45, no. 5,
pp. 280-281, 2009.
[61] E. Myslivets, et. al., "1.56-μs continuously tunable parametric delay line for
a 40-Gbit/s signal," Optics Express, vol. 17, pp. 11958-11964, 2009.

139
[62] S. Namiki, "Wide-Band and -Range Tunable Dispersion Compensation
Through Parametric Wavelength Conversion and Dispersive Optical
Fibers," Journal of Lightwave Technology, vol. 26, pp. 28-35, 2008.
[63] S. Namiki, et. al., "17 ns Tunable Delay for Picosecond Pulses through
Simultaneous and Independent Control of Delay and Dispersion Using
Cascaded Parametric Processes," in European Conference on Optical
Communications, 2008, p. Th.3.C.3.
[64] R. Noé, "Phase noise-tolerant synchronous QPSK/BPSK baseband-type
intradyne receiver concept with feedforward carrier recovery," Journal of
Lightwave Technology, vol. 23, pp. 802-808, 2005.
[65] S. R. Nuccio, et. al., "Experimental Demonstration of All-Optical
Polarization Multiplexing and Polarization Demultiplexing between Two
50-Gbit/s Channels and a Single 100-Gbit/s Channel," in Optical Fiber
Communications Conference, 2010.
[66] S. R. Nuccio, et. al., "Tunable 503 ns Optical Delay of 40 Gbit/s RZ-OOK
and RZ-DPSK Using a Wavelength Scheme for Phase Conjugation to
Reduce Residual Dispersion and Increase Delay," Optics Letters, vol. 34,
pp. 1903-1905, 2009.
[67] S. R. Nuccio, et. al., "Higher-order dispersion compensation to enable a 3.6
μs wavelength-maintaining delay of a 100 Gb/s DQPSK signal," Optics
Letters, vol. 35, pp. 2985-2987, 2010.
[68] S. R. Nuccio, et. al., "Continuously tunable 1.16 μs optical delay of 100
Gbit/s DQPSK and 50 Gbit/s DPSK signals using wavelength conversion
and chromatic dispersion," Optics Letters, vol. 35, pp. 1819-1821, 2010.
[69] S. R. Nuccio, et. al., "Fine tuning of conversion/dispersion based optical
delays with a 1 pm tunable laser using cascaded acousto-optic mixing,"
Optics Letters, vol. 35, pp. 523-525, 2010.
[70] Y. Okawachi, et. al., "Large Tunable Delays Using Parametric Mixing and
Phase Conjugation in Si Nanowaveguides," Optics Express, vol. 16, pp.
10349-10357, 2008.
[71] Y. Okawachi, et. al., "Large Tunable Optical Delays via Self-Phase
Modulation and Dispersion," in Conference on Lasers and Electro-
Optics, May 2007, pp. 1-2.
[72] M. J. O'Mahony, et. al., "The application of optical packet switching in
future communications networks," Communications Magazine, vol. 39,
no. 3, pp. 129-135, March 2001.

140
[73] T. Pfau, et. al., "First real-time data recovery for synchronous QPSK
transmission with standard DFB lasers," Photonic Technology Letters,
vol. 18, pp. 1907–1909, 2006.
[74] G. Raybon, et. al., "100 Gbit/s Challenges and Solutions," in Proceedings of
Optical Fiber Communications Conference, San Diego, 2008, p. paper
OTuG1.
[75] J. Ren, et. al., "12.47ns Continuously-Tunable Two-Pump Parametric
Delay," in European Conference on Optical Communications, 2006, p.
PDP Th4.4.3.
[76] M. Saruwatari, "All-Optical Signal Processing for Terabit/Second Optical
Transmission," Journal On Selected Topics In Quantum Electronics, vol.
6, pp. 1363-1374, 2000.
[77] S. J. Savory, et. al., "Transmission of 42.8 Gbit/s polarization multiplexed
NRZ-QPSK over 6400 km of standard fiber with no optical dispersion
compensation," in Conference on Optical Fiber Communications,
Anaheim, 2007, p. OTuA1.
[78] K. Sekine, et. al., "40 Gbit/s; 16-ary (4 bit/symbol) optical
modulation/demodulation scheme," Electronics Letters, vol. 41, pp. 430–
432, 2005.
[79] M. Serbay, "Implementation of Differential Precoder for High-Speed
Optical DQPSK Transmission," Electronic Letters, vol. 40, no. 20, 2004.
[80] C. E. Shannon, "A mathematical theory of communication," Bell. Syst. Tech.
J, vol. 27, pp. 379-423, 1948.
[81] J. E. Sharping, et. al., "All-optical, wavelength and bandwidth preserving,
pulse delay based on parametric wavelength conversion and dispersion,"
Optics Express, vol. 13, pp. 7872-7877, 2005.
[82] C. A. Siller, SONET/SDH.: IEEE Press, 1996.
[83] B. A. Small, et. al., "A modular, scalable, extensible, and transparent optical
packet buffer," Journal of Lightwave Technology, vol. 25, no. 4, pp. 978-
985, April 2007.
[84] R. H. Stolen, "Issues in Raman gain measurements," Tech. Dig. Symp.
Optical Fiber Measurements; NIST Special Publication 953 , pp. 139–
142, 2000.
[85] R. H. Stolen, et. al., "Raman gain in glass optical waveguides," Applied
Physics Letters, vol. 22, no. 6, 1973.

141
[86] C. Tian, et. al., "Dual-wavelength packets buffering in dual-loop optical
buffer," Photonic Technology Letters, vol. 20, no. 8, pp. 578-580, April
2008.
[87] S. Tsukamoto, et. al., "Coherent demodulation of optical multilevel phase-
shift-keying signals using homodyne detection and digital signal
processing," Photonic Technology Letters , vol. 18, pp. 1131–1133, 2006.
[88] R. S. Tucker, et. al., "Slow-Light Optical Buffers: Capabilities and
Fundamental Limitations," Journal of Lightwave Technology, vol. 23, pp.
4046-4066, 2005.
[89] Y. Ueno, et. al., "3.8-THz Wavelength Conversion of Picosecond Pulses
Using a Semiconductor Delayed-Interference Signal-Wavelength
Converter (DISC)," Photonics Technology Letters, vol. 10, pp. 346-348,
1998.
[90] R. S. Vodhanel, et. al., "Performance of directly modulated DFB lasers in
10-Gbit/s ASK, FSK, and DPSK Systems," Journal of Lightwave
Technology, vol. 8, no. 9, pp. 1379-1386, 1990.
[91] J. Wang, et. al., "Impact of chromatic and polarization-mode dispersion on
DPSK Systems Using Interferometric Demodulation and Direct
Detection," Journal of Lightwave Technology, vol. 22, no. 2, 2004.
[92] Y. Wang, et. al., "44-ns Continuously Tunable Dispersionless Optical Delay
Element Using a PPLN Waveguide With Two-Pump Configuration,
DCF, and a Dispersion Compensator," Photonics Technology Letters, vol.
19, no. 11, pp. 861-863, June 2007.
[93] S. Watanabe, et. al., "Compensation of chromatic dispersion in a single-
mode fiber by optical phase conjugation," Photonic Technology Letters,
vol. 5, no. 1, pp. 92-95, January 1993.
[94] A. E. Willner, "Stable and Reconfigurable Optical Networks," in IEEE
Lasers and Electro-Optics Society Annual Meeting 2007, Piscataway, NJ,
USA, 2007, p. paper WFF1.
[95] P. J. Winzer, "Generation and 1,200-km Transmission of 448-Gb/s ETDM
56-Gbaud PDM 16-QAM using a Single I-Q Modulator," in ECOC,
2010, p. paper PD2.2.
[96] P. J. Winzer, "Impact of Filtering on RZ-DPSK Reception," Photonics
Technology Letters, vol. 15, no. 6, 2003.
[97] P. J. Winzer, et. al., "107-Gbit/s optical ETDM transmitter for 100G
Ethernet transport," in Proceedings of European Conference on Optical
Communication, Glasgow, 2005, pp. 1-2.

142
[98] X. Wu, et. al., "Optical WDM-to-TDM Multiplexing of 40-Gbit/s Data
Channels into a Single-Wavelength 320-Gbit/s Data Channel using
Cross-Phase Modulation in Highly Nonlinear Fiber," in European
Conference on optical Communicaitons, 2009.
[99] X. Wu, et. al., "Tunable optical wavelength conversion of OFDM signal
using a periodically-poled lithium niobate waveguide," Optics Express,
vol. 17, 2009.
[100] A. Yariv, et. al., Optical Waves in Crystals: Propagation and Control of
Laser Radiation.: Wiley-Interscience, 2002.
[101] Y.-K. Yeo, et. al., "A dynamically reconfigurable folded-path time delay
buffer for optical packet switching," Photonics Technology Letters, vol.
16, no. 11, pp. 2559-2561, November 2004.
[102] O. F. Yilmaz, et. al., "Time-Slot-Interchange of 40 Gbit/s Variable Length
Optical Packets Using Conversion/Dispersion-Based Tunable Delays,"
Optics Letters, vol. 33, pp. 1954-1956, 2008.
[103] O. F. Yilmaz, et. al., "Optical Multiplexing of Two 21.5 Gbit/s DPSK
Signals into a Single 43 Gbit/s DQPSK Channel with Simultaneous 7-
Fold Multicasting in a Single PPLN Waveguide," in Optical Fiber
Communications Conference, 2009.
[104] O. F. Yilmaz, et. al., "40 Gbit/s Optical Packet Buffer Using
Conversion/Dispersion Based Delays," Journal of Lightwave Technology,
vol. 28, pp. 616-623, 2010.
[105] S. J. B. Yoo, "Wavelength conversion technologies for WDM network
applications," Journal of Lightwave Technology, vol. 14, pp. 955-966,
June 1996.
[106] J. Yu, et. al., "In-band Wavelength Conversion of 16x114-Gbps Polarization
Multiplexed RZ-8PSK Signals with Digital Coherent Detection," in
Optical Fiber Communications Conference, 2009.

Abstract (if available)

Abstract Increased data traffic demands, along with a continual push to minimize cost per bit, have recently motivated a paradigm shift away from traditional on-off keying (OOK) fiber transmission links towards systems utilizing more advanced modulation formats. In particular, modulation formats that utilize the phase of the optical signal, including differential phase shift keying (DPSK) and differential quadrature phase shift keying (DQPSK) along with polarization multiplexing (Pol-MUX), have recently emerged as the most popular means for transmitting information over longhaul

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Detection and optical signal processing using phase based optical modulation formats

PDF

High-speed and reconfigurable all-optical signal processing for phase and amplitude modulated signals

PDF

All-optical signal processing toward reconfigurable optical networks

PDF

Optical signal processing for enabling high-speed, highly spectrally efficient and high capacity optical systems

PDF

Optical signal processing for high-speed, reconfigurable fiber optic networks

PDF

Nonlinear optical signal processing for high-speed, spectrally efficient fiber optic systems and networks

PDF

Reconfigurable high‐speed optical signal processing and high‐capacity optical transmitter

PDF

Reconfigurable high speed optical signal processing for optical communication and modulation format manipulation

PDF

Advanced modulation, detection, and monitoring techniques for optical communication systems

PDF

Optical wave mixing for tunable delays and high‐speed signal processing

PDF

Optical modulators for integrated photonic systems

PDF

Silicon photonics integrated circuits for analog and digital optical signal processing

PDF

Optical communications, optical ranging, and optical signal processing using tunable and reconfigurable technologies

PDF

Microresonator-based Kerr frequency comb for high-speed optical communications and signal processing

PDF

Reconfigurable optical signal processing for efficient spectrum utilization in high-speed optical communication systems

PDF

Reconfigurable and flexible high-speed optical signal processing and spectrally shared optical subsystems

PDF

Silicon integrated devices for optical system applications

PDF

On-chip Kerr frequency comb generation and its effects on the application of optical communications

PDF

Microdisk-waveguide system for the control of the polarization state of light

PDF

Reconfigurable high-speed processing and noise mitigation of optical data

Asset Metadata

Creator Nuccio, Scott R. (author)

Core Title Applications of all optical signal processing for advanced optical modulation formats

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Electrical Engineering

Publication Date 01/26/2011

Defense Date 12/10/2010

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag fiber optics,non-linear optics,OAI-PMH Harvest,optical delay,optical modulation,optical signal processing,optics

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Willner, Alan E. (committee chair), Armani, Andrea M. (committee member), Steier, William H. (committee member)

Creator Email nuccio@usc.edu,scott.nuccio@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m3625

Unique identifier UC1149124

Identifier etd-Nuccio-4231 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-432059 (legacy record id),usctheses-m3625 (legacy record id)

Legacy Identifier etd-Nuccio-4231.pdf

Dmrecord 432059

Document Type Dissertation

Rights Nuccio, Scott R.

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu