Ultrafast Forwarding Architecture Using a Single Optical Processor for Multiple SAC-Label Recognition Based on FWM

Abstract

We propose and demonstrate a novel ultrafast label processor that can recognize multiple spectral amplitude coded (SAC) labels using four wave mixing (FWM) sideband allocation and selective optical filtering. Our proposed solution favors hardware simplicity over bandwidth efficiency in order to achieve ultra- fast label recognition at reasonable cost. Our implementation, unlike all other optical label processing techniques, does not require time gating, envelope detectors, or serial-to-parallel converters. Labels are transmitted simultaneously with the payload, improving temporal efficiency at the expense of spectral efficiency. Note that bandwidth efficiency can be improved through a frequency management scheme that uses irregular spacing of wavelengths for payload and label, a complexity overhead in management similar to that in long-haul networks employing irregular spacing of carriers to avoid FWM products. We present two experiments of the single processor for ultrafast forwarding using first optoelectronic and then all-optical switches. In the first experiment, we use 10 SAC labels with minimum bin separation of 25 GHz, 10 Gb/s variable-length data packets, and forward packets over 200 km using electrooptical switches. In the second experiment, all-optical switching at 40 Gb/s is demonstrated using a SAC family for up to 36 labels. We present details on the families of spectral codes for label recognition, using unequally spaced frequency bins. A code family with weight 2 and length 9 uniquely identifies 36 labels. Hardware complexity is moderate compared with short-pulse code labels (mode-locked laser) techniques. Two stable tunable lasers are required for label generation of this code family; all other hardware is commercial, off-the-shelf components such as semiconductor optical amplifiers, array waveguide gratings, optoelectronic switches, and photodetectors.

Full text

868 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 14, NO. 3, MAY/JUNE 2008
Ultrafast Forwarding Architecture Using a Single
Optical Processor for Multiple SAC-Label
Recognition Based on FWM
Jos
´
e Bernardo Rosas-Fern
´
andez, Member, IEEE, Simon Ayotte, Student Member, IEEE,
Leslie A. Rusch, Senior Member, IEEE, and Sophie LaRochelle, Member, IEEE
Abstract—We propose and demonstrate a novel ultrafast label
processor that can recognize multiple spectral amplitude coded
(SAC) labels using four wave mixing (FWM) sideband allocation
and selective optical filtering. Our proposed solution favors hard-
ware simplicity over bandwidth efficiency in order to achieve ultra-
fast label recognition at reasonable cost. Our implementation, un-
like all other optical label processing techniques, does not require
time gating, envelope detectors, or serial-to-parallel converters.
Labels are transmitted simultaneously with the payload, improv-
ing temporal efficiency at the expense of spectral efficiency. Note
that bandwidth efficiency can be improved through a frequency
management scheme that uses irregular spacing of wavelengths
for payload and label, a complexity overhead in management sim-
ilar to that in long-haul networks employing irregular spacing
of carriers to avoid FWM products. We present two experiments
of the single processor for ultrafast forwarding using first opto-
electronic and then all-optical switches. In the first experiment,
we use 10 SAC labels with minimum bin separation of 25 GHz,
10 Gb/s variable-length data packets, and forward packets over
200 km using electrooptical switches. In the second experiment,
all-optical switching at 40 Gb/s is demonstrated using a SAC fam-
ily for up to 36 labels. We present details on the families of spec-
tral codes for label recognition, using unequally spaced frequency
bins. A code family with weight 2 and length 9 uniquely identi-
fies 36 labels. Hardware complexity is moderate compared with
short-pulse code labels (mode-locked laser) techniques. Two stable
tunable lasers are required for label generation of this code family;
all other hardware is commercial, off-the-shelf components such
as semiconductor optical amplifiers, array waveguide gratings, op-
toelectronic switches, and photodetectors.
Index Terms—Four wave mixing (FWM), generalized multipro-
tocol label switching networks (GMPLS), label recognition, packet
switching, spectral amplitude codes (SACs).
Manuscript received November 1, 2007; revised January 4, 2008. This work
was supported in part by the Canadian Institute for Photonic Innovations (CIPI).
Dr. J. B. Rosas-Fernandez acknowledge financial support from the U.K. Engi-
neering and Physical Sciences Research Council (EPSRC). This work was pre-
sented in part at the European Conference on Optical Communication (ECOC),
2006 and 2007.
J. B. Rosas-Fern
´
andez was with the Centre d’Optique Photonique et Laser
(COPL), Department of Electrical and Computer Engineering, Universit
´
e Laval,
Quebec, QC G1K 7P4, Canada. He is now with the Group for Photonic Com-
munications, Centre for Advanced Photonics and Electronics, University of
Cambridge, Cambridge CB3 0FA, U.K. (e-mail: jbr28@cam.ac.uk).
S. Ayotte, L. A. Rusch, and S. LaRochelle are with the Centre d’Optique
Photonique et Laser (COPL), Department of Electrical and Computer Engineer-
ing, Universit
´
e Laval, Quebec, QC G1K 7P4, Canada (e-mail: simon.ayotte.l@
ulaval.ca; leslie.rusch@gel.ulaval.ca; sophie.larochelle@gel.ulaval.ca).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/JSTQE.2008.916751
I. INTRODUCTION
D
UE to the fast growth of Internet traffic and other
high-bandwidth data communications applications such
as blade server interconnections, storage area networks,
telemedicine, and tele-education, all-optical packet switched
networks are required where voice, video, and data traffic
will be integrated under the IP [1]. These services will re-
quire high-performance optical core and access networks to
provide high bandwidth to industrial and residential users. A
promising approach is optical label switching employing the
following subsystems: label generation, label recognition, label
swapping (erasing and inserting a new label), synchronization,
and contention resolution (e.g., buffering, deflection). Several
challenges exist in the development of efficient network ar-
chitectures and devices necessary for their implementation in
commercial systems. Some of these challenges are fast label
recognition, fast and inexpensive electrooptical (E/O) switches,
and all optical buffering.
To achieve high throughput at the optical router, ultrafast la-
bel recognition on the order of picoseconds is needed, as well
as minimal buffering. One of the most promising advances in
packet switching systems in recent years has been the devel-
opment of generalized multiprotocol label switching networks
(GMPLS) for high-speed forwarding and routing [2]. However,
GMPLS implementation for optical networks is hampered by
electronic processing times, still in the range of several nanosec-
onds or milliseconds per packet. Different optical label strate-
gies with conventional electronic recognition modules have been
investigated: time-domain [4], [5], bit-parallel [6], amplitude-
shift keying [7], frequency-shift keying [8], subcarrier modula-
tion [9], and bit stacking [10]. These E/O labels are limited to
label modulation rates in the range of a few gigabits per second.
In 1987, the information from an optical code (OC) was
first demonstrated as a control signal for a photonic switch,
establishing that OCs can be employed as headers or labels for
switching [11]. This approach is sometimes known as photonic-
code-based routing in the literature [12], where an OC label
is assigned to each packet. Each label could contain forward-
ing QoS and other information. The photonic code labels are
designed to provide a high autocorrelation peak for ultrafast op-
tical detection. While technology choices abound for OC labels,
all share a common property when it comes to label recognition;
typically, recognition time is only limited by the propagation
time of light in the correlation device.
1077-260X/$25.00 © 2008 IEEE
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.

ROSAS-FERN
´
ANDEZ et al.: ULTRAFAST FORWARDING ARCHITECTURE USING A SINGLE OPTICAL PROCESSOR 869
As with OCDMA data transmissions, photonic code labels
can be implemented in several ways: time domain (includ-
ing spread spectrum and optical orthogonal codes), wavelength
domain [including spread-time and spectral amplitude coding
(SAC)], or wavelength/time domain (varieties include 2-D, λ–t,
and fast frequency hopping). An efficient exploitation of pho-
tonic code labels requires the development of encoders/decoders
specifically adapted to switching. To date, most label switch-
ing proposals use encoders/decoders originally designed for
OCDMA data transmission systems [16]–[24], leading to high-
component count and splitting loss that limit the network scala-
bility. Consider [13] using wavelength-domain OC labels based
on binary phase-shift keying (BPSK) codes, and [14] and [15]
using wavelength/time labels; in all three cases, parallel correla-
tion techniques achieved ultrafast recognition. To provide inputs
to the bank of parallel correlators, the photonic label was split or
copied M times, where M is the number of possible labels. For
a given packet, one correlator achieves an autocorrelation peak,
all other correlators exhibit only a low-power cross-correlation
signal. Since no more than one packet is present at detection, no
multiuser interference is present (in contrast to OCDMA data
transmission systems).
For packet switching, the switch must identify which of all
possible codes is present, while in OCDMA data transmission
systems, the code is known and only the code carrying the de-
sired data is processed. To overcome high component count and
splitting loss, we choose to find a single processor that can de-
tect any of the possible codes without resorting to parallelism.
Several other strategies choose instead to reduce the number
of labels to be detected by: 1) examining only the statistically
most probable labels [25] or 2) exploiting label stacking [26].
In [25], the 100 most popular labels (generating 90% of all
Internet traffic) are forwarded optically, resulting in a reduced
and more manageable number of parallel code correlators per
node. However, the interaction between the electronic and op-
tical caches may result in the undesirable loss of packet order.
The use of hierarchical addressing also reduces the number of
code correlators per node; a stack of labels contains a label for
each region in the hierarchy. In each region a smaller number
of parallel code correlators is required per node. We previously
proposed optical label stacking using wavelength domain, SAC
labels. We demonstrated an “out-of-band” solution [27] with
SAC labels on wavelengths distinct from that of the payload,
and a “self-routing, intraband” solution where the payload it-
self was encoded by the stack of labels [28]. Both systems
perform ultrafast label recognition using parallelism: one fiber
Bragg Grating (FBG) encoder/decoder per label. In this paper,
we again propose the use of spectral amplitude codes; however,
we avoid parallel correlators for label recognition.
Single processor, nonparallel, solutions have been proposed
in the literature for various code types. A silica planar lightwave
circuit (PLC) device was proposed to process phase modulated,
time domain, and short pulses [29]. In this case, parallelism
was achieved on the board, reducing component count, but still
incurring splitting losses. The BPSK time-domain codes are
Walsh codes so that splitting losses are linear in the number
of codes. Also, this processor typically generates significant
intensity noise when detecting the combined pulses; a novel
technique is introduced to reduce this noise, at the expense
of efficiency, i.e., the same number of pulses must be more
widely distributed in time. Scaling the processor beyond the
demonstrated length of four codes generating four labels would
be technologically very challenging.
Another encoder/decoder was proposed by Cincotti [30], [31]
that can generate and/or recognize all code labels via a single
processor. This structure is based in an arrayed waveguide grat-
ing (AWG) router and was experimentally demonstrated [32] for
the generation and processing of 16 phase-shift keying codes,
each consisting of 16 chips at 200 Gchip/s. The method was
extended for wavelength/space coding [33]. This method re-
quires picosecond pulses from a mode-locked laser and phase
modulation, but avoids the pitfalls of parallelism: high com-
ponent cost and/or high splitting losses. The wavelength/space
codes provide impressive code cardinality, but require higher
cost and complexity: multiple mode-locked lasers with center
wavelengths managed to avoid beating effects.
In this paper, we focus on SAC labels due to the high perfor-
mance and low cost of label hardware. The OC-label processor
recognizes multiple SAC labels using four wave mixing (FWM)
and selective filtering.
This paper is organized as follows. In Section II, we explain
the concept of SAC label recognition using FWM and selective
optical filtering. An experimental demonstration of switching
of 10 SAC labels with payloads at 10 Gb/s with E/O switches
is presented in Section III. The design and implementation of a
computer algorithm to search for larger codes families is pre-
sented in Section IV, as well as an experimental validation of
these codes in an optically controlled switch carrying 40 Gb/s
payloads. Finally, in Section V, we conclude the paper.
II. U
LTRAFAST LABEL RECOGNITION USING A SINGLE
PROCESSOR FOR MULTIPLE SAC-LABELS
The proposed all-optical label processor for ultrafast for-
warding exploits low-complexity spectral amplitude code la-
bels. Fig. 1 shows the schematic of the SAC-label processor.
The input packet consists of a payload with modulated data on
one wavelength, and adjacent wavebands (bins) carrying the un-
modulated SAC labels (equivalent to be constant at the packet
rate). A nonlinear optical medium is employed to generate FWM
products from the label. Proper choice of the spacing of spectral
components (also known as bin wavelengths or chips) assures
that a unique FWM product will exist for each label. A bank
of optical filters or AWG is used to isolate these unique prod-
ucts and act as the control signal demultiplexer. The schematic
shows that the splitting of the incoming OC labels is completely
avoided and high-speed recognition is achieved due to the fast
nonlinear process.
If the set of SAC labels is carefully selected, each label gener-
ates at least one unique FWM sideband. We identify the presence
of the label by filtering its unique sideband. The careful choice
of frequency sidebands is not unlike WDM systems that seek
to reduce FWM effects by using unequal wavelength spacing to
avoid FWM products falling on an occupied wavelength [34].
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.

870 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 14, NO. 3, MAY/JUNE 2008
Fig. 1. Schematic of the SAC-label processor for fast forwarding.
Our scheme is more challenging in code design since FWM
products fall not only in occupied wavelengths, but also in at
least one waveband not occupied by any other FWM product
from any other label at the label processor.
Let L be the number of bins available for labels, which is also
the length of the code. The spectral amplitude coded label is
formed by selecting a subset of the available bins. We assume
binary codes, that is, all wavelengths have the same amplitude,
logical “1,” or a suitably low amplitude for logical “0” (normally
equals zero). Let W be the code weight, i.e., the number of
frequencies present in a given code. We assume that all codes
have the same weight. The code can be represented as a vector of
length L, with W nonzero elements, whose ith element is equal
to the wavelength of the ith bin if that bin is present in the code.
For example, one SAC code with W = 2is
SAC = {0, λ
2
, 0,..., 0,..., λ
L−1
}. (1)
For our application, a label with sufficient optical power in
each of its frequency components generates FWM products
when passed through a nonlinear medium. For any two input
frequencies λ
1
and λ
2
, FWM produces sidebands at 2λ
1
− λ
2
and 2λ
2
− λ
1
[35]. In Fig. 2, we consider two labels from a fam-
ily of weight-2 codes of length 3. Each spectral amplitude code
will produce two additional sidebands after the FWM process.
Codes are designed to assure that the FWM-sidebands (sb) are
nonoverlapping in at least one sideband. In this example, we use
a minimum bin separation of 100 GHz (one unit of separation)
simply as an example. The first code uses one unit of separation
and the second uses two units. The FWM-sb will also appear
in multiples of these separation units. The first label generates
products sb
1
and sb
2
, and the second produces sb
3
and sb
4
.
We note that sb
2
and code wavelength λ
3
coincide; hence, sb
2
cannot be used as an identifier. In this example, code 1 can be
uniquely identified (label identifier, LI
1
)bysb
1
, i.e., LI
1
=sb
1
.
Analogously, sb
3
and sb
4
do not overlap with any active bin of
SAC label 1, nor one of its FWM-SBs, so either can be used as
LI
2
. Clearly, as we increase the number of labels, finding unique
identifiers becomes more challenging. Care must be taken in se-
lecting label power levels. The efficiency of the FWM mixing
will impact the accuracy of label detection, and sufficient label
power levels must be maintained to assure good efficiency.
For OCDMA systems using correlation techniques, autocor-
relation peaks should be equal to distinguish the desired user
among interferers; our application has no interferers, so equal-
power FWM products are not required. The range of the identi-
Fig. 2. SAC-label recognition by unique sideband wavelength allocation.
fier FWM products power (analogous to autocorrelation peaks)
need only be sufficient to: 1) drive the switch and 2) keep false
alarm and miss probability low, i.e., the lowest power FWM
product must be well above the noise. We will demonstrate that
this is indeed achievable. Second-order FWM products must be
kept low, especially if tight spacing of the wavelengths is envis-
aged. In the present paper, we use power control at the transmit-
ter to assure that only first-order FWM products are generated.
If power control was not feasible, optical amplitude limiters or
equalizers could be used to avoid second-order effects.
Normally, it would be desirable to use closely packed frequen-
cies for smaller spectral occupancy and to simplify the design of
the bank of filters (or the AWG) generating the control signals
for the crossbar switch. As the number of labels increases, the
wavelengths will necessary be more widely spaced in order to
generate unique identifiers, leading to a tradeoff between spec-
tral occupancy and the number of labels supported. Nonetheless,
in some cases, the spectral occupancy can be comparable to that
of OC labels using short pulse as chips. See Section I for an
example.
The recuperation of interbin wavelengths for other uses is also
an option for decreasing spectral occupancy, although at the cost
of increased complexity. The label processor would be required
to reject the interbin wavelengths before entering the nonlinear
medium. For instance, the input signal could pass through an
AWG or an FBG in transmission, so valid label wavelengths
would be forwarded to input ports of a second AWG whose
single output would then enter the nonlinear medium.
Label stacking is an important method of reducing the com-
plexity of network nodes. The proposed label processor would
not be able to create wavelength-domain stacked codes. Such
stacking would lead to more than W wavelengths entering the
nonlinear medium, and label identifiers would no longer neces-
sarily be unique. Stacking in the time domain remains a valid
stacking approach, especially given that the labels are not modu-
lated, but constant during the packet interval. Stacking labels in
time, i.e., using time-domain multiplexing for the labels would
lead to moderate increase in device complexity.
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.

ROSAS-FERN
´
ANDEZ et al.: ULTRAFAST FORWARDING ARCHITECTURE USING A SINGLE OPTICAL PROCESSOR 871
In summary, our label recognition technique consists in using
the FWM process to produce two (or more) additional frequen-
cies for each SAC. These spectral codes are chosen such that,
for any given code, it produces a unique FWM frequency com-
ponent in the SACs family that is different from other frequency
components produced by any other code in the set, and also, dif-
ferent from the incident frequency bins. It is, therefore, possible
to determine the incoming code by extracting specific FWM
components using optical filters. Advantages of the proposed
method are:
1) ultrafast recognition due to the fast FWM process;
2) no ultrashort pulses are required;
3) no copying of incoming labels, i.e., no splitting losses;
4) label separation by simple optical filtering (avoiding time
gating used with time domain labels [36]);
5) variable rate packets are easily implemented, as the spec-
tral label is modulated at the packet rate [37];
6) no-serial-to-parallel conversion for label processing;
7) the single, all-optical label processor can be implemented
with off-the-shelf optical components;
8) compatibility with label stacking in time domain.
Drawbacks of the proposed method are:
1) spectral occupancy tradeoff against label cardinality;
2) requires sufficient label power for efficient FWM;
3) second-order FWM to be avoided in codes with close bins;
4) polarization among bins must be controlled to optimize
the FWM efficiency;
5) spectral code generation requires stable tunable lasers or
array (multiwavelength) lasers.
III. E
XPERIMENT USING A SINGLE PROCESSOR FOR MULTIPLE
SAC LABELS AND E/O SWITCHES
As a first demonstration of the proposed architecture, we
chose an SAC family with ten codes suitable to be recognized
by FWM and selective filtering. Table I shows the code set: the
ITU, 25-GHz frequency grid was used, i.e., the unit frequency
separation was 25 GHz. The codes have two active wavelengths
(code weight W = 2) in any of the five wavelength bins (code
length L = 5) and have a cardinality M (number of codes in
the code family) equal to 10. Also indicated in Table I are the
two first-order FWM-SBs for each label. The column labeled
“identifier” indicates which of the two first-order FWM-SBs is
unique to the code and used for label recognition. Consider the
spectral occupancy of this example compared to a system using
one wavelength per label (MPλS). For this example, our system
has a smaller spectral occupancy and only requires two laser
sources instead of ten for an MPλS.
Fig. 3 shows the experimental setup and results using all-
optical recognition of the label. In order to decrease the spectral
occupancy, the laser payload (LP) is located at “intraband” of
the code (at 1554.94 nm) and is filtered out before the label
processor. The total optical bandwidth is 200 GHz and is less
than that used in [29] and [32], where short pulses of 2 ps are
used for 16 OC labels. Pattern generator 1 (PG
1
) generating a
length 2
10
− 1 pseudorandom binary sequence (PRBS) drives
modulator 1 (Mod
1
) to create the optical payload. Four tunable
TABLE I
W
AVELENGTH ALLOCATION FOR THE SAC-LABELS,FWMSIDEBANDS,
AND LABEL IDENTIFIERS
lasers (TL
A
to TL
D
) generate the labels, two per code label.
For simplicity, two packets are generated by splitting a payload
and combining each branch with a distinct label. Modulators
2 and 3 transform the optical signals into two packets of dif-
ferent lengths (0.5 and 1 µs with 0.25 µs guard time between
them). The packets are transmitted through a total of 200 km
of a single-mode fiber (SMF) and the corresponding dispersion
compensating fiber (DCF). At the forwarding node, the signal
is split such that the entire packet is forwarded to the optical
switch; only the OC label is forwarded to the label processor as
the payload is filtered out by a FBG.
The label is then passed through a highly nonlinear semi-
conductor optical amplifier [nonlinear-semiconductor optical
amplifier (NL-SOA), from the Centre for Integrative Physi-
ology (CIP), U.K.]. The characterization of FWM efficiency
of the NL-SOA, for the left (L) and right (R) SBs, is shown
in Fig. 4(a). We used two input continuous-wave (CW) sig-
nals and frequency separations of 12, 25, 50, 100, 150, and
250 GHz. The SOA has 30 dB gain and 6 dBm saturated output
power. The SOA generates at least one unique FWM-SB for
each SAC-label. All measured FWM spectrums are shown for
all the codes in Fig. 4(b). We aligned the polarizations between
the lasers of the label at the transmitter. We achieved good FWM
even after 200 km of transmission, indicating that the polariza-
tion detuning was not excessive. The SOA output is connected
to a 32-channel AWG (25 GHz channel separation) to select the
sideband of the desired code according to the identifier column
of Table I. Since we generate an optical control signal, this could
drive an all-optical switch or it could be photodetected and con-
trol an E/O switch. In the latter case, we used a 1 × 2 LiNBO
3
E/O switch; one label at a time enters the label processor for
optical forwarding.
The EOSPACE optical switch has a switching time of less
100 ps, insertion loss of 3 dB (pass through port) and a crosstalk
level (rejection port) of 23 dB. We used a balanced photodi-
ode (BPD) to overcome variations in the amplified spontaneous
emission noise due to gain saturation in the SOA. A slow elec-
tronic thresholder (ETh) is used to drive +13 V need for the
E/O switch. This also has a desirable effect on the control sig-
nal: very flat square pulses are generated after the Eth. All
the SAC labels have the same optical power at transmitter, but
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.

872 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 14, NO. 3, MAY/JUNE 2008
Fig. 3. Experimental setup and results of packet-switched node using a single processor for multiple SAC labels. Switching is performed after electronic detection
and thresholding of the optical control signals; the electrical control signals bias 1 × 2 LiNBO
3
switches. Insets (a)–(j) show optical signals at AWG outputs
1–10, respectively, corresponding to OC labels 1–10; there are 60 mV and 200 ns per division. A packet train is sent that alternates a longer packet with label 4 and
a shorter packet with label 6, with a brief silent period between packets; the brief periods of oscillation that are visible occur during silent periods between packets
due to cross-gain modulation of ASE in the SOA. Outputs 4 and 6 [insets (d) and (f)] show autocorrelations, while all other outputs display cross-correlations.
Inset (f) has a 1000 ns time offset compared with other insets to center the label of the shorter packet in the inset. The electrical threshold level of 70 mv (the same
for all ports) yields electrical control signals (after detection and thresholding) given in insets (k)–(t) for ports 1–10, respectively; these are 50 mV and 500 ns per
division. Autocorrelation shapes are square; all cross-correlation signals are shifted to the zero level after thresholding. Inset (u) shows three packets at the input of
the switches with labels 6, 4, and 6, respectively; (v) and (w) show the packets switched by control signals from ports 4 and 6 (i.e., OC labels 4 and 6), respectively;
there are 100 mV and 500 ns per division. Note inset (v) has a time offset compared to (u) and (w); in (v), the time shift allows two longer packets to be displayed.
powers of the FWM-SBs differ among the ten codes. The thresh-
old voltage ETh was carefully adjusted to ensure that all ten
codes were successfully recognized without false detection.
With the now fixed Eth, we verified for each label in turn that the
remaining nine AWG ports (valid identifier FWM products other
than that of the label transmitted) did not generate false alarms
in the EThs. Fig. 3 also shows the characteristics of the auto-
and cross-correlated optical signals at each AWG output and the
electrical signals after thresholding. Power from second-order
FWM does not interfere with any label recognition. In this setup,
when the OC label is present, the desired packet is switched to
port 1; if the OC label is not present, the packet is sent to port 2.
Since, the spectral code labels are modulated at the packet rate,
the switch is open during a time equivalent to the corresponding
packet length, thus making the scheme practical for processing
variable length packets. In Fig. 3, we see that the short 0.5 µs
packet is switched to port D, and the 1 µs packet to port B
after the 1 × 2 switches. We verified packet switching for all
the labels and we present BER results of packets for two of the
labels. All labels produce ten different control signals with the
same electrical characteristics following the ETh (both autocor-
relations are square shapes at the respective packet rate and all
the cross-correlations are transformed to a zero level), so the
bit error rates (BERs) of the switched packets are independent
of the label number. Fig. 5 shows the BER measured for the
payloads of both packets using representative labels 4 and 6;
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.

ROSAS-FERN
´
ANDEZ et al.: ULTRAFAST FORWARDING ARCHITECTURE USING A SINGLE OPTICAL PROCESSOR 873
Fig. 4. (a) FWM efficiency for the NL SOA. (b) FWM products for the 10
SAC-labels, showing the unique FWM sideband allocation for each code.
label 4 has greater frequency difference between bins than label
6. We employed a gating signal of 0.2 µs. Error-free transmis-
sion is achieved in all cases (BER < 2 × 10
−−11
) for more than
5 × 10
10
bits, and were transmitted over 200 km of fiber with
1 dB penalty. Hence, we have demonstrated the feasibility of
SAC labels for fast forwarding of optical packets.
IV. D
ESIGN OF SAC FAMILIES FOR FWM LABEL
RECOGNITION
As briefly discussed in the previous sections, SAC families
with unequal frequency spacing are required to guarantee that
unique FWM-SB identifiers can be found for each code. Spectral
occupancy optimization of SAC therefore consists of maximiz-
ing code cardinality for a given spectral band, or equivalently,
finding the minimum spectral band required for a given cardi-
nality. Such an optimization problem was previously studied in
the context of WDM transmission systems for which unequal
channel spacing was proposed and studied as a mean to mitigate
FWM-induced crosstalk [39]–[50]. For example, transmission
experiments using unequal frequency spacing channels at rates
of several gegabits per second and over thousands of kilometers
are presented in [39].
Fig. 5. BER versus average received power for the payloads of the switched
packets using labels 4 and 6 at ports D and B, respectively.
Many of the various studies performed to reduce FWM
crosstalk through unequal channel spacing are extensions of
techniques developed to reduce third-order intermodulation dis-
tortion in radio systems. An algorithm was developed in [34]
and implemented to search for the optimal spectral position-
ing of 10 transmitting channels. The modeling led to an integer
linear programming problem of a nondeterministic polynomial
type that had to be solved through computer search. Channel
allocation strategies were also developed by taking advantage
of optimization techniques established for the determination of
optical orthogonal codes (OOCs) [40]. Analytical tools are de-
veloped in [41]–[45] using frequency difference triangle (DFT),
frequency difference sets (FDS), and optimal Goulomb rulers
to find the number and position of FWM products.
Optimization constraints in the design of WDM channel plans
differ from the requirements of SAC labels compatible with
FWM recognition. For example, suboptimal channel allocation,
where some FMW-SBs overlap with some carriers, is studied
in [46] and [47]. In the present case, these sidebands could
not be used for SAC identification and this strategy cannot be
adopted. Additionally, depending on the launch power, higher
order FWM products can be a concern to WDM systems [48],
but this is usually not an issue in the present case. Finally, in
the case of SAC identification, the overlap of FWM-SBs should
be avoided for at least one of the products while such overlaps
between FWM-SBs only are not detrimental to WDM systems.
Given the aforementioned considerations, we use a simple
algorithm to search for optimum SAC families that are com-
patible with label identification through FWM. The algorithm
is described in Fig. 6. We begin by defining S to be a vector
containing the frequency slot allocation
S = {S
0
,S
1
,S
2
,S
3
,..., S
L−1
} (2)
where L was already defined in Section II as the maximum
number of frequency bins for an SAC label, i.e., the code length.
The 2
L
labels are composed from the presence/absence of these
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.

874 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 14, NO. 3, MAY/JUNE 2008
Fig. 6. Algorithm for searching spectral amplitude code families for label
recognition based on FWM.
slots. We also defined a minimum (∆S
min
) and a maximum
(∆S
max
) frequency slot spacing to limit the solution set and
the total bandwidth. Any combinations where two or more S
i
elements are equal are automatically eliminated since that would
mean that the vector S has two bins at the same frequency.
The program considers each of the vectors and calculates the
FWM-SB for each binary set B = {b
0
,b
1
,b
2
...,b
L−1
} where
b
i
=1when that frequency is part of the code and 1 when it
is not. The simulation stops when a solution that satisfies the
specified code length and weight is found while minimizing
the maximum frequency spacing and required bandwidth. For
example, with L =4and W =2, and ∆S
max
= 2, the program
stops once it finds that the vector S = {0234} produces
FWM-SB suitable to identify all the possible labels with the
smallest total spectral bandwidth. If no solution is found with
the given parameters, the program stops and ∆S
max
must be
increased to resume searching.
For a given code length and code weight, a code family is
defined by specifying the S vector together with the binary code
set. For example, in Table I, the M =10 codes consists of
the binary combinations OC-label #1 (11000)toOC-label
#10 (0 0 0 1 1). The code family is independent of the phys-
ical value of the frequency (wavelength) spacing ∆f between
two contiguous slots. When a solution is found, the frequency
(wavelength) allocation can be determined from the desired
TABLE II
W
AVELENGTH ALLOCATION FOR THE SAC-LABELS WITH DIFFERENT CODE
LENGTHS AND WEIGHTS
∆f. For example, in Table I, ∆f is 25 GHz (0.2 nm), and the S
vector is S
table−1
= {0, 1, 3, 6, 8}, where 0 is the reference
wavelength. The computer algorithm seeks a label family with
minimum and unequal bin separations that generate unique SB
identifiers, assuming an ideal communications channel. Future
work on identifying label families could include optical proper-
ties such as power conversion efficiency, power equalization, or
optical SNR.
V. S A C R
ECOGNITION AND ALL-OPTICAL SWITCHING
OF
40 GB/S PACKETS WITH LONGER CODE SETS
The numerically generated SAC labels are shown in Table II
for lengths L = 5–9 and all using weight 2. It was found that, for
an SAC family with L = 9 and W = 2, up to M = 36 (C
9
2
)SAC
labels can be produced. Comparing with MPλS networks using
one wavelength for one label, this is four times the reduction in
spectral occupancy. Table II also shows that, if a bin bandwidth
is as wide as a payload bandwidth, the label would span as
much bandwidth as an extremely dense MPλS network. Clearly
unused bandwidth must be exploited in order to capitalize on
the bandwidth advantage. Contrary to WDM systems limited
by FWM, in which the unoccupied frequency slots cannot be
used for other carriers because of possible nonlinear crosstalk,
in this SAC-label strategy, any unused frequency slot can be
used for other WDM users without interference (the FWM-SB
are only generated at the optical correlator). The unused label
frequency slots are available for exploitation. For example, for
the demonstrated setup of Fig. 2, L =5and there is one iso-
lated empty slot, two sets of two consecutive empty slots, and
one set of three consecutive slots, with 25 GHz per slot. Sev-
eral 10 G channels could be located in the unused bandwidth.
We implemented a SAC-label processor for a W = 2 and L = 9
(36 labels) code with ∆f = 0.1 nm (12.5 GHz). For these labels,
there are two sets of two consecutive empty slots, and five sets
of three consecutive slots, with 12.5 GHz per slot. The unused
frequency slots can be used for low-speed (10 Gb/s and lower)
data transmission, or for other labels. With 11 bins, two coded
labels could be transmitted (e.g., codes 0, 1, 3, 6, 8, and codes
2, 4, 7, 9, 10). That is 10 labels × 2 = 20 labels in 11 bins.
Close packing of frequency slots will also compact the band-
width available for other purposes falling between frequency
slots. This compression makes the exploitation of unused band-
width more problematic by increasing the complexity and cost
of filtering technologies required.
For our all-optical switching experiment, the code set and LIs
are summarized in Table III and Fig. 7 shows the experimental
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.

ROSAS-FERN
´
ANDEZ et al.: ULTRAFAST FORWARDING ARCHITECTURE USING A SINGLE OPTICAL PROCESSOR 875
TABLE III
W
AVELENGTH ALLOCATION FOR THE SAC-LABELS WITH
l
= 9 AND
w
= 2
setup used for all-optical processing and switching of labels and
packets. The correlator optical output directly controls an all-
optical switch (thus avoiding electronic detection) that consists
of integrated SOAs in Mach–Zender (MZI) configuration. The
correlator output spectrums for all codes are shown in Fig. 7(a)
where the FWM-SBs corresponding to each LIs are marked.
All the codes have the same input power. For code 1, with
the minimum bin separation of 0.1 nm, second-order FWM is
produced but it does not interfere with any other LI. FWM
power conversion efficiency varies among codes. For all the
labels, the efficiencies are between −11 and −34 dB from the
peak powers of the input bins with minimum (0.1 nm) and
maximum (3.5 nm) bin separations corresponding to codes 1
and 8, respectively. Two tunable lasers (TL
1
and TL
2
) generate
the spectral code label. We use SAC labels 1 and 8 since FWM
power conversion for the 36 codes will fall between these two
values. Both labels are transmitted with the same optical power.
A waveform generator modulates the label to produce a square
4 µs window with 1 µs guard time. The payload is modulated
with a 40 Gb/s PRBS signal (2
31
− 1). At the switching node, the
unique FWM-SB is produced by passing the SAC-label through
a nonlinear SOA with a 15 dB gain and saturation power of
10 dBm. The payload is filtered out by a notch filter.
The unique FWM-SB corresponding to the LIs are isolated
by an AWG followed by a narrowband optical filter of less than
8 GHz bandwidth to obtain the optical control signals (OCS)
for the all-optical switch. The OCS is sent in counterpropaga-
tion direction through the SOA-MZI switch (CIP, U.K.) produc-
ing a phase shift in the interferometer, thus switching a 4 µs,
40 Gb/s payload signal to output port 2. The SOA-MZI has a 5
dB insertion loss and a switching energy of <100 fJ. Fig. 7(b)
also shows the OCS spectrum of labels 1 and 8 with peak powers
Fig. 7. (a) Label recognition for 36 SAC labels using FWM sidebands.
(b) Setup for 40 Gb/s all-optical switching using the 36-port single corela-
tor for code label recognition. Extinction ratios of 7 and 9.8 dB were obtained
for switched 40 Gb/s packets for labels with the maximum and minimum label
separation, respectively.
of 3.2 and −3 dBm, respectively. We obtained the suppression of
adjacent LI by more than 20 dB for both SAC labels after demul-
tiplexing. Due to the low FWM power conversion efficiency, the
filtered FWM-SBs need to be amplified. The addition of ASE
noise to the OCS does not impair all-optical switching of the
40 Gb/s payload by the OCS. Switching extinction ratios of 7
and 9.8 dB were achieved.
For each of the labels recognized, we measured cross correla-
tion on all other LI AWG ports; the absence of any FWM power
leads to a very low-level, noise-only optical signal. We verified
that this power was insufficient to trigger the all-optical switch,
i.e., there were no false detections. If another type of optical
switch is used requiring equal optical input power, an SOA with
low saturation power could be used to equalize the input optical
power for all labels.
We chose to modulate the label at the packet rate to demon-
strate that the label could directly control an optical switch. This
choice has the advantage of natively adapting the control signal
to the packet duration, i.e., in support of variable packet length.
Furthermore, in this case, the optical signal at the correlator
output is compatible with optically controlled switches, i.e., no
further signal processing is required to open and close the optical
gates. The modulation rate of the spectral label is flexible and
could be chosen in a manner other than that in our experiment.
The label optical bandwidth can be freed for the packets time slot
if, for example, the spectral code precedes the packet. However,
this would require electronic or optical envelope detection for
controlling the switch, higher speed detectors, an accurate syn-
chronization mechanism, and time gating techniques for label
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.

876 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 14, NO. 3, MAY/JUNE 2008
separation. For W = 2 forwarding in 32- or 64-port switches is
achievable. It is reported [46] that the number of FWM products
for W wavelengths is FWM
SB
= W
2
× (W − 1). The exact
maximum number of labels for W > 2 and for different FWM
efficiencies should be investigated; however, it is reasonable to
expect that our method will be limited to the number of labels
close to those demonstrated.
On the other hand, increasing the dimensionality of the code
(either via time or space coding) could increase the label car-
dinality at the cost of some increased hardware complexity.
Also, as with any optical signal processing method using FWM,
polarization control is a great challenge for practical implemen-
tation. Complex methods such as tracking the polarization could
be considered. As the label rates are low, a simple polarization
scrambler at the transmitter might be sufficient to ensure that
a constant FWM average power is produced by the SOA. This
issue also deserves further investigation and is beyond the scope
of this paper. In the case of AWG technology, AWGs with un-
equal spaced channels have been demonstrated that can be used
for this application [49]. Thus, in order to minimize the num-
ber of components for system applications, all the labels could
be detected with only one optical amplifier located before a
high-efficiency nonlinear element together with one AWG with
unequally spaced narrowband channels.
Two stable tunable lasers are required for label generation
of a weight 2 code family; modulation rates are extremely low
(packet rate, not bit rate) facilitating stability, but tuning must be
accomplished within the guard time between packets. Increased
tunability speed in low-cost lasers would greatly facilitate the
adoption of our proposed architecture; promising technologies
can be found in [50]. State-of-the-art tunable lasers (for exam-
ple, from Bookham and Agility) provide switching speeds in
microseconds and more than 45 nm of tuning. Alternatively, a
bar of nine fixed lasers could be used to generate the labels,
with only two lasers being active for any given label. For both
experiments, due to the low label rate, chromatic dispersion ef-
fects can be neglected even for the longest bin separations. In
any case, the transmission of high-speed packets will require
dispersion compensation for the data packets.
VI. C
ONCLUSION
We presented a novel approach for all-optical label process-
ing and packet forwarding using a family of spectral amplitude
coded labels that can be recognized by a single-optical-code
correlator. The code correlator performs nonlinear frequency
mixing and selective optical filtering to demultiplex a control
signal. Since the labels are in the frequency domain, they can
be modulated at any packet rate and the generated optical con-
trol signal that gates the all-optical switch will natively match
the packet length. This simple technique, therefore, does not re-
quire ultrashort pulses, high-speed electronics or any envelope
detector. Our method does not insert splitting losses. In contrast,
parallel processing introduces a 10 × log 10(M) dB loss per M
labels; for 10 and 36 labels, parallel processing would incur a
total loss of 10 and 15.6 dB, respectively. In our case, suffi-
cient power for efficient FWM is needed; we suppose the use of
one amplification stage to overcome system losses and assure
enough power to the processor. Also, when using a SOA as the
nonlinear FWM device, it is possible to integrate two SOAs, one
optimized for amplification and one for nonlinear behavior, on
the same device. Most importantly, a single processor method
remains very attractive as the cost and complexity of a single
processor is much lower than parallel processors using one cor-
relator per label. The SAC-label identification was demonstrated
with standard off-the-shelf components. More novel technolo-
gies or dedicated devices could eventually be employed. For
example, the demonstration was done with a nonlinear SOA but
other nonlinear media such as photonic crystal and chalcogenide
fibers could be investigated. The code generation required two
tunable CW lasers that could also be step-tunable or discretely
addressable lasers.
We performed two packet switching experiments with 10 and
40 Gb/s payloads using E/O switching and all-optical switch-
ing, respectively. In the first experiment, we used an EThs after
label identification to control the switch and we demonstrated
the switching of 10 Gb/s packets with only 1 dB penalty after
200 km transmission and switching. Label recognition without
any errors was accomplished after a careful selection of the
threshold voltage. In the second experiment, the optical con-
trol signal was successfully used to control a SOA-MZI optical
switch once again without any packet loss. In this paper, we
demonstrated code generation and identification of up to 36
labels for a code length of 9.
The proposed SAC labels therefore present a spectral occu-
pancy improvement of more than four times over wavelength
usage for MλPS systems. OC with ultrashort pulses can have
comparable or larger spectral occupancy but at higher cost and
with increased complexity. Finally, we found that code optimiza-
tion must take into account the constraints of the label recogni-
tion technique. We estimate that the design of SAC families with
larger weight and length can be improved with more sophisti-
cated mathematical and numerical tools. We believe that this
all-optical packet processing approach can find application in
high-speed blade servers’ interconnections, where high amount
of data must be rapidly forwarded between servers.
A
CKNOWLEDGMENT
The authors acknowledge the Canadian Institute for Photonics
Innovation (CIPI) for funding the project on packet switched
networks with photonic-code-based processing. Also, they are
thankful to W. Mathlouthi and P. Larochelle of the Universit
´
e
Laval, M. Presi of Scuola Superiore Stant’Anna in Pisa, and
A. Wonfor and Prof. I. H. White of Cambridge University for
fruitful discussions and technical help. The all-optical switching
experiment was performed in the University of Cambridge.
R
EFERENCES
[1] M. J. O’Mahony, D. Simeonidou, D. K. Hunter, and A. Tzanakaki, “The
application of optical packet switching in future communication net-
works,” IEEE Commun. Mag., vol. 19, no. 3, pp. 128–135, Mar. 2001.
[2] U. Black, MPLS and Label Switching Networks, 2nd ed. Englewood
Cliffs, NJ: Prentice-Hall, 2002.
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.

ROSAS-FERN
´
ANDEZ et al.: ULTRAFAST FORWARDING ARCHITECTURE USING A SINGLE OPTICAL PROCESSOR 877
[3] A. Leon-Garcia and I. Widjaja, Communication Networks, 2nd ed. New
York: McGraw-Hill, 2004.
[4] K. Chan, F. Tong, C. K. Chan, L. K. Chen, and W. Hung, “An all-optical
packet header recognition scheme for self-routing packet networks,” in
Proc. OFC 2002, Anaheim, CA, Mar., 2008, pp. 284–285.
[5] M. C. Cardakli, S. Lee, A. E. Willner, V. Grubsky, D. Starodubov, and
J. Feinberg, “Reconfigurable optical packet header recognition and routing
using time-to-wavelength mapping and tunable fiber Bragg gratings for
correlation decoding,” Photon. Technol. Lett., vol. 12, pp. 552–554, May
2000.
[6] T. Lin, K. A. Williams, R. V. Penty, I. H. White, M. Glick, and D. McAuley,
“Performance and scalability of a single-stage SOA switch for 10 ×
10 Gb/s wavelength striped packet routing,” Photon. Technol. Lett.,
vol. 18, pp. 691–693, Mar. 2001.
[7] Y. Lin, M. Yuang, S. Lee, and W. Way, “Using superimposed ASK label
in a 10-Gb/s multihop all-optical label swapping system,” J. Lightw.
Technol., vol. 22, no. 2, pp. 351–361, Feb. 2004.
[8] E. N. Lallas, N. Skarmoutsos, and D. Syvridis, “Coherent encoding of
optical FSK header for all optical label swapping systems,” J. Lightw.
Technol., vol. 23, no. 3, pp. 1199–1209, Mar. 2005.
[9] M. Y. Jeon, Z. Pan, J. Cao, Y. Bansal, J. Taylor, Z. Wang, V. Akella,
K. Okamoto, S. Kamei, J. Pan, and S. J. B. Yoo, “Demonstration of
all-optical packet switching routers with optical label swapping and 2R
regeneration for scalable optical label switching network applications,” J.
Ligthw. Technol., vol. 21, no. 11, pp. 2723–2733, Nov. 2003.
[10] A. E Willner, D. Gurkan, A. B. Sahin, J. E. McGeehan, and M. C. Hauer,
“All-optical address recognition for optically assisted routing in next gen-
eration networks,” IEEE Commun. Mag., vol. 41, no. 5, pp. S38–S44,
May 2003.
[11] P. Prucnal, D. Blumenthal, and P. Perrier, “Photonic-switch with optically
self-routed bit switching,” IEEE Commun. Mag., vol. 25, no. 5, pp. 547–
554, May 1987.
[12] N. Wada and F. Kubota, “Cutting edge technologies on broadband and scal-
able photonic-network—Packet switched networks based on all-optical
label processing,” Opt. Rev., vol. 11, no. 2, pp. 106–112, 2004.
[13] K. Kitayama, N. Wada, and H. Sotobayashi, “Architectural considerations
for photonic IP router based upon optical code correlation,” J. Lightw.
Technol., vol. 18, pp. 1834–1844, Dec. 2000.
[14] A. Teixeira, P. Andre, P. Monteiro, M. Lima, R. Nogucira, and J. da Rocha,
“All-optical routing based on OCDMA headers,” in Proc. 16th Annu.
Meeting IEEE Lasers Electro-Opt. Soc. (LEOS 2003), vol. 2, pp. 1046–
1047.
[15] N. Wada, H. Harai, and W. Chujo, “Multi-hop, 40 Gbit/s variable length
photonic packet routing based on multi-wavelength label switching, wave-
band routing and label swapping,” in Proc. OFC 2002, Anaheim, CA,
Mar., pp. 216–217.
[16] D. B. Hunter and R. A. Minasian, “Programmable high-speed optical code
recognition using fiber Bragg grating arrays,” Electron. Lett., vol. 35,
pp. 412–414, Mar. 1999.
[17] M. C. Cardaki, S. Lee, A. E. Willner, V. Grubsky, D. Starodubov, and
J. Feinberg, “Reconfigurable optical packet header recognition and routing
using time to wavelength mapping and tunable fiber Bragg gratings corre-
lation decoding,” IEEE Photon. Technol. Lett., vol. 12, no. 5, pp. 552–554,
May 2000.
[18] M. C Hauer, J. E. Mcgeehan, S. Kumar, J. D. Touch, J. Bannister,
E. R. Lyons, C. H. Lin, A. A. Au, H. P. Lee, D. S. Starodubov, and
A. E. Willner, “Optically assisted internet routing using arrays of novel
dynamically reconfigurable FBG-based correlators,” J. Lightw. Technol.,
vol. 21, no. 11, pp. 2765–2777, Dec. 2003.
[19] N. Wada and K. I. Kitayama, “A 10 Gb/s optical code division multiplexing
using 8-chip optical bipolar code and coherent detection,” J. Lightw.
Technol., vol. 17, no. 10, pp. 1758–1765, Oct. 1999.
[20] N. Wada, H. Harai, W. Chujo, and F. Kubota, “Multihop 40 Gbit/s vari-
able length photonic packet routing based on multiple wavelength label
switching, waveband routing, and label swapping,” in Proc. Opt. Fiber
Conf. (OFC 2002), vol. 1, pp. 216–217, Paper WG-3.
[21] A. Teixera, M. Lima, R. Andre, and J. da Rocha, “All optical time-
wavelength code router for optical CDMA networks,” in Proc. LEOS
Ann. Meeting. San Diego, CA, 2001, vol. 1, pp. 93–94, Paper ML3.
[22] D. Gurkan, M. C. Hauer, A.B. Sahin, Z. Pan, S. Lee, and A. E.
Willner, “Demonstration of multi-wavelength all optical header recog-
nition using PPLN and optical correlators,” in Proc. Eur. Conf. Opt. Com-
mun. (ECOC), 2001, vol. 3, pp. 312–313, Paper WeB.2.5.
[23] J. Widjija, “IP router using optical address recognition with second har-
monic generation,” Electron. Lett., vol. 38, no. 21, pp. 1268–1269, Oct.
2002.
[24] H. Furukawa, T. Konishi, O. Yu, K. Itoh, and Y. Ichioka, “Design of
header recognition filter for binary phase shift keying in header recognition
unit time-space conversion,” Opt. Rev., vol. 11, no. 2, pp. 119–125,
2004.
[25] M. C. Hauer, J. E. McGeehan, S. Kumar, J. D. Touch, J. Bannister,
E. R. Lyons, C. H. Lin, A. A. Au, H. P. Lee, D. S. Starodubov, and
A. E. Willner, “Optically assisted Internet routing using arrays of novel
dynamically reconfigurable FBG-based correlators,” J. Lightw. Technol.,
vol. 21, no. 11, pp. 2765–2776, Nov. 2003.
[26] P. Seddighian, S. Ayotte, J. B. Rosas–Fern
´
andez, J. Penon, L. A. Rusch,
and S. LaRochelle, “Label stacking in photonic packet switched networks
with using spectral amplitude code labels,” J. Lightw. Technol., vol. 25,
no. 2, pp. 463–471, Feb. 2007.
[27] P. Seddighian, S. Ayotte, J. B. Rosas-Fern
´
andez, J. Penon, S. LaRochelle,
and L. A. Rusch, “Label stacking using spectral amplitude code labels for
optical packet switching,” presented at the OFC 2006, Anaheim, CA,
Mar.
[28] P. Seddighian, S. Ayotte, J. B. Rosas-Fern
´
andez, J. Penon, S. LaRochelle,
and L. A. Rusch, “Self-routed packets with encoded payload and stacked
optical code labels,” presented at the ECOC 2006, Cannes, France,
Sep.
[29] O. Moriwaki, T. Kitoh, T. Sakamoto, and A. Okada, “Novel PLC-based
optical correlator for multiple phase-modulated labels,” Photon. Technol.
Lett., vol. 17, no. 2, pp. 489–491, Feb. 2005.
[30] G. Cincotti, “Full optical encoders/decoders for photonic IP routers,” J.
Lightw. Technol., vol. 22, no. 2, pp. 337–342, Feb. 2004.
[31] G. Cincotti, “Design of optical full encoders/decoders for code-based
photonic routers,” J. Lightw. Technol., vol. 22, no. 7, pp. 1642–1650, Jul.
2004.
[32] G. Cincotti, N. Wada, and K. Kitayama, “Characterization of a full
encoder/decoder in the AWG configuration for code-based photonic
routers—Part I: Modeling and design,” J. Lightw. Technol., vol. 24, no. 1,
pp. 103–112, Jan. 2006.
[33] N. Wada, G. Cincotti, S. Yoshima, N. Kataoka, and K. Kitayama, “Char-
acterization of a full encoder/decoder in the AWG configuration for code-
based photonic routers—Part II: Experiments and applications,” J. Lightw.
Technol., vol. 24, no. 1, pp. 113–121, Jan. 2006.
[34] F. Forghieri, R. W. Tkach, and A. R. Chaplyvy, “WDM systems with
unequally spaced channels,” J. Lightw. Technol., vol. 13, no. 5, pp. 889–
897, May 1995.
[35] G. A. Agrawal, Nonlinear Fiber Optics, 3rd ed. New York: Academic,
2001.
[36] H. Hiura and N. Goto, “All-optical label recognition using self-routing
architecture of Mach–Zehnder interferometer optical switches with semi-
conductor optical amplifiers,” IEICE Trans. Electr., vol. E90-C, no. 8,
pp. 1619–1626, Aug. 2007.
[37] B. R. Koch, Z. Hu, J. E. Bowers, and D. J. Blumenthal, “All-optical
payload envelope detection for variable length 40-Gb/s optically labeled
packets,” IEEE Photon. Technol. Lett., vol. 18, no. 17, pp. 1846–1848,
May 2006.
[38] D. Klonidis, C. Politi, M. O’Mahony, and D. Simeonidou, “Fast and
widely tunable optical packet switching scheme based on tunable laser
and dual-pump four wave mixing,” IEEE Photon. Technol. Lett., vol. 16,
no. 5, pp. 1412–1414, May 2004.
[39] H. Toga, “Long distance transmission experiments using the WDM tech-
nology,” J. Lightw. Technol., vol. 14, no. 6, pp. 1287–1297, Jul. 1996.
[40] W. C. Kong and C. C. Yang, “An algebraic approach to the unequal-
spaced-channel allocation problem in WDM ligthwave systems,” IEEE
Trans. Commun., vol. 45, no. 3, pp. 352–359, Mar. 1997.
[41] S. Song, “Evaluating high order four-wave-mixing using a two chan-
nel WDM system,” in Proc. Conf. Lasers Electro-Opt. (CLEO),San
Francisco, CA, 2000, pp. 204–205, Paper CTuG2.
[42] K. D. Chang, C. C. Yang, and W. C. Kong, “Determination of the FWM
products in unequal spaced channel WDM ligthwave systems,” J. Lightw.
Technol., vol. 18, no. 12, pp. 2113–2122, Dec. 2000.
[43] V. L. Thing, P. Shum, and M. Rao, “Channel allocation algorithm for
WDM systems,” Opt. Exp., vol. 11, no. 11, pp. 1322–1327, 2003.
[44] J. G. Zhang and A. B. Sharma, “Fast frequency allocation in WDM systems
with unequally space channels,” Electron. Lett., vol. 39, pp. 450–452, Mar.
2003.
[45] L. L. Vrizlyn, P. Thing, P. Shun, and M. K. Rao, “Bandwidth-efficient
WDM channel allocation for four-wave-mixing-effect minimization,”
IEEE Trans. Commun., vol. 52, no. 12, pp. 2184–2189, Dec. 2004.
[46] B. Hwang and O. K. Tonguz, “A generalized suboptimum unequally chan-
nel allocation technique—Part I: In IM/DD WDM systems,” IEEE Trans.
Commun., vol. 46, no. 8, pp. 1027–1037, Aug. 1998.
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.

878 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 14, NO. 3, MAY/JUNE 2008
[47] B. Hwang and O. K. Tonguz, “A generalized suboptimum unequally chan-
nel allocation technique—Part II: In coherent WDM systems,” IEEE
Trans. Commun., vol. 46, no. 8, pp. 1027–1037, Aug. 1998.
[48] S. Song, “The number of four wave mixing waves in WDM systems and
its applications,” in Proc. LEOS 2001, vol. 1, pp. 283–284.
[49] K. Okamoto, M. Ishii, Y. Hibino, Y. Ohmori, and H. Toba, “Fabrication
of unequal channel spacing arrayed-waveguide grating multiplexer mod-
ules,” Electron. Lett., vol. 31, no. 17, pp. 1464–1466, Aug. 2008.
[50] J. Buus and E. J. Murphy, “Tunable lasers in optical networks,” J. Lightw.
Technol, vol. 24, no. 1, pp. 5–11, Jan. 2006.
Jos
´
e Bernardo Rosas-Fern
´
andez (M’98) received
the B.S. degree in electronics and telecommunica-
tions engineering from the Autonomous Metropoli-
tan University, Iztapalapa, Mexico City, Mexico, in
1995, and the Ph.D. degree in engineering from
St. Catharine’s College, University of Cambridge,
Cambridge, U.K., in 2005.
From 1995 to 2000, he was with Nippon Electric
Company (NEC), Mexico, where he has also been the
Chief of Engineering for fiber and access communi-
cations systems for Central America market. From
2005 to February 2007, he was a Postdoctoral Research Fellow at the Universit
´
e
Laval, Quebec, QC, Canada. He is currently a Research Associate at the Centre
for Advanced Photonics and Electronics, University of Cambridge. His current
research interests include optical-code-division multiple access (OCDMA), op-
tical packet switching, radio-over-fiber, nonlinear optics, and semiconductor
amplifiers and lasers.
Dr. Rosas-Fern
´
andez is member of the Society of Photographic Instrumen-
tation Engineers (SPIE).
Simon Ayotte (S’04) was born in Quebec, QC,
Canada, in 1979. He received the B.S. degree in en-
gineering physics in 2002 from the Universit
´
e Laval,
Quebec, where he is currently working toward the
Ph.D. degree in optical telecommunications at the
Centre d’Optique, Photonique et Lasers, Department
of Electrical and Computer Engineering.
His current research interests include fiber-optic
communication systems, optical code-division mul-
tiple access (OCDMA), and optical packet switched
networks.
Leslie A. Rusch (S’91–M’94–SM’00) received the
B.S.E.E. (Hons.) degree from the California Insti-
tute of Technology, Pasadena, in 1980, and the M.A.
and Ph.D. degrees in electrical engineering from
Princeton University, Princeton, NJ, in 1992 and
1994, respectively.
From 2001 to 2002, she was the Manager of a
group researching on new wireless technologies at
Intel Corporation. She is currently a Full Professor in
the Department of Electrical and Computer Engineer-
ing, Universit
´
e Laval, Quebec, QC, Canada, where
she is engaged in research in wireless and optical communications. Her current
research interests include optical code-division multiple access using noncoher-
ent sources for metropolitan area networks, semiconductor and erbium-doped
optical amplifiers and their dynamics, and in wireless communications, high per-
formance, reduced complexity receivers for ultra-wideband systems employing
code-division multiple access.
Sophie LaRochelle (M’00) received the Bachelor’s
degree in engineering physics from the Universit
´
e
Laval, Quebec, QC, Canada, in 1987, and the Ph.D.
degree in optics from the University of Arizona,
Tempe, in 1992.
From 1992 to 1996, she was a Research Scientist
at the Defense Research and Development Canada,
Valcartier, QC, where she was engaged in electroop-
tical systems. She is currently a Professor in the
Department of Electrical and Computer Engineer-
ing, Universit
´
e Laval, where she is also a Canada
Research Chair in Optical Fibre Communications and Components. Her current
research interests include active and passive fiber optics components for opti-
cal communication systems including fiber Bragg gratings, optical amplifiers,
multiwavelength, and pulsed fiber lasers.
Dr. LaRochelle is a member of the Optical Society of America (OSA) and
the IEEE-Lasers and Electro-Optics Society (LEOS).
Authorized licensed use limited to: BIBLIOTHEQUE DE L'UNIVERSITE LAVAL. Downloaded on June 23,2010 at 13:57:01 UTC from IEEE Xplore. Restrictions apply.