All-optical Multi-wavelength Virtual Memory Architecture

Design and Performances Analysis

Selma Batti

, Mourad Zghal

and Noureddine Boudriga

Communication Networks and Security Laboratory (CNAS), Engineering School of Communication of Tunis (Sup’Com),

University of Carthage, Ariana, Tunisia

CIRTA’COM Laboratory, Engineering School of Communication of Tunis (Sup’Com), University of Carthage,

Ariana, Tunisia

Keywords: Optical Buffering, All-optical Memory, Fiber Bragg Grating, Tunable Wavelength Conversion.

Abstract: As all-optical memory represents one of the most important lacks in evolution of optical networks; this

paper presents an all-optical virtual memory based on a recirculation loop, with the goal of providing optical

data unit storage in all-optical switching networks. The concept of multi-wavelength signal buffering is

adopted, to realize a shared buffer with an important storage capacity. We propose the organization of the

buffer in two loops, the first as a delay loop and the second as an amplification loop, to improve the

buffering duration and performances. The memory implementation is demonstrated using optical

components such as fiber Bragg gratings (FBG), circulator and tunable wavelength converter. An all-optical

control unit is designed to provide a dynamic and automatic signal buffer managing. An analytical model is

implemented and a simulations set is done to prove that the proposed architecture is able to confine several

signals for a relatively long time as a memory and signals can leave the architecture for a reasonably short

delay after the departure decision is taken. The low penalty observed shows good system reliability.

1 INTRODUCTION

To resolve the increasing need of capacity,

networking is interesting more and more in optical

technologies. However, the absence of optical

memory causes a bottleneck due to the optical-

electronic-optical signal conversion. To improve

optical network performances, all-optical memory is

considered as a crucial point. The design of optical

memory has important effect on the development of

all-optical network and more especially on optical

switching node by reducing the impact of some

technical problems such as contention resolution

(Mack et al., 2010).

We define all-optical virtual memory as a device

able to deliver a signal identical to the received one

but after a certain delay. Optical data would be kept

in optical format throughout the storage time without

being converted into electronic format. The multi-

wavelength memory must be able to enclose several

signals having different wavelengths values at the

same time and each one can be delivered

independently of the others. To be considered as

acceptable, optical buffers must minimally provide

some criteria (Burmeister et al., 2008), such as being

bit rate scalable to greater than 40 Gbits/s, being

able to memorize data units having at least 40 bytes

as size additional to their guard bands and providing

dynamically variable memorization time.

A wide variety of architectures has been

proposed in the literature for the design of optical

buffer. Two main techniques were developed: fiber

delay line (FDL) and recirculation loop. The time

buffer described in (LeGrange et al., 2007) is based

on FDL. The architecture is proposed particularly to

be implemented inside an all-optical router. It

utilizes fast wavelength switching in combination

with an arrayed waveguide grating (AWG) to select

a particular FDL from an array of FDL of varying

length. An integrated optical device consisting of the

silicon delay line and the gate matrix operating as a

buffer at 40 Gbits/s is demonstrated in (Park et al.,

2008). Once optical packets are routed into the delay

line, they are stored in the delay line until the gate

matrix switch re-routes them to the output. The gate

matrix switch is controlled by an electronic device.

Offering a particular small size and a low cross talk,

the chosen components enable a 1.1 ns delay of 40

Gbits/s data packets after one turn.

388

Selma Batti S., Zghal M. and Boudriga N..

All-optical Multi-wavelength Virtual Memory Architecture - Design and Performances Analysis.

DOI: 10.5220/0004071403880395

In Proceedings of the International Conference on Data Communication Networking, e-Business and Optical Communication Systems (OPTICS-2012),

pages 388-395

ISBN: 978-989-8565-23-5

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

In this work, we propose an all-optical multi-

wavelength virtual memory (MWVM). Using all-

optical technologies even for the control part, the

virtual memory is designed to confine optical signals

without resort to electrical conversion during the

stay time in the buffer. The multi-wavelength

concept allows the share of the unique buffer

resource by several signals. Organizing the memory

in two loops by isolating the amplifier in a second

loop improves the memory performances through

guarantee adequate power level and avoiding

additional noise amplification. Moreover, we size

the duration of a first loop to reduce the delay

between the moment of the exit decision and the

signal deliverance and to provide an enough long

buffering duration. Work presented in (Batti et al.,

2010) describes an all-optical virtual memory

allowing the confinement of a single signal. In the

new all-optical multi-wavelength virtual memory,

several signals having different wavelengths can be

buffered at the same time. Each signal is delayed for

time duration independently of the other signals

delays. The designed control unit provides a

dynamic managing of the signal exit without resort

to external intervention. Virtual memory physical

implementation and optical control part are designed

to manage the delay of each signal independently of

the other signals. To dimension the proposed virtual

memory, three criteria are addressed: first loop size,

admitted wavelength number and system bit rate.

The architecture performances are evaluated by an

analysis study and a set of simulations.

The remainder of this paper is organized as

follow: Section 2 is dedicated to the description of

the design of the novel all-optical multi-wavelength

virtual memory; Section 3 presents the analytical

model to evaluate the communication performances;

simulation and results are given in Section 4; finally,

Section 5 concludes this paper.

2 MWVM ARCHITECTURE

In this section, we describe the new architecture of

the all-optical multi-wavelength virtual memory.

Signals arriving at the memory entrance have to be

delayed by turning in the loop. So, each signal has to

perform a number of turn independently of the other

signals. The information about the required turn

number for each arriving signal is generated by a

component upstream from the memory and sent to

the memory control unit. The maximum signal

number turning in the same time in the virtual

memory K represents the memory capacity.

Signals arriving to the input block are considered

synchronized. This can be ensured by the insertion

of synchronizer (Batti et al., 2009) upstream from

the virtual memory. That is why we suppose that the

data units have the same size or, at least, have a

fixed maximum size equivalent to slot duration (S).

The synchronizer is an all-optical device able to

align data units on a slot starting indicated by a

clock signal.

2.1 Loops Architecture and

Functioning

Organized in two loops as illustrated in Figure 1, the

virtual memory is used to buffer several signals

during different time duration. This memorization is

done by enclosing signals in the first loop, named

delay loop (L1 in Figure 1). When the power of a

signal reaches an unacceptable level, it is amplified

in the second loop, named amplification loop (L2 in

Figure 1). Implemented using all-optical

components, the virtual memory consists of an

optical combiner, an optical fiber, a wavelength

converters block, an optical amplifier, three

circulators and two FBG arrays.

Figure 1: Multi-wavelength virtual memory architecture.

The proposed architecture uses a relatively

reduced components number, according to existing

architecture. Moreover, the included components are

well known devices with reduced complexity

functioning. This criterion guarantees that the buffer

architecture is implementable.

The fiber inserted in the delay loop has a fixed

length L. This length can be assimilated to the first

loop length. The second loop encloses only the

amplifier, so its length can be considered as quasi

null. Each revolution signals are delayed by a time

duration

T proportional to the fiber length L; this

duration is named memory sensitivity. The total

delay accumulated by the k

signal after n

revolutions in the buffering block is equal to (n

T).

All-opticalMulti-wavelengthVirtualMemoryArchitecture-DesignandPerformancesAnalysis

389

The fiber length L must be proportional to the

slot duration S, which corresponds to the maximum

data unit length, to confine the total signals. The

physical implementation of the buffer takes the

maximum data unit length into account but remains

independent of each data unit length.

An array of FGBs (Srivastava et al., 2008) is

inserted at the virtual memory output path and a

second array is inserted between the delay loop and

the amplification loop. As the buffer can enclose K

signals at the same time, each array constitutes a

sequence of K reflection bands (RB): the array at the

output path constitutes the first sequence of K RB

named RB

, k

∈

[0, K[ and the array between both

loops constitutes the second sequence of K RB

named RB

, k

∈

[0, K[. As shown in Figure 2, the

and RB

are overlapped and the K bands

consisting of the union of RB

and RB

are

disjointed one of the other.

Figure 2: Organization of the reflection bands sequences.

Signals arriving on the input path pass to the

delay loop throw the combiner. At this moment,

theirs wavelengths are in the band B

, it means that

each k

signal has as wavelength λ

(∈ RB

∩

), which corresponds to overlapping zone

between RB

and RB

. Having these wavelengths,

signals remain enclosed in the delay loop until theirs

wavelengths are converted. Signals pass throw the

optical fiber which induces a delay equal to

T each

revolution. Then, they pass throw the wavelength

converter block. The signal number k leaves the

converter block with wavelength equal to: λ

if it

has to make another turn in the delay loop, λ

(

∈

-(RB

∩

)) if it has to be amplified and λ

(

∈

-(RB

∩

)) if it has to exit the buffering

block. Then, signals reach the first port of the first

circulator and they are led to its second port. The

signals pass throw the second FBGs array which

constitutes the reflection bands RB

, k

∈

[0, K[. Two

cases may occur: signals with wavelengths out of the

, that means signals with wavelengths equal to

, are transmitted to the second circulator; but

signals with wavelengths inside the RB

, that means

signals with wavelengths equal to λ

or λ

, are

reflected to the first circulator. The transmitted

signals arrive at the second port of the second

circulator and are led to its third port. So these

signals enter to the second loop where they are

amplified. Passing from the first port to the second

port of the second circulator, these signals leave the

amplification loop, cross the reflection bands RB

and are led from the second port to the third port of

the first circulator. Also the reflected signals are led

from the second port to the third port of the first

circulator. Signals arriving at the first port of the

third circulator are led to its second port. Then, they

pass throw the first FBGs array which constitutes the

reflection bands RB

, k

∈

[0, K[. As happen with the

first reflection bands, two cases may occur: signals

with wavelengths out of the RB

, that means signals

with wavelengths equal to λ

, are transmitted on the

virtual memory output path; but signals with

wavelengths inside the RB

, meaning signals with

wavelengths equal to λ

or λ

, k

∈

[0, K[, are

reflected to the second port of the third circulator.

Transmitted to the third port of the third circulator,

the reflected signals reenter to the delay loop and are

passed by the combiner to the optical fiber. Signals

remain enclosed in the delay loop until the

wavelength converter block shifts theirs wavelengths

to: (a) λ

and so they are passed to the amplification

loop, or (b) λ

to be exited from the memory.

2.2 Wavelength Converter Block

A wavelength converter block is inserted inside the

delay loop. This block is used to convert the signals

wavelengths independently one of each other. In

fact, several signals can be enclosed in the virtual

memory at the same time (at most, K signals);

signals having to be amplified must have theirs

wavelengths converted to λ

∈

[0, K[); signals

which were just amplified, during the previous turn,

must have theirs wavelengths reconverted to λ

∈

[0, K[); signals having to exit the virtual memory

to the switching block must have theirs wavelengths

converted to λ

∈

[0, K[) and signals having to

turn again in the delay loop must have theirs

wavelengths maintained to λ

(no conversions are

needed).

As shown in Figure 3, the wavelength converter

block consists in a set of K tunable wavelength

converter (Wang et al., 2006) managed by a memory

control unit. When signals turning in the delay loop

arrive at the wavelength converter block, a

demultiplexer separates them on K paths. On each

path an optical sensor is used to detect the arrival

moment of each signal. These sensors inform the

memory control unit of the arrival moment of

OPTICS2012-InternationalConferenceonOpticalCommunicationSystems

390

signals. Each sensor sends an information signal

named sens

, where k

∈

[0, K[ is the range of the

sensor. The wavelength converters make the needed

conversion or still idle according to the decision of

the memory control unit. The control unit generates

K control signals organized in an array of control

wavelength named 







0..−1



. Then, the K paths

are regrouped by the multiplexer.

Figure 3: Wavelength converter block architecture.

2.3 Memory Control Unit

The control unit is the component that manages the

signals paths by delivering synchronous signals to

pilot the wavelength converters. According to the

previously received turn number information (sent

by a component upstream from the virtual memory),

it generates the three kinds of wavelengths

summarized in Equation 1 at different moments.

Using all-optical technologies, the control unit

consists of three main components as shown in

Figure 4: calculator, synchronizer and multi-

wavelength sources (MW sources).

The calculator receives the information about

required turn numbers for each enclosed signal.

According to the physical implementation of the

loops, it computes the amplifications number p

for

each signal. Each time a signal crosses the

wavelength converter block, the calculator is

advertised by the synchronizer and memorizes the

turn number for this signal. According to this

number, the calculator can let the k

MW source

ideal, if no conversions are required; also it can

generates a signal indicating to the MW Sources

∈

[0, K-1], which kind of wavelengths they must

generate (λ

, λ

). So, the calculator has to

realize only two simple operations (increment and

comparison) which can be carried out in optical

domain.

The synchronizer is used to command the

calculator to start the signal generation exactly when

the k

signal arrives at the wavelength converter

block. In fact, at its arrival, signal crosses the k

optical sensor to inform the control unit of its

arrival. If a conversion is required, the control unit

must start generating a continuous wave intended for

the wavelength converter.

The MW sources are a set of multi-wavelength

laser sources generating the wavelengths required to

signal conversion. These components receive control

signals from the calculator, and generate the optical

continuous signals organized in the array of control

wavelength 







0..−1



. The multi-wavelength

laser source can be chosen from the existing

components. However, its implementation should

pay attention to its response time, since it can affect

the system performances.

Figure 4: Memory control unit architecture.

Several technologies can be used to provide a

wavelength generator for the control unit. Among

the most important technologies, one can distinguish

two components; the optical flip-flop memory

(Kurobe et al., 2007) and the laser neural network

(Liu et al., 2004). Due to the described memory

control unit, the proposed virtual memory provides a

dynamical variable memorization time without

resort to electrical domain.

3 PERFORMANCES ANALYSIS

We evaluate performances of the system, and more

precisely of the buffering block, by assessing four

communication parameters: the delay, the

attenuation, the dispersion and the signal to noise

All-opticalMulti-wavelengthVirtualMemoryArchitecture-DesignandPerformancesAnalysis

391

ratio (SNR). In fact, each component crossed by

signals adds some delay and signal distortion.

For simplification reasons, we suppose that: (a)

the circulator and the combiner have identical effect

on signals independently of their input port; (b) all

used FBGs have the same effects on the

communication performances; (c) the dispersion

induced by the cross of the FBGs, the wavelength

converter block and the optical amplifier is

negligible according to the dispersion induced by the

other components; (d) only the optical amplifier and

the wavelength converter block in active mode

increase the noise.

In this work, n represents the turn number in the

delay loop. Each revolution, the crossed components

increase the signal distortion by increasing the

attenuation, the dispersion and the noise. After a

number of turns, the signals power reaches a critical

level. That’s why amplification must be performed.

After n turns in the memorization loop and p

turns in the amplification loop, the k

signal

cumulates the total delay given by equation 1, where

c, n

fib

and L are respectively the light celerity,

refractive index and fiber length; and the T

conv,on

conv,off

, T

FBG,ref

and T

FBG,trx

are respectively the

response times of the active wavelength converter,

the inactive wavelength converter, the reflecting

FBG and the transmitting FBG. One can see that the

delay induced by the optical fiber is the most

important one according to the delay of the

converter and the FBGs. So, the signal rang (k) has a

neglected effect on the cumulated delay.

()

() ( )

()( )( )

[]

21 21

212 1 21

fib

kn FBGref

vonv on vonv off

BG trx

fib

Tn npT

pT npT

knpKpT

=+−−

++ +−−

+− −−+ +

≈

(1)

Equation 2 evaluates the power attenuation occurred

to the k

signal after n turns in the memorization

loop and p turns in the amplification loop. In this

equation, A

comb

, A

fib

, A

cir

, A

conv,on

, A

conv,off

, A

FBG,ref

and

FBG,trx

, respectively appoint the attenuation of the

combiner, fiber, circulator, wavelength convertor in

active mode, wavelength convertor in inactive mode,

FBG when signals are reflected and the FBG when

signals are transmitted. The G

amp

is the optical

amplifier gain.

()

[]

(4 2 1)

(2 1) ( 2 1)

(2 1)

2 ( 1) ( 2 1) (2 1)

kn comb fib cir

conv on conv off

FBG ref amp

BG trx

AnA AL npA

pA npA

np A pG

knpKpA

=+++−

++ +−−

+−− −

+ − −−+ +

(2)

The accumulated dispersion is given by equation 3

for the k

signal after n turns in the memorization

loop and p turns in the amplification loop. The

symbols D

comb

, D

fib

and D

cir

give respectively the

dispersion of the combiner, fiber and circulator. The

λ is the spectral line width of the laser source.

(

)

cirfibcombnk

DpnLDDnD )124(

−++Δ+=

(3)

The equation 4 listed below gives the SNR

expression after n turns in the memorization loop,

inducing a cascade of p amplifications and (2p+1)

wavelength conversions, where P

is the input

signal power, SNR

conv,on

is the SNR introduced by

the wavelength converter in active mode, η

is the

ratio of electrons in higher and lower states, h is the

Plank’s constant,

∇

f is the bandwidth that measures

the noise figure and G

amp

is the optical amplifier

gain.

[]

inonconvkampsp

onconvin

PpSNRGfhp

SNRP

SNR

)12()1(2

++−∇

λη

(4)

4 SIMULATIONS AND RESULTS

4.1 Simulation Model and

Parameterization

To demonstrate the proposed memorization

function, a multi-wavelength model implemented

using the OptiSystem simulator and Matlab is

presented in this section. The laser sources are

modeled as pseudo-random bit sequence generators

with variable data unit size. Mach-Zehnder

modulators are used with continuous wave lasers

having a power of 1mW, and NRZ generators. The

wavelength converter block is represented by

developed Matbab co-simulator components. The

FBG arrays are modeled by subsystems of FBGs.

The optical receivers are modeled as PIN

photodetectors and low pass Bessel filters. The data

unit size is fixed to 1500 bytes.

To evaluate the performances limits of the

system, we need to implement the worst case, which

means, when all the transmitters generate signals

having to be enclosed in the buffering block for the

maximum time duration.

A set of simulations is performed for several

buffering fiber length, capacity and bit rate values.

The variations of the delay, attenuation, dispersion

and SNR while the turn number is increasing are

collected from the list of signal port data of the

simulation model layout at the output ports. After

OPTICS2012-InternationalConferenceonOpticalCommunicationSystems

392

receptions, signals are translated to electrical domain

and analyzers show their eye diagram and calculate

the maximum Q factor.

4.2 Results Analysis

To evaluate the communication parameters variation

in function of the turn number and the signal rank, a

first set of simulations is performed where the

buffering fiber length is fixed to be equivalent to one

data unit size (59,95m), the buffering capacity is

equal to four wavelengths and the system bit rate is

40 Gbits/s. The attenuation, the dispersion and the

SNR of each signal is depicted while the turn

number is increasing.

By examining the curves of Figure 5, one can say

that the attenuation depends of the signal rank (k). In

fact, as previously illustrated in equation 6, the

number of crossed FBG depends of the signal rank.

Also, this variation can be explained by the effect of

the amplifier on each signal according to its

wavelength value.

Figure 6 illustrates the variation of the dispersion

while the turn number is increasing for the different

signal ranks. It is evident that when turn number

increases, the dispersion increases also, as the

number of crossed equipments increases. As the

fiber, one of most equipment affecting the signal

dispersion, adds varied dispersion on signals

according to theirs wavelengths, the signal

dispersions depend of the signal ranks.

The SNR variation of each signal when turn

number is growing is depicted in Figure 7. As each p

turns in the first loop, signals have theirs

wavelengths converted and are amplified, it is clear

that the SNR value decreases. As the amplifier

contribution in the SNR depends of the signal

wavelength value, curves in Figure 7 are

distinguishable.

Figure 5: Variation of the attenuation for various turn

number and signal ranks.

Figure 6: Variation of the dispersion for various turn

numbers and signal ranks.

Figure 7: Variation of the SNR for various turn numbers

and signal ranks.

To evaluate the effect of the architecture

characteristics on signals quality, three simulation

sets are performed by fixing two characteristics and

varying a third one: the system bit rate, the buffering

fiber length and buffering capacity. Firstly, the

buffering fiber length and the buffering capacity are

fixed respectively to the equivalent one data unit

size (1500 bytes) and four wavelengths and the

system bit rate is varied to 2,5, 10 and 40 Gbits/s,

secondly, the capacity and bit rate are fixed

respectively to four wavelengths and 40 Gbits/s and

the buffering fiber length is varied to the equivalent

of 1, 5 and 10 data unit length, and thirdly, the bit

rate and the buffering fiber length are fixed to 40

Gbits/S and the equivalent of one data unit length

(59,95 m) and the buffering capacity is varied to

one, two, four and height wavelengths. The

maximum Q factor is depicted for each signal while

the turn number is increasing. Then, the maximum Q

factor average is calculated.

In Figure 8, the maximum Q factor average

variation for several system bit rates is showed. One

can see that the signal quality decreases when the

turn number growth. But it remains in acceptable

All-opticalMulti-wavelengthVirtualMemoryArchitecture-DesignandPerformancesAnalysis

393

row for the three bit rates.

Figure 8: Maximum Q factor average variation for various

system bit rates.

Figure 9: Maximum Q factor average variation for various

buffering fiber lengths.

The Maximum Q factor average variation for

several buffering fiber lengths is illustrated in Figure

9. While the fiber length (size of the delay loop) is

growing, it degrades the signal quality. This fact can

be explained by the attenuation and dispersion rising

when the crossed fiber length increases.

Figure 10 shows the Maximum Q factor average

variation for several buffering capacities. Curves

prove that the buffering capacity decreases the

signals quality while increasing. This fact can be

justified by the crosstalk effect of signals one on

another.

Architecture presented in (Park et al., 2008)

operates at 40Gbits/s with a single wavelength

capacity and 9cm fibre length. After one turn, signal

is delayed by 1.1ns with a power penalty of 2.4dB

and a Q factor of 6. Authors predict that the

architecture is able to achieve 9 or 10 recirculations

with the same power penalty. To obtain more delay,

using a longer delay line requires the insertion of

amplification inside the delay line.

As our architecture includes amplifications only

when needed, in 40Gbits/s system and with four

wavelengths capacity, it is able to delay signals for

more than 30μs with a Q factor average of 8.

Figure 10: Maximum Q factor average variation for

various buffering capacities.

To summarize, one can say that simulations

prove that the proposed all-optical multi-wavelength

virtual memory is able to memorize several data

units having variable sizes in scalable bit rate

systems. Due to the simulation limits, no simulations

for more than 8 wavelengths as buffering capacity

were performed. However, using the aggregation of

several virtual memories can provide an important

number of memorized signals.

5 CONCLUSIONS

This work presents a novel architecture of an all-

optical multi-wavelength virtual memory based on

tunable wavelength converter and FBGs. The

architecture is organized in two loops: the first loop

encloses the signals during the buffering time, and

the second one amplifies the signal optical powers

just when they reach a critical level. Several signals

can be memorized at the same time and the entrance

and the exit of each one is managed independently

of the others. The proposition includes all-optical

control parts making the virtual memory

independent of external events. All decisions are

made according to required turn number for each

signal calculated on the base of previously received

information. The feasibility of this proposition is

demonstrated by analyzing the performances of the

buffering block. This analysis is confirmed by

simulation studies. Our architecture can memorize

an important number of signals at the same time

with a fine response time (sensitivity) quasi equal to

the data unit duration

T in a system of 40Gbits/s

with a relatively acceptable signal distortion.

OPTICS2012-InternationalConferenceonOpticalCommunicationSystems

394

The all-optical multi-wavelength virtual memory

can be implemented inside a switching node to be

used as a solution to various traffic engineering

tasks. Several traffic engineering applications can be

addressed in optical burst switching network such

as: contention resolution, delay based quality of

service (QoS) provision, call admission control and

congestion control.

REFERENCES

Batti, S., Zghal, M., Boudriga, N., Hall, T., 2009. An all-

optical synchronizer for switching node using single-

sideband modulator and fiber Bragg gratings. In ISCC,

Tunisia, 5-8 July.

Batti, S., Zghal, M., Boudriga, N., 2010. A new all-optical

switching node including virtual memory and

synchronizer. In Journal of Networks, 5(2), pp. 165-

179.

Burmeister, E.F., Blumenthal, D.J., Bowers, J.E., 2008. A

comparison of optical buffering technologies. In

Optical Switching and Networking, 5(1), pp. 10-18.

Kurobe, T., Kimoto, T., Muranushi, K., Kagimoto, T.,

Kagi, N., Kasukawa, A., Wu, J., Otani, E., Arimoto,

H., Tsuji, S., 2007. Tunable laser source for fast

wavelength switching using a short-cavity DBR laser

packaged with wavelength locker. In National Fiber

Optic Engineers Conference, 25-29 March.

LeGrange, J. D., Bernasconi, P., Simsarian, J. E., Neilson,

D. T., Stiliadis, D., Gripp, J., Zirngibl, M., 2007.

Demonstration of a time buffer for an all-optical

packet router. In Journal of Optical Networking, 6(8),

pp. 975-983.

Liu, Y., Hill, M.T., Geldenhuys, R., Canbretta, N., de

Waardt, H., Khoe, G,D., Dorren, J.S., 2004.

Demonstration of a variable optical delay for a

recirculating buffer by using all-optical signal

processing. In IEEE Photonics Technology Letters,

16(7), pp. 1748-1750.

Mack, J.P., Burmeister, E.F., Garcia, J.M., Poulsen, H.N.,

Biljana Stamenic, Kurczveil, G., Nguyen, K.N.,

Hollar, K.,Bowers, J.E., Blumenthal, D.J., 2010.

Synchronous optical packet buffers. In, IEEE Journal

of Selected Topics in Quantum Electronics, 16(5), pp.

1413-1421.

Park, H., Mack, J. P., Blumenthal, D. J., Bowers, J. E.,

2008. An integrated recirculating optical buffer. In

Optical Express, 16, pp. 11124-11131.

Srivastava, R., Singh, R. K., Singh, Y. N., 2008. Fiber-

optic switch based on fiber Bragg gratings. In IEEE

Photonics Technology Letters, 20(18), pp. 1581-1583.

Wang, J., Sun, J., Kurz, J. R., Fejer, M. M., 2006. Tunable

wavelength conversion of ps-pulses exploiting

cascaded sum- and difference frequency generation in

a PPLN-fiber ring laser. In IEEE Photonics

Technology Letters, 18(20), pp. 2093-2095.

All-opticalMulti-wavelengthVirtualMemoryArchitecture-DesignandPerformancesAnalysis

395