Multirate Multiservice All-Optical Code Switched GMPLS ...hshalaby/pub/farghaljocnA.pdf · Switched GMPLS Core Network Utilizing Multicode Variable-Weight Optical Code-Division Multiplexing

Multirate Multiservice All-Optical CodeSwitched GMPLS Core NetworkUtilizing Multicode Variable-Weight Optical Code-Division

MultiplexingAhmed E. Farghal, Hossam M. H. Shalaby, and Zen Kawasaki

Abstract—A multicode variable-weight (MCVW) tech-nique is proposed for generalized multiprotocol labelswitching (GMPLS) optical networks in order to supportmultirate and integrated multimedia services. Under thistechnique, the number of simultaneously assigned code-words to each user is a function of the data rate of the ser-vice class, while quality-of-service differentiation isachieved using variable-weight codewords for each serviceclass. The traffic behavior of the network is modeled usinga multiservice loss model, and the probability density func-tions of the number of busy codes in the fiber link areobtained by the Kaufman–Roberts algorithm. In order toanalyze the performance of the proposed multiservice mul-ticode GMPLS optical network, several measures arederived and investigated, specifically, the bit-error rate,probability of degradation, blocking probability, andsteady-state throughput. These performance measuresare obtained for two different receiver structures, namely,correlation receivers with and without hard limiters. Theperformance of our optical GMPLS network, based onthe multicode switching path, is compared with that oftraditional optical GMPLS networks, based on the labelswitching path. The results show the superiority of theproposed technique when compared to traditional ones.

Index Terms—Generalized multiprotocol label switching(GMPLS); Multicode switching path (MCSP); Multirate;Optical code-division multiplexing (OCDM); Optical net-works; Quality of service (QoS); Throughput.

I. INTRODUCTION

D riven by the insatiable demand for bandwidth in-crease, as a consequence of the emergence of

multimedia applications, optical networks are the superiorcandidates to support huge bandwidth as well as diverseservice demands in a cost-effective manner. Wavelength-division multiplexing (WDM) techniques provide platformsto exploit the potential huge capacity of the fiber-optictransmission medium by simultaneously transmittingdata on multiple wavelengths on a single fiber. Due tooptical–electrical–optical (OEO) conversion at everyswitching or routing node, WDM capability is not com-pletely exploited in a whole network due to the slownessof data processing in the electronic domain. All-optical net-works (AONs), e.g., optical circuit switching (OCS), opticalburst switching (OBS), and optical packet switching (OPS)networks in which data processing is carried entirely inthe optical domain without OEO conversion, have beenproposed to cope with the electronic switching bottleneck.OPS is viewed as the most promising technology in futureAONs, due to its characteristics of practical implementa-tion of IPs in the optical domain, high-speed data transmis-sion, data rate/format transparency, power efficiency,fine granularity, and flexibility [1].

One of the most efficient protocols, proposed for futureAONs, is generalized multiprotocol label switching(GMPLS) [2,3]. GMPLS protocols are connection-orientedcontrol-plane protocols for setting up label switching paths(LSPs) that are identified by labels. A GMPLS protocol ex-tends multiprotocol label switching (MPLS) to supportdevices that perform packet switching, time-division multi-plexing (TDM), wavelength (λ) switching, wavebandswitching, and fiber switching. GMPLS protocols offereffective network resource management and traffic engi-neering (constrained based routing), and simplify thenetwork implementation as a whole [4,5].

Utilizing optical code-division multiplexing (OCDM)technology for data switching in optical core networkshas been discussed in [6–9], where the code of the receivedoptical data is considered as a label. Advantages such asasynchronous accessibility [10], distributed control, anddifferentiated services with quality of service (QoS) atthe physical layer have made OCDM rather attractivetoward all-optical communications. Moreover, OCDMhttp://dx.doi.org/10.1364/JOCN.6.000670

Manuscript received December 4, 2013; revised April 8, 2014; acceptedJune 11, 2014; published July 14, 2014 (Doc. ID 201977).

A. E. Farghal (e-mail: [email protected]) is with the Depart-ment of Electronics and Communications Engineering, Egypt-Japan Uni-versity of Science and Technology (E-JUST), Alexandria 21934, Egypt.

H. M. H. Shalaby is with the Department of Electronics and Communi-cations Engineering, Egypt-Japan University of Science and Technology(E-JUST), Alexandria 21934, Egypt, and is on leave from the Electrical En-gineering Department, Alexandria University, Alexandria 21544, Egypt.

Z. Kawasaki is with the Graduate School of Engineering, OsakaUniversity, Osaka 565-0871, Japan.

670 J. OPT. COMMUN. NETW./VOL. 6, NO. 8/AUGUST 2014 Farghal et al.

1943-0620/14/080670-14$15.00/0 © 2014 Optical Society of America

http://dx.doi.org/10.1364/JOCN.6.000670

provides more label space and fine subwavelength datagranularity for GMPLS networks. Two approaches, opticalcode (OC)-labeled path and OCDM path, have been studied[6]. In OC-labeled path, an OC label is only attached tothe head of a packet. This limits the throughput becausethe packets on the same wavelength have to be processedserially. In OCDM path, the whole data packet includingthe header is encoded using a distinct OC. OCDM pathscan be multiplexed onto the same wavelength and enablethe router to process packets in parallel, which results in asimple architecture with no need for optical buffers.Furthermore, hybrid WDM/OCDM-based routing, in whichboth the wavelength and the OC of packets (λ; C) are con-sidered as the identifying label, has been studied in [6,7].

Future optical networks are expected to providebroadband access to a high number of users with very dif-ferent transmission rates and QoS traffic requirements.Recently, optical orthogonal frequency division multiplex-ing (OFDM)-based elastic optical network architecturewith flexible data rate and spectrum allocation called spec-trum-sliced elastic optical path (SLICE) was proposed in[11]. SLICE architecture can provide fine-granularitycapacity to connections by elastically allocating spectrumusing a variable number of low-rate OFDM subcarriersaccording to the transmitted data rate.

Various techniques have been proposed to support multi-rate services provision using OCDM, e.g., varying the codelength [12,13], adopting optical fast frequency hopping[14], and adopting multicode-keying (MK) schemes [15,16].

A different approach to provide data rate differentiationis to use multicode (MC) techniques, in which a data ratedifferentiation is performed by assigning a set of codesaccording to the required data rate of each service class[17–20]. Due to the fixed code length of different classes,the system performance is not degraded. However, becauseof the need for transmitting many codes simultaneously,the number of supported users in the MC technique isdecreased, which limits its applicability. Code families withlarge cardinalities, e.g., two-dimensional (2D) wavelength-hopping time-spreading (WHTS) codes, are required inthe MC scheme. Hybrid OCDM/WDM is an alternativetechnique to resolve the problem of limited codes, whereevery code is reusable and can be sent simultaneously atdifferent wavelengths [21]. The bit-error rate (BER) atthe receiver side can be controlled using variable-weight(VW) optical codes, providing a provision of QoS differen-tiation in OCDM networks [22].

In this paper we propose to use multicode variable-weight optical-orthogonal codes (MCVW-OOCs) for sup-porting multirate and multi-QoS transmission andproviding path switching in optical GMPLS networks,where different service classes are accommodated. We referto the path switching and GMPLS network based on theMC scheme as the multicode switching path (MCSP) andoptical multicode switching-GMPLS (OMCS-GMPLS) net-work, respectively. The OMCS-GMPLS network providesthe support of various data rates and multi-QoS by elasti-cally assigning a number of fixed length codewords andchoosing an appropriate code weight according to the

connection demands. In the proposed network, we considertraffic with a fixed bandwidth and service requirementscase, while the case of traffic that, upon arrival, may havedifferent possible bandwidth requirements depending onthe bandwidth availability will be investigated in futurestudies.

In order to investigate the OMCS-GMPLS networkperformance, three important performance metrics arederived for correlation receivers with and without hardlimiters, which are the blocking probability, the BER, andthe steady-state throughput. The effects of the number ofsimultaneous active users, user activity, and offered trafficloads on these performance metrics are also investigated.

The rest of this paper is organized as follows. Section IIis devoted to the architecture and the operation of theproposed OMCS-GMPLS network. The performance ofthe OMCS-GMPLS network is analyzed in Section III.Section IV presents the details of designing a two-classOMCS-GMPLS network and the numerical results, includ-ing a comparison between the performance of the proposedOMCS-GMPLS network and that of the OCS-GMPLS net-work, which utilizes MLVW-OOCs. Finally, the paper isconcluded in Section V.

II. MULTISERVICE OMCS-GMPLS NETWORK

ARCHITECTURE AND OPERATION

In this section we describe the multiservice OMCS-GMPLS network architecture and introduce a wavelengthsharing code partitioning (WSCP) policy. In addition, wederive an expression of the blocking probability for eachclass of traffic.

A. Network Architecture and Multicode SwitchingMechanism

In the proposed multiservice OMCS-GMPLS network,MCVW-OOC is utilized as the signature sequences.The MCVW-OOC family is characterized by �L;w�fw1;w2;…;wQg;D�fd1;d2;…;dQg;F�fF1;F2;…;FQg;Q;I�,where L is the fixed code length for all classes, Q is thenumber of specified classes in the network, and wj, dj,and Fj (for any j ∈ Ω � f1; 2;…; Qg) denote the code weight,the fraction of codewords with weight wj in class j, and thenumber of parallel codes assigned to class j users, respec-tively. In addition, I indicates the cross-correlation matrix,which is defined as

I �def �I�n;m�;∀ n;m ∈ Ω�; (1)

where I�n;m� denotes the maximum cross correlation be-tween the codewords in classes n and m. It is clear thatthere are exactly di · jCj codewords with weight wj in classj, where jCj represents the overall cardinality of availablecodewords per wavelength.

In a multiclass OMCS-GMPLS network, core labeledtraffic is identified by the input/output port, wavelength,

Farghal et al. VOL. 6, NO. 8/AUGUST 2014/J. OPT. COMMUN. NETW. 671

and optical code, �i; λ; C�, which are considered as the avail-able network resources. Multicode data forwarding along apredefined path, called MCSP, is performed using wave-length switching, OC switching, or both. MCSP consistsof a number of coded packets that follow the same pathfrom source to destination. The number of codes in eachMCSP is determined based on the required data rate foreach forward-equivalent class (FEC). The label distributionprotocol (LDP) determines the incoming and outgoing labelof a path in each intermediate router and records this in-formation in the label-lookup table of the corresponding in-termediate routers. In order to provide multiclass MCSP,the label-lookup table is classified based on the data rateand QoS of each FEC. Moreover, LDP should be modified toassign simultaneously multiple labels based on each FECrequired data rate. An edge optical multicode switchingrouter (EOMCSR), which connects the OMCS-GMPLS corenetwork to other networks, assigns a number of codes(labels) to the incoming packets based on their FECsand destination addresses. In the core network, the coreoptical multicode switching router (OMCSR) performsrouting and forwarding functions by only recognizing thelabel �iin; λin; Cin� of the incoming optical data in the optical

domain and determines the outgoing label �iout; λout; Cout�from its internal label-lookup table based on the incominglabel.

Figure 1 shows a MC-based routing example in a simplemultiservice OMCS-GMPLS network. It is assumed thatthere are two FECs, the high-bit rate class (FEC1), whichassigns two codes simultaneously, and the low-bit rateclass (FEC2), which assigns only one code. Moreover, fourMCSPs are established, and their incoming/outgoing labelsare recorded in the label-lookup table of the core OMCSRs.The number of labels assigned to each MCSP depends onits FEC. For instance, FEC1-MCSP1 assigns two labels andthe packets are coded from EOMCSR1 to EOMCSR4 by��λ2; C1

1�; �λ2; C12��, ��λ2; C1

3�; �λ2; C14��, and ��λ1; C1

3�; �λ1; C14��.

Based on the wavelength and code conversion of the incom-ing packets, four scenarios of label swapping can be used inMCSP; as shown in the OMCSR1 label-lookup table, FEC1-MCSP1 and FEC2-MCSP4 packets are routed by changingthe code and wavelength of the incoming packets, respec-tively, while FEC2-MCSP2 packets are forwarded withoutany conversion. From the OMCSR3 label-lookup table,FEC1-MCSP3 packets are forwarded by both wavelengthand code conversion.

Fig. 1. Multiservice OMCS-GMPLS network architecture with end-to-end operation.


The architecture of OMCSR is similar to that of an op-tical code switching router (OCSR) presented in [23] withthe difference that for class j, which requests Fj codes, thecontroller allocates the required Fj decoders and encoderssimultaneously as shown in Fig. 2.

In Fig. 2, the wavelength demultiplexer first separatesthe received optical signal into W components based onthe wavelength. Every wavelength is split using an opticalsplitter through a Q decoder array of the defined serviceclasses, and in each decoder array C�j� different tapped de-lay line (TDL) decoders, each of which is matched to one ofthe C�j� designed codewords in class j. After decoding theincoming signals, the controller searches the label-lookuptable to make a routing decision to determine the numberof labels and which outgoing labels are assigned to the out-going packets considering incoming label �iin; λin; Cin� andtheir FEC. If the incoming label wavelength λin and out-going label wavelength λout are different, an opticalcross-connect (OXC) first fed the output of the decoder intoa tunable wavelength converter (TWC), which modulates iton a λout selected by the controller and then feeds the out-put to the input of the corresponding encoder of Cout. Afterinput label swapping, the outputs of the encoder arrays ofdifferent service classes are combined using an optical cou-pler, resulting in a new transmission signal in the wave-length. Then different wavelengths are multiplexed bythe wavelength multiplexer, and forwarded to the nextswitching node.

B. Wavelength Sharing Code Partitioning Policy

Figure 3 illustrates the fiber bandwidth partitioning us-ing the WSCP policy to provide Q service classes in aOMCS-GMPLS network. We assume that the total numberof wavelengths available in a single-fiber link is W, whichare shared among all service classes. Isolation between dif-ferent classes is performed by dividing the available code-words in each wavelength into Q different code sets, eachwith C�j�; j ∈ Ω available codewords. The codes of each code

set are assigned to one of the Q classes according to therequired QoS. In the WSCP policy, the performance andcapacity of the network are limited by multiple access in-terference (MAI) from codes from the same service classand from other classes in the same wavelength.

The number of available codewords in class j in eachwavelength should achieve the following bound:

0 < C�j� < jCj; j ∈ Ω; (2)

with the additional constraint on the overall cardinality jCj:

jCj �XQj�1

C�j� �XQj�1

djjCj: (3)

In addition, we have the following constraint [22]:

jCj ≤ �L − 1�PQj�1 djwj�wj − 1�

: (4)

It should be noticed that the number of available codes inclass j is

Decoder 1

Decoder

Decoder

Decoder

Channel

Other classes channels

Class 1

Class 2

Class Q

Channel 1

Channel 2

Splitter . . . . .

Fiber

DM

UX

. . .

. . .

. . . . . . . . .

. . .

(a)

Encoder 1

Encoder

Encoder

Encoder

Channel

Other classes channels

Channel 2

. . .

. . .

. . . . . .

. . .

Channel 1

Class 1

Class 2

Class Q

Com

biner . . . . . . .

Fiber

MU

X

(b)

Fig. 2. Multiservice multicode decoder and encoder arrays used in OMCS-GMPLS router. (a) Multiservice multicode decoder array.(b) Multiservice multicode encoder array.

Fig. 3. Fiber bandwidth partitioning of OMCS-GMPLS networkinto Q service classes using WSCP policy.


L�j� � W × C�j�: (5)

In Fig. 3, we assume that a connection request of each ofthe classes with Fj, j ∈ Ω simultaneous codewords (servers)requirements arrives according to a Poisson process witharrival rate γj and exponentially distributed holding timewith mean 1∕μj. Hence, class j users are assigned simulta-neous free Fj codewords (i.e., Cj

1;…; CjFj) from the available

code set in class j in any wavelength.

C. Blocking Probability

Here, we are interested in the blocking probabilityof each class of traffic. A fiber link using the WSCPpolicy operates as Q independent service classes. In thiscase, the OMCS-GMPLS link forms a multiservice losssystem, where users of some classes may require morethan one code (server) simultaneously. Notice that theM∕M∕L�j�∕L�j� Erlang’s model is a multiserver single-service loss model, while here we consider a multiservermultiservice loss model. The number of valid states forclass j traffic is Γ�j� � ⌊L�j�∕Fj⌋, and the state transitiondiagram is as shown in Fig. 4. A transition from state�n − Fj� to state �n� is due to a new arrival of class j requestrequiring any Fj free codes with rate γj, while transitionfrom state �n� to state �n − Fj� is due to a completion ofclass j service with rate Fjμj. Here n denotes the number ofoccupied codes in class j. The unnormalized steady-stateprobability ~P�j�

N �n� of being in state n can be calculatedby Kaufman–Roberts recursion as follows [24,25]:

~P�j�N �n� �

8<:0; n < 0;1; n � 0;FjAj

n~P�j�N �n − Fj�; 0 < n < L�j�;

(6)

where Aj � γj∕μj denotes the offered traffic per idle user.The above probabilities are then normalized such thatPL�j�

n�0 P�j�N �n� � 1:

P�j�N �n� �

~P�j�N �n�PL�j�

k�0~P�j�N �k�

: (7)

When a connection request of service class j arrives andcannot find Fj free codes (servers), its service is denied andit is blocked and cleared from the system. Hence, the block-ing probability P�j�

B for class j can be found by adding up thesteady-state probabilities of all states with code occupan-cies higher than L� j� − Fj:

P� j�B �

XL� j�n�L� j�−Fj�1

P� j�N �n�: (8)

The OMCS-GMPLS network can provide fine subwave-length granularity capacity to connections. Moreover, serv-ing a class j connection request that requires Fj codewordsis done by finding any Fj unoccupied codewords in anywavelength, hence removing the spectrum contiguity con-straints found in an OFDM-based elastic network [11].However, the OMCS-GMPLS network poses some draw-backs; for example, serving class j connection requests re-quires Fj encoders/decoders in each OMCSR. In addition,high-rate classes will experience higher blocking probabil-ity than low-rate classes.

III. PERFORMANCE EVALUATION

In this section we derive expressions for the BER of theproposed network. Furthermore, we obtain expressionsfor both the probability of degradation and systemthroughput.

A. Multiservice Multicode BER Evaluation

In this subsection, the BER is derived for a multiserviceMCVW-OOC based OMCS-GMPLS network for tworeceiver structures, namely, correlation receivers withand without hard limiters. In general, the performanceof an OCDM-MCSP is primarily affected byMAI from otherpaths and from the codes in the same path. In this paper,we neglect the effects of shot noise, thermal noise, andother fiber impairments, and consider MAI as the maindegrading factor to the system performance. This makesthe OCDM-MCSP provisioning more straightforward. Inaddition, we assume a chip synchronous interferencemodel, which results in an upper bound on the system per-formance [26]. We further assume that the network ispower-controlled so that all packets arrive at the receiverwith equal power; therefore all connections would create anequal amount of interference.

1) BER of a Correlation Receiver: In an �L;w;D; F;Q; I�MCVW-OOC with correlation constraints equal to 1, i.e.,�I�n;m� � 1; ∀ n;m ∈ Ω�, each interfering code may con-tribute only one chip interference with the intended user.Let pkq denote the probability that a codeword of class qmakes one chip interference with a codeword of class k.Since in a codeword of class q, wq marked chips out of Lchips may interfere with any of the wk marked chips ofa class k user, we can write

pkq � wkwq

L; k; q ∈ Ω: (9)

We assume that there are Ki active users in class i, i ∈ Ω. Auser in class i is assigned Fi codewords, each of weight wi.The total number of active codes in the network, K, is thus

K � F1K1 � F2K2 � � � � � FQKQ: (10)Fig. 4. State transition diagram for class j with Fj parallel codes.


Furthermore, there are

F1K1 � � � � � Fk−1Kk−1 � FkKk − 1

� Fk�1Kk�1 � � � � � FQKQ � K − 1 (11)

codes that could interfere with the desired user in class k.The bit-error probability Pe�k� of a class k user, withwk codeweight and Fk parallel codes, given K codes in the networkis evaluated as

Pe�k� � PrferrorjK codesg

�XF1K1

l1�0

…

XFkKk−1

lk�0

…

XFQKQ

lQ�0

Prfl interferersjK codesg

× Prfa bit errorjl interferers;K codesg; (12)

where we have defined the interfering vector

l �def f�l1;l2;…;lQ�∶lj ∈ f0;1;…; FjKjg∀ j ∈ f1;…k − 1; k� 1;…; Qg;lk ∈ f0;1;…; FkKk − 1gg: (13)

The first conditional probability in the right-hand side ofEq. (12) can be evaluated as follows:

PrflinterferersjKcodesg��FkKk−1

lk

�plkkk�1−pkk�FkKk−1−lk

×YQ

q�1q≠k

�FqKq

lq

�plqkq�1−pkq�FqKq−lq ;

(14)

where pkk is given by Eq. (13) with q � k.

Assuming that data bits “1” and “0” are equally likely�Prf0g � Prf1g � 1∕2�, the last conditional probability inEq. (12) is calculated (for a class k user) as follows. The cor-relation receiver decides a data bit “1” was transmitted ifthe total received pulses Z from all weighted chips is notless than a threshold Th. A data bit “0” is decided other-wise. For optimal operation, the decision threshold Th isset to code weight of the intended service class wk [26].Thus,

Prfa bit errorjl interferers;K codesg

� 12Prfa bit errorjl interferers;K codes;1was sentg

� 12Prfa bit errorjl interferers;K codes;0was sentg

� 12PrfZ ≥ wkjl interferers;K codes;0was sentg

� 12

1QQi�1 2

li

Xi1;i2 ;…;iQ∈Ψ

YQi�1

�li

ii

�; (15)

where

Ψ�def��i1; i2;…; iQ�∶ij ∈ f0;1;…;FjKjg

∀ j∈ f1;…k−1;k�1;…;Qg; ik ∈ f0;1;…;FkKk −1g;

wk ≤XQq�1

iq ≤K−1�: (16)

Finally, substituting in Eq. (12), we get

Pe�k� �XF1K1

l1�0

…

XFkKk−1

lk�0

…

XFQKQ

lQ�0

×�FkKk − 1

lk

�plkkk�1 − pkk�FkKk−1−lk

×YQ

q�1q≠k

�FqKq

lq

�plq

kq�1 − pkq�FqKq−lq

×�12

1QQi�1 2

li

Xi1 ;i2;…;iQ∈Ψ

YQi�1

�li

ii

��: (17)

2) BER of a Correlation Receiver With a Hard Limiter:For the correlation receiver with a hard limiter, an erroroccurs only when a data bit “0” is transmitted while thenumber of interfering pulses in every weighted chip posi-tion of the codeword of intended service class k is nonzero.We define

χk �def f1;2;…; wkg: (18)

Given that there are l interfering users, each interfering atexactly one chip position, there is a set of possible interfer-ence patterns. To describe these patterns, we define a Q ×wk interference matrix L whose element lj

i; i ∈ Ω, j ∈ χk,represents the number of users (out of available li classi users) that interfere with the jth weighted chip of thedesired user:

L �def

0BBBBB@

l11 l2

1 … lwk1

l12 l2

2 … lwk2

… …

l1Q l2

Q … lwkQ

1CCCCCA: (19)

Notice that the sum over the rows of L gives the vector las defined in Eq. (13). On the other hand, the sum over thecolumns of L gives the vector α, defined as

α�fα1;α2;…;αwkg�def

(XQi�1

l1i ;XQi�1

l2i ;…;

XQi�1

lwki

): �20�

Here αi; i ∈ χk represents the number of interfering usersthat overlap with the ith pulse position of the class kdesired user. Since every interfering user contributes oneand only one pulse, the interference matrix elements mustsatisfy


lji ≥ 0; any i ∈ Ω; j ∈ χk;

XQi�1

Xwk

j�1

lji �

XQi�1

li �Xwk

j�1

αj � l: (21)

For a given l there is a set of matrices F l that satisfyEq. (21):

F l �def�L ∈ NQ×wk∶

XQi�1

Xwk

j�1

lji � l

and lji ≥ 0; any i ∈ Ω; j ∈ χk

�; (22)

where N is the set of natural numbers. Based on theanalyses in [27,28], the bit-error probability Pe�k� of a classk user with wk code weight and Fk parallel codes givenK codes in a multiservice multicode network can be evalu-ated as follows:

Pe�k� � PrferrorjK codesg

�XF1K1

l1�0

…

XFkKk−1

lk�0

…

XFQKQ

lQ�0

Prfl interferersjK codesg

×XL∈Fl

P�L;F l� · Pbe�k��L�; (23)

where P�L;F l� is the probability of occurrence of interfer-ence pattern L ∈ F l given l � �l1;l2;…;lQ� interferingusers, and Pbe�k��L� is the BER of a class k user given in-terference pattern L. A user is equally likely to interfereat any one of thewk pulse positions independent of all otherusers. Thus, pattern L can be produced in

l!PQi�1 l

1i !PQ

i�1 l2i !…

PQi�1 l

wki !

� l!α1!α2!…αwk

!�24�

ways, each with probability 1∕wlk. So P�L;F l� is given by

P�L;F l� �1wl

k

l!α1!α2!…αwk

!: (25)

The bit-error probability Pbe�k��L� given interference pat-tern L is evaluated as follows:

Pbe�k��L� �12Prfa bit errorjL interferers;1was sentg

� 12Prfa bit errorjL interferers;0was sentg

� 12Prfαi ≥ 1; ∀ i ∈ χkjL interferers;0was sentg:

(26)

The last equation can be rewritten as

Pbe�k��L�

� 12−12Prfαi � 0; some i ∈ χkjL interferers;0was sentg

� 12−12

�Xwk

m�1

12αm

−

Xwk−1

m�1

Xwk

n�m�1

12αm�αn

� � � � � �−1�wk−112l

�:

(27)

In the above relation, we have used the inclusion–exclusionproperty of the probability of a union of events. Substitut-ing in Eq. (23), we obtain the BER of a class k user Pe�k� for acorrelation receiver with a hard limiter as

Pe�k� �XF1K1

l1�0

…

XFkKk−1

lk�0

…

XFQKQ

lQ�0

×�FkKk − 1

lk

�plkkk�1 − pkk�FkKk−1−lk

×YQ

q�1q≠k

�FqKq

lq

�plq

kq�1 − pkq�FqKq−lq

×X

L∈Fl∶αi≥1∀ i∈χk

1wl

k

l!α1!α2!…αwk

!

×�12−12

�Xwk

m�1

12αm

−

Xwk−1

m�1

Xwk

n�m�1

12αm�αn

� � � � � �−1�wk−112l

��: (28)

B. Probability of Degradation

The OMCS-GMPLS network must guarantee the QoSrequested by each user, and at the same time maximizethe network capacity by maximizing the number of usersadmitted to the network. In an OMCS-GMPLS network,the performance is limited by MAI and the BER is a func-tion of the number of simultaneously active MCSPs. Con-sequently, the allowable number of simultaneous MCSPsmust be controlled according to themaximumBER for eachclass; otherwise, the desired QoS would be drastically af-fected. Capacity can be improved significantly by allowinggraceful QoS degradation. In this paper, we present a calladmission control (CAC) protocol that depends on degrad-ing the desired QoS (in terms of BER) of class j active usersby increasing the number of admitted users over thedegradation threshold Γ�j�

Th of class j, which representsthe maximum number of users (each with Fj parallel codes)that may be simultaneously active in each wavelength onthe network for a given BER threshold.

If ρ represents the probability that a connected MCSPtransmits data, then the probability density function(PDF) of the number of simultaneously active codes x inclass j in a wavelength is given by [23]

P�j�x �x� �

XC�j�m�x

P�xactive pathsjm connected paths�

× P�j�M �m connected paths�

�XC�j�m�x

�mx

�ρx�1 − ρ�m−xP�j�

M �m�; (29)

where P�j�M �·� denotes the PDF of the number of occupied

codes per wavelength in class j. It can be evaluated asfollows. When a connection request arrives, a wavelengthis randomly chosen and Fj of its unused codes is assigned


simultaneously to the arrived service-class j connection re-quest. Assuming that C�j� is a multiple of Fj, we get the con-ditional PDF of the number of occupied codes in a givenwavelength by modifying the hypergeometric distributionto account for simultaneously assigning Fj codes from thesame wavelength to each incoming connection request asfollows:

P�j�M �mjn� �

�C�j�m

�Pr∈R

QW−1i�1

�C�j�riFj

�Pn∈N

QWi�1

�C�j�niFj

� ; (30)

where n and m are the number of occupied codes in a fiberand a wavelength, respectively, and the two sets N and Rare defined as

N �def�n � �n1; n2;…; nW �∶

XWi�1

ni �nFj

and

ni ∈�0;1;…;

C�j�Fj

�∀ i ∈ f1;2;…;Wg

�;

R �def�r � �r1; r2;…; rW−1�∶

XW−1

i�1

ri �n −mFj

and

ri ∈�0; 1;…;

C�j�Fj

�∀ i ∈ f1; 2;…;W − 1g

�: (31)

Using the PDF of n, we have

P�j�M �m� �

XL�j�n�m

�C�j�m

�P∀ kisets

QW−1i�1

�C�j�kiFj

�P

∀ nisets

QWi�1

�C�j�niFj

� P�j�N �n�: (32)

In the WSCP policy, if at least the number of active codes inone service class exceeds FjΓ

�j�Th, the desired QoS is

degraded. In this case, P�j�deg is the same for all classes

and (after dropping the index j) is given as

Pdeg � Prfx�j� > FjΓ�j�Th; some j ∈ Ωg

� 1 − Prfx�j� ≤ FjΓ�j�Th; ∀ j ∈ Ωg: (33)

Considering the statistical independence among the differ-ent classes, the total probability of degradation is obtainedby using the probability of degradation of each class asfollows:

Pdeg � 1 −

YQj�1

�P�j�x �x�j� ≤ FjΓ

�j�Th��; (34)

where P�j�x �·� denotes the probability of getting the required

QoS for class j. Using Eq. (29) and assuming that ρ is thesame for all classes, we can write

P�j�x �x�j� ≤ FjΓ

�j�Th�

�XFjΓ

�j�Th

x�j��0

24 XC�j�

m�x�j�

� m

x�j�

�ρx�j��1 − ρ�m−x�j�P�j�

M �m�35: (35)

Hence, Pdeg can be computed as follows:

Pdeg � 1 −

YQj�1

0@XFjΓ

�j�Th

x�j��0

24 XC�j�

m�x�j�

� m

x�j�

�

×ρx�j��1 − ρ�m−x�j�P�j�M �m�

351A: (36)

C. Call Admission Control

With CAC, the network blocks some new connection re-quests in order to reduce interference on the network sothat the degradation probability decreases. However, thisimprovement in service availability comes at the cost ofincreasing the blocking probability.

For an OCDM-based OMCS-GMPLS network to be oper-ated with perfect service availability, so that Pdeg � 0 forany load, it follows that no more than ΓTh connectedMCSPs can ever be connected to the network. However,this scheme can severely limit the number of availablecodes, since it eliminates the capacity gained throughstatistical multiplexing. On the other hand, the capacityof the network can be greatly increased if degradationsare allowed to occur with some small probability. DefineΛTh as a blocking threshold that determines the upperbound of the number of active codes per wavelength. Thevalue of ΛTh is determined based on the activity coefficientsuch that Pdeg does not exceed a given degradation con-straint Pmax

deg , defined by the network operator. In otherwords, new connection requests for class j are blocked whenactive codes x�j� > W × Λ�j�

Th − Fj, and as before, degradation

occurs when x�j� > FjΓ�j�Th. Thus, the blocking and degrada-

tion probabilities for class j are given by

P�j�B �

XW×Λ�j�Th

x�j��W×Λ�j�Th−Fj�1

P�j�N �x�j��;

Pdeg � 1 −

YQj�1

0@XFjΓ

�j�Th

x�j��0

24 XΛ�j�

Th

m�x�j�

� m

x�j�

�

× ρx�j��1 − ρ�m−x�j�P�j�M �m�

351A; (37)

respectively. It should be noticed that the degradation be-comes more probable and blocking becomes less probableas the blocking threshold Λ�j�

Th increases. That is, there isa trade-off between the degradation and blocking proper-ties of the system.

D. System Throughput

We consider a synchronous random access system,where packet transmissions start at the beginning of timeslots, each of duration T. All packets are assumed to have afixed length of LB bits, which corresponds to the slot


duration T. When a class j user becomes active, it is as-signed Fj codewords as long as the total number of occupiedcodes is less than the maximum number of available code-words. The active class j user starts Fj simultaneous pack-ets transmission (with probability po) and enters athinking mode. For unsuccessfully received packets, theuser enters a backlogged mode and attempts to retransmitthese packets after a random time delay with average dtime slots (with probability pr � 1∕d) [28]. A data packetis successfully received at the destination node if it passesthrough all intermediate core nodes without blocking andall its bits are received without errors.

Let lk; k ∈ Ω denote the number of class k transmittingusers. Each user has Fk simultaneous packets per timeslot. Assuming that po � pr and lk is large enough, thecomposite arrival (new and retransmitted packets) distri-bution of a class k packet is Poisson with a finite arrivalrate γk packets/s. The steady-state probability distributionof class k composite arrivals is [29]

f �lk� ��Gk∕Fk�lk

lk!exp�−Gk∕Fk�; (38)

where Gk � γkT denotes the offered load of class k packets,which is the average number of generated packets fromclass k in one slot duration. The per node class k steady-state throughput is thus

βk �XQ

i�1;i≠k

X∞li�0

X∞lk�0

�lkFk ∧ WC�k��

× PSucc�k�YQi�1

�Gi∕Fi�lili!

exp�−Gi∕Fi�; (39)

where x ∧ y denotes the minimum of two numbers x and y,and PSucc�k� is the overall packet success probability of classk packet transmission, given by

PSucc�k� � �1 − P�k�B �PS�k�

��HW

��; (40)

where H � l1F1 ∧ WC�1� � � � � � lQFQ ∧ WC�Q�, d·e de-notes the ceiling function, and PS�k��K� represents thepacket success probability (the probability that all ofthe bits of the class k packet are correctly received atthe destination) given K codes. For the correlation receiver,PS�k��K� is given as

PS�k��K� �XF1K1

l1�0

…

XFkKk−1

lk�0

…

XFQKQ

lQ�0

× P�l interferersjK codes��Pbc�k��l��LB ; (41)

where P�l interferersjK codes� is given in Eq. (14) and theconditional bit-correct probability Pbc�k��l� is given as

Pbc�k��l� � Pbc�k��a bit successjl interferers;K codes�� 1 − Pbe�k��a bit errorjl interferers;K codes�

� 1 −12

1QQi�1 2

li

Xi1 ;i2;…;iQ∈Ψ

YQi�1

�li

ii

�: (42)

For the correlation receiver with a hard limiter, on theother hand, the packet success probability is given as

PS�k��K� �XF1K1

l1�0

…

XFkKk−1

lk�0

…

XFQKQ

lQ�0

× P�l interferersjK codes�×

XL∈F l

P�L;F l��Pbc�k��L��LB ; (43)

where the conditional bit-correct probability Pbc�k��L� isgiven as

Pbc�k��L� � 1 − Pbe�k��L� (44)

and Pbe�k��L� is given by Eq. (27).

IV. NUMERICAL RESULTS

In this section we investigate the performance of theOMCS-GMPLS network. Considering only two levels,namely, high and low levels, for data rate and QoS perfor-mancemetrics, there are 22 possible service classes. For thesake of simplicity, we will consider only two service classesdefined as follows:

1) Class 1: High rate and high QoS �BER ≤ 10−9�,2) Class 2: Low rate and low QoS �BER ≤ 10−6�,where class 1 is suitable for real-time video streamingand class 2 is appropriate for real-time voice transmissions.

For the sake of comparison, we use the previously pro-posed MLVW-OOC technique [23] for providing multiratemultiservice in the OCS-GMPLS network. For this net-work, the following parameters are adopted: L �f600;1200g, w � f7; 5g, D � f14∕42;28∕42g, F � f1;1g,and Q � 2, I � 1. In order to provide the same transmis-sion rate and QoS using our proposed MCVW-OOC tech-nique, the following parameters are adopted: L �f1200; 1200g, w � f7;5g, D � f14∕42;28∕42g, F � f2;1g,and Q � 2, I � 1. Taking W � 4, the number of the avail-able codewords in classes 1 and 2 are L�1� � W × C�1� � 56and L�2� � W × C�2� � 112, respectively.

A. Bit-Error Rate

The BER performance of the two-class system usingMLVW-OOC and MCVW-OOC is plotted in Fig. 5(a) as afunction of the number of class 1 (high-QoS) active codes.The number of class 2 active codes is fixed to 10 and 14 forcorrelation receivers without and with a hard limiter, re-spectively, for both MLVW-OOC and MCVW-OOC systems.In Fig. 5(b), the BER is plotted versus the number of class 2(low-QoS) active codes. For the case of the MLVW-OOCsystem, the number of class 1 active users is fixed to 4and 8 with F1 � 1 for correlation receivers without andwith a hard limiter, respectively. For the case of the


MCVW-OOC system, however, the number of class 1 usersis fixed to 3 and 7 with F1 � 2 for correlation receiverswithout and with a hard limiter, respectively. It can be seenfrom the figure that the proposed MCVW-OOC system out-performs MLVW-OOC in terms of BER. Indeed, in order tomeet the BER requirement (≤10−9 for class 1 users and≤10−6 for class 2 users), the degradation threshold of class1 and class 2 cannot exceed Γ�1�

Th � 4 and Γ�2�Th � 10 for the

correlation receiver, and Γ�1�Th � 8 and Γ�2�

Th � 14 for thecorrelation receiver with the hard limiter in the case ofthe MLVW-OOC system. On the other hand, for the caseof the MCVW-OOC system, in order to meet the sameBER requirements, the degradation threshold of class 1and class 2 cannot exceed Γ�1�

Th � 3 (6 codes) and Γ�2�Th � 10

for the correlation receiver, and Γ�1�Th � 7 (14 codes) and

Γ�2�Th � 14 for the correlation receiver with a hard limiter.

That is, it is clear that the number of supported high-QoS users is almost the same for both MCVW-OCC andMLVW-OOC systems using a correlation receiver with ahard limiter. In contrast, the number of available codesin the MCVW-OOC system that meet the BER require-ments for each class is larger than that in the MLVW-OOC system. This is due to decreased interference betweendifferent classes in the MCVW-OOC system.

B. Probability of Degradation

The Pdeg of both OCS-GMPLS and OMCS-GMPLS net-works, using correlation receivers with and without a hardlimiter, is plotted in Fig. 6 versus the offered load for

0 2 4 6 8 10 12 1410

−12

10−11

10−10

10−9

10−8

10−7

10−6

10−5

10−4

Number of simultaneous class 1 active codes F1K

1 (F

1=1 (SC), F

1=2 (MC))

Pro

babi

lity

of e

rror

Pe

MLVW−OOCMCVW−OOC

Correlation Receiver with HL

Correlation receiver

Pe(2)

Pe(1)

(a)

0 5 10 15 20 2510

−12

10−11

10−10

10−9

10−8

10−7

10−6

10−5

10−4

Number of simultaneous class 2 active codes F2K

2 (F

2=1 for SC and MC)

Pro

babi

lity

of e

rror

Pe

MLVW−OOCMCVW−OOC

Pe(2)

Pe(1)

Correlation Receiver with HL


(b)

Fig. 5. BER versus number of simultaneous active codes for MLVW-OOC and MCVW-OOC systems.

0 50 100 15010

−8

10−7

10−6

10−5

10−4

10−3

10−2

10−1

100

Offered load, A

Pro

babi

lity

of d

egra

datio

n, P

deg

Pdeg

for Correlation receiver with hard−limiter

ρ=0.3 SC

ρ=0.5 SC

ρ=0.8 SC

ρ=0.3 MC

ρ=0.5 MC

ρ=0.8 MC

(a)

0 10 20 30 40 50 60 7010

−6

10−5

10−4

10−3

10−2

10−1

100

Offered load, A

Pro

babi

lity

of d

egra

datio

n, P

deg

Pdeg

for Correlation receiver

ρ=0.3 SC

ρ=0.5 SC

ρ=0.8 SC

ρ=0.3 MC

ρ=0.5 MC

ρ=0.8 MC

(b)

Fig. 6. Pdeg as a function of offered load A for different activity coefficients ρ. (a) Correlation receiver with hard limiter. (b) Correlationreceiver.


different activity coefficients. In these plots we consider thevalues of Γ�i�

th obtained from Fig. 5. Since the OCDMA-basedGMPLS network is interference limited, when transmis-sion is very bursty �ρ → 0�, the total interference on the net-work decreases, causing the Pdeg to decrease and thenumber of available codes to increase. When the transmis-sion is less bursty �ρ → 1�, however, the Pdeg increasesowing to the increase of the number of simultaneouslyactive paths. Moreover, as the offered load increases, thenumber of connected paths increases, causing the degrada-tion Pdeg to increase. Comparing the two receiver models,one can observe the superiority of the correlation receiverwith a hard limiter. This is owing to the fact that the num-ber of available codes when using the correlation receiverwith a hard limiter is more than that when using the cor-relation receiver without a hard limiter. Furthermore, the

results show the superiority of the OMCS-GMPLS net-work. This is due to the fact that the available codesin OMCS-GMPLS systems are more than those inOCS-GMPLS systems due to reduced interference, whichresults in increasing the number of available codes andreducing Pdeg.

Pdeg is plotted in Fig. 7 versus the activity coefficient ρ atan offered load A � 50 for both MLVW-OOC and MCVW-OOC systems. This indicates how the performance canbe improved by limiting the number of connected paths.It can be seen from the figure that a decrease in blockingthreshold ΛTh leads to an improvement in performance.The results show that the MCVW-OOC system outper-forms the MLVW-OOC system since it can support morecodes for the same QoS. As an example, for the case ofthe lowest possible blocking threshold, Pdeg is the samefor both systems. However, the MLVW-OOC system cansupport 8 and 15 codes in class 1 (high-QoS) and class 2(low-QoS), respectively, while the MCVW-OOC systemcan support �7 × 2 � 14� and 15 codes in class 1 and class2, respectively. For the case of the largest possible blockingthreshold, both systems support the same number of codes(14 and 28 for class 1 and class 2, respectively); however,the performance of the MCVW-OOC system is better dueto the reduced interference. Moreover, for the MCVW-OOC system, the class 1 blocking threshold equals the deg-radation threshold �Λ�1�

Th � Γ�1�Th�, and, hence, the QoS degra-

dation occurs only if the number of active connectedMCSPs in class 2 exceeds its corresponding degradationthreshold Λ�2�

Th > Γ�2�Th.

C. Blocking Probability

The blocking probabilities PB of class 1 and class 2 usersare plotted in Fig. 8 versus the offered load A for bothMLVW-OOC and MCVW-OOC systems under differentblocking thresholds ΛTh. As can be seen from the figure,

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.910

−12

10−10

10−8

10−6

10−4

10−2

100

Activity coefficient, ρ

Pro

babi

lity

of d

egra

datio

n, P

deg

Pdeg

for Correlation receiver with hard−limiter

ΛTh(1)=8, Λ

Th(2)=15

ΛTh(1)=12,Λ

Th(2)=22

ΛTh(1)=14,Λ

Th(2)=28

ΛTh(1)=7, Λ

Th(2)=15

ΛTh(1)=7, Λ

Th(2)=22

ΛTh(1)=7, Λ

Th(2)=28

MLVW−OOC

MCVW−OOC

Fig. 7. Effect of limiting the number of connected paths on theperformance.

Fig. 8. Blocking probability versus offered load for different blocking thresholds. (a) For class 1 users. (b) For class 2 users.


higher offered loads can be supported for class 2 usersrather than class 1 users. This is clear as class 2 users havemore available codes. Furthermore, PB increases as Λ�i�

Th de-creases. This is because the decrease in Λ�i�

Th would reducethe number of available codes. In addition, it is clearfrom the figure that for class 1 users at BER ≤10−9, theMCVW-OOC system with Λ�1�

Th � 7 �7 × 2 � 14� experiencesa higher blocking probability than that of the MLVW-OOCsystem with Λ�1�

Th � 8. This is because the blocking inthe MCVW-OOC system occurs when the connection re-quest finds less than two free codes, while it occurs inthe MLVW-OOC system when there is no free code.On the other hand, for class 2 users, the blocking probabil-ity is the same for both MCVW-OOC and MLVW-OOCsystems since class 2 users in both systems require onlyone code.

D. System Throughput

The steady-state throughputs of both classes 1 and 2 areplotted as functions of the offered packet load in Figs. 9 and10, respectively. Both MLVW-OOC and MCVW-OOC sys-tems, using correlation receivers with and without a hardlimiter, are adopted. In order to meet the BER requirement(≤10−9 for class 1 users and ≤10−6 for class 2 users), we con-sider the values of Γ�i�

th obtained from Fig. 5 as the availablecodes in each class. For the case of no blocking (NB), as Aincreases above zero, more packets become available withlow interference. As a result, class 1 (2) throughput in-creases until it reaches a saturation value that is alwaysless than the number of class 1 (2) available codes. For largeenough A, the number of active users requesting free codesincreases until there are not enough codes and no more

0 20 40 60 80 100 1200

10

20

30

40

50

60

Offered load of class 1, G1 [packets/slot]

Thr

ough

put,

β 1 [pac

kets

/slo

t]Correlation receiver with hard−limiter

SC: ΓTh(1)=8, Γ

Th(2)=14

MC: ΓTh(1)=7, Γ

Th(2)=14

G2=0 (SC NB)

G2=120 (SC NB)

G2=0 (SC B)

G2=120 (SC B)

G2=0 (MC NB)

G2=120 (MC NB)

G2=0 (MC B)

G2=120 (MC B)

LB=127

(a)

0 20 40 60 80 100 1200

5

10

15

20

25


Thr

ough

put,

β 1 [pac

kets

/slo

t]


SC: ΓTh(1)=4, Γ

Th(2)=10

MC: ΓTh(1)=3, Γ

Th(2)=10

G2=0 (SC NB)

G2=120 (SC NB)

G2=0 (SC B)

G2=120 (SC B)

G2=0 (MC NB)

G2=120 (MC NB)

G2=0 (MC B)

G2=120 (MC B)

LB=127

(b)

Fig. 9. Class 1 throughput versus offered packet load. (a) Correlation receiver with hard limiter. (b) Correlation receiver.

0 20 40 60 80 100 1200

10

20

30

40

50

60


Thr

ough

put,

β 2 [pac

kets

/slo

t]

Correlation receiver with hard−limiter

SC: ΓTh(1)=8, Γ

Th(2)=14

MC: ΓTh(1)=7, Γ

Th(2)=14

G1=0 (SC NB)

G1=60 (SC NB)

G1=0 (SC B)

G1=60 (SC B)

G1=0 (MC NB)

G1=120 (MC NB)

G1=0 (MC B)

G1=120 (MC B)

LB=127

(a)

0 20 40 60 80 100 1200

5

10

15

20

25

30

35

40

45


Thr

ough

put,

β 2 [pac

kets

/slo

t]Correlation receiver

SC: ΓTh(1)=4, Γ

Th(2)=10

MC: ΓTh(1)=3, Γ

Th(2)=10

G1=0 (SC NB)

G1=60 (SC NB)

G1=0 (SC B)

G1=60 (SC B)

G1=0 (MC NB)

G1=120 (MC NB)

G1=0 (MC B)

G1=120 (MC B)

LB=127

(b)

Fig. 10. Class 2 throughput versus offered packet load. (a) Correlation receiver with hard limiter. (b) Correlation receiver.


users (no more interference as well) can transmit theirdata, and hence saturation occurs. However, at higherloads, the throughput starts to decrease, as more and morepackets fail to find free codes, especially for the MLVW-OOC system. When considering blocking (B) at a core node,it is seen that as A increases, the blocking increases, thuslimiting the throughput. From Fig. 9, class 1 throughputusing MCVW-OOCs is nearly twice that using MLVW-OOCs, and hence the number of supported high-QoS usersis almost equal to that supported using MLVW-OOCs.Moreover, MCVW-OOC can achieve better throughput athigher offered loads. This of course is due to the reducedMAI when using MCVW-OOCs. Furthermore, it is noticed(from these two figures) that when using the correlationreceiver with a hard limiter, the system throughputs ofthe two classes outperform those of the correlation receiverwithout a hard limiter. From Fig. 10, class 2 throughputsare equal for both MLVW-OOC and MCVW-OOC systems,as they support the same number of codes and have thesame blocking probability [see Fig. 8(b)] in both systems.Comparing the traffic throughput from both classes, class2 throughput outperforms that of class 1 for both MLVW-OOC and MCVW-OOC systems. This of course is due to thelarger number of codes supported by class 2.

V. CONCLUDING REMARKS

An OMCS-GMPLS network employing MCVW-OOCshas been proposed, and its performance has been analyzed.The OMCS-GMPLS network architecture and multicodeswitching mechanism have been described, and the block-ing probability has been obtained using a multiservice lossmodel. The BER, the probability of degradation, and thesteady-state throughput have been derived for both corre-lation receivers with and without hard limiters. In our der-ivation, MAI has been considered as the main performancelimiting factor. The performance of our OMCS-GMPLS net-work has been compared to the traditional OCS-GMPLSnetwork, which employs MLVW-OOCs. We can extractthe following concluding remarks.

1) The MCVW-OOC system outperforms the MLVW-OOCsystem in terms of BER. However, the number of sup-ported high-QoS users is almost the same for both sys-tems using a correlation receiver with a hard limiter.

2) The MCVW-OOC probability of degradation is less thanthat of the MLVW-OOC system, as the number of avail-able codes in the MCVW-OOC system that meets theBER requirements is larger than that in the MLVW-OOC system.

3) The MCVW-OOC system experiences a higher blockingprobability than that of the MLVW-OOC system.

4) Steady-state throughput results show that the MCVW-OOC system performs better than the MLVW-OOCsystem—specifically at higher offered loads.

ACKNOWLEDGMENTS

This work was financially supported by the EgyptianMinistry of Higher Education (MoHE).

REFERENCES

[1] S. J. B. Yoo, “Optical packet and burst switching technologiesfor the future photonic Internet,” J. Lightwave Technol.,vol. 24, no. 12, pp. 4468–4492, Dec. 2006.

[2] M. J. O’Mahoney, C. Politi, D. Klonidis, R. Nejabati, and D.Simeonidou, “Future optical networks,” J. LightwaveTechnol., vol. 24, no. 12, pp. 4684–4696, Dec. 2006.

[3] R. Van Caenegem, D. Colle, M. Pickavet, P. Demeester, K.Christodoulopoulos, K. Vlachos, E. Varvarigos, L.Stampoulidis, D. Roccato, and R. Vilar, “The design of anall-optical packet switching network,” IEEE Commun.Mag., vol. 45, no. 11, pp. 52–61, Nov. 2007.

[4] A. Banerjee, J. Drake, J. P. Lang, and B. Turner, “Generalizedmultiprotocol label switching: An overview of routing andmanagement enhancements,” IEEE Commun. Mag., vol. 39,no. 1, pp. 144–150, Jan. 2001.

[5] A. Banerjee, J. Drake, J. P. Lang, and B. Turner, “Generalizedmultiprotocol label switching: An overview of signalingenhancements and recovery techniques,” IEEE Commun.Mag., vol. 39, no. 7, pp. 144–151, July 2001.

[6] S. Huang, K. Baba, M. Murata, and K. Kitayama, “Variable-bandwidth optical paths: Comparison between optical code-labelled path and OCDMpath,” J. Lightwave Technol., vol. 24,no. 10, pp. 3563–3573, Oct. 2006.

[7] K. Kitayama, “Code division multiplexing lightwave net-works based upon optical code conversion,” IEEE J. Sel. AreasCommun., vol. 16, pp. 1309–1319, Sept. 1998.

[8] K. I. Kitayama and M. Murata, “Versatile optical code-basedMPLS for circuit, burst, and packet,” J. Lightwave Technol.,vol. 21, no. 11, pp. 2753–2764, Dec. 2003.

[9] Y. Wang and B. Li, “Optical code-labeled router based onOCDM,” J. Opt. Commun. Netw., vol. 2, no. 2, pp. 111–116,Feb. 2010.

[10] D. D. Sampson, G. J. Pendock, and R. A. Griffin, “Photoniccode division multiple-access communications,” Fiber Integr.Opt., vol. 16, no. 2, pp. 129–157, Mar. 1997.

[11] M. Jinno,H. Takara, B.Kozicki, Y. Tsukishima, Y. Sone, andS.Matsuoka, “Spectrum-efficient and scalable elastic opticalpath network: Architecture, benefits, and enabling technolo-gies,” IEEE Commun. Mag., vol. 47, no. 11, pp. 66–73,Nov. 2009.

[12] S. V. Maric, O. Moreno, and C. J. Corrada, “Multimediatransmission in fiber-optic LANs using optical CDMA,” J.Lightwave Technol., vol. 14, no. 10, pp. 2149–2153,Oct. 1996.

[13] W. C. Kwong and G.-C. Yang, “Multiple-length extendedcarrier-hopping prime codes for optical CDMA systemssupporting multirate multimedia services,” J. LightwaveTechnol., vol. 23, no. 11, pp. 3653–3662, Nov. 2005.

[14] E. Intay, H. M. H. Shalaby, P. Fortier, and L. A. Rusch, “Multi-rate optical fast frequency-hopping CDMA system usingpower control,” J. Lightwave Technol., vol. 20, no. 2,pp. 166–177, Feb. 2002.

[15] E. Narimanov, W. C. Kwong, G.-C. Yang, and P. R. Prucnal,“Shifted carrier-hopping prime codes for multicode keyingin wavelength-time O-CDMA,” IEEE Trans. Commun.,vol. 53, no. 12, pp. 2150–2156, Dec. 2005.

[16] C.-Y. Chang, G.-C. Yang, and W. C. Kwong, “Wavelength-timecodes with maximum cross-correlation function of two formulticode-keying optical CDMA,” J. Lightwave Technol.,vol. 24, no. 3, pp. 1093–1100, Mar. 2006.


[17] T. Ohtsuki, “Performance of multicode direct-detectionoptical CDMA systems,” in Proc. IEEE GLOBECOM, 1998,pp. 3227–3232.

[18] S. V. Maric and V. K. Lau, “Multirate fiber-optic CDMA: Sys-tem design and performance analysis,” J. Lightwave Technol.,vol. 16, no. 1, pp. 9–17, Jan. 1998.

[19] A. R. Forouzan, N.-K. Masoumeh, and N. Rezaee, “Frametime-hopping patterns in multirate optical CDMA networksusing conventional and multicode schemes,” IEEE Trans.Commun., vol. 53, no. 5, pp. 863–875, May 2005.

[20] J. S. Vardakas, I. D. Moscholios, M. D. Logothetis, and V. G.Stylianakis, “Performance analysis of OCDMA PONs sup-porting multi-rate bursty traffic,” IEEE Trans. Commun.,vol. 61, no. 8, pp. 3374–3384, Aug. 2013.

[21] G.-C. Yang and W. C. Kwong, “Performance comparison ofmultiwavelength CDMA and WDMA+OCDMA for fiber-optics networks,” IEEE Trans. Commun., vol. 45, no. 11,pp. 1426–1434, Nov. 1997.

[22] G.-C. Yang, “Variable-weight optical orthogonal codes forCDMA network with multiple performance requirements,”IEEE Trans. Commun., vol. 44, no. 1, pp. 47–55, Jan. 1996.

[23] H. Beyranvand and J. A. Salehi, “All-optical multi-servicepath switching in optical code switched GMPLS corenetworks,” J. Lightwave Technol., vol. 27, no. 12,pp. 2001–2012, June 2009.

[24] J. S. Kaufman, “Blocking in a shared resource environment,”IEEE Trans. Commun., vol. 29, no. 10, pp. 1474–1481,Oct. 1981.

[25] J. W. Roberts, “A service system with heterogeneous user re-quirements: Application to multi-service telecommunicationssystems,” in Proc. Performance of Data CommunicationsSystems and Their Applications, G. Pujolle, Ed. Holland:Amsterdam, 1981, pp. 423–431.

[26] J. A. Salehi and C. A. Brackett, “Code divisionmultiple-accesstechniques in optical fiber networks—Part II: Systems perfor-mance analysis,” IEEE Trans. Commun., vol. 37, pp. 834–842,Aug. 1989.

[27] M. Azizoglu, J. Salehi, and Y. Li, “Optical CDMAvia temporalcodes,” IEEE Trans. Commun., vol. 40, pp. 1162–1170,July 1992.

[28] H. M. Shalaby, “Optical CDMA random access protocols withand without pretransmission coordination,” J. LightwaveTechnol., vol. 21, no. 11, pp. 2455–2462, Nov. 2003.

[29] A. Sandouk, H. Okada, T. Yamazato, M. Katayama, and A.Ogawa, “Throughput improvement of a dual-class multicodeCDMA ALOHA system with modified channel load sensingprotocol,” in IEEE Int. Conf. Communication (ICC),June 1999, pp. 1079–1083.

AhmedE. Farghal received his B.S. andM.S. degrees in electricalengineering from the Faculty of Electronic Engineering, MenufiyaUniversity, Menouf 32952, Egypt, in 2006 and 2011, respectively.In 2006 he joined the Electronics and Electrical CommunicationsEngineering Department, Menufiya University, Egypt, and was

promoted to the position of Lecturer Assistant in 2011. He is cur-rently working toward a Ph.D. degree in electrical engineering atthe Graduate School of Engineering, Egypt-Japan University forScience and Technology (E-JUST), New Borg El-Arab City, Alexan-dria 21934, Egypt. His research interests include all-optical net-works (AONs), optical CDMA, elastic optical networks, andnano-optoelectronic devices.

Hossam M. H. Shalaby (S’83–M’91–SM’99) was born in Giza,Egypt, in 1961. He received B.S. andM.S. degrees from AlexandriaUniversity, Alexandria, Egypt, in 1983 and 1986, respectively, anda Ph.D. degree from the University of Maryland, College Park, MD,USA, in 1991, all in electrical engineering. In 1991, he joined theElectrical Engineering Department, Alexandria University, andwas promoted to a Professor in 2001. He is currently on leave fromAlexandria University, where he is the Acting Dean of the School ofElectronics, Communications, and Computer Engineering, Egypt-Japan University of Science and Technology (E-JUST), New BorgEl-Arab City, Alexandria, Egypt. From December 2000 to 2004, hewas an Adjunct Professor with the Faculty of Sciences and Engi-neering, Department of Electrical and Information Engineering,Laval University, Quebec, QC, Canada. From September 1996to February 2001, he was on leave from Alexandria University,when he was in the following places. From September 1996 toJanuary 1998, he was with the Electrical and Computer Engineer-ing Department, International Islamic University Malaysia, andfrom February 1998 to February 2001, he was with the Schoolof Electrical and Electronic Engineering, Nanyang TechnologicalUniversity, Singapore. He worked as a Consultant at SysDSoftcompany, Alexandria, Egypt, from 2007 to 2010. His researchinterests include optical communications, optical CDMA, opticalburst-switching, OFDM technology, and information theory.

Prof. Shalaby has served as a Student Branch Counselor atAlexandria University, the IEEE Alexandria and North DeltaSubsection, from 2002 to 2006, and served as a Chairman of thestudent activities committee of the IEEE Alexandria Subsectionfrom 1995 to 1996. He received an SRC Fellowship from 1987 to1991 from the Systems Research Center, Maryland; the StateExcellence Award in Engineering Sciences in 2007 from the Acad-emy of Scientific Research and Technology, Egypt; the ShomanPrize for Young Arab Researchers in 2002 from the Abdul HameedShoman Foundation, Amman, Jordan; the State Incentive Awardin Engineering Sciences in 1995 and 2001 from the Academy ofScientific Research and Technology, Egypt; the UniversityExcellence Award in 2009 from Alexandria University; and theUniversity Incentive Award in 1996 from Alexandria University.He is a Senior Member of both the IEEE Photonics Society andthe Optical Society (OSA).

Zen Kawasaki (M’72–SM’01) received Bachelor, Master, andPh.D. degrees from the Graduate School of Engineering, OsakaUniversity, Suita, Japan, in 1973, 1975, and 1978, respectively.He started his career as a Research Associate at Nagoya Univer-sity, Nagoya, Japan, in 1979. He moved to Osaka University in1989 to be a Lecturer, where he was promoted to Associate Profes-sor and Full Professor in the Department of Electrical, Electronicsand Information Engineering, Graduate School of Engineering, in1991 and 2000, respectively. He is a Fellow of the Institute of Elec-trical Engineering Japan. He is currently the President of theInternational Committee of Atmospheric Electricity. He is also aSenior Member of the IEEE.


Multirate Multiservice All-Optical Code Switched GMPLS ...hshalaby/pub/farghaljocnA.pdf · Switched GMPLS Core Network Utilizing Multicode Variable-Weight Optical Code-Division Multiplexing

Documents