Optical Cross Connects Architecturewith per-Node Add&Drop Functionality

Paolo Ghelfi(1), Filippo Cugini(1), Luca Potì(1), Antonella Bogoni(1), Piero Castoldi(2),Rodolfo Di Muro(3), Bimal Nayar(3)

(1) CNIT, Pisa, Italy; (2) Scuola Superiore Sant’Anna, Pisa, Italy;(3) Ericsson Ltd. - New Century Park, Coventry, CV3 1JG, UKpaolo.ghelfi@cnit.it

Abstract: We propose a cost-effective implementation of reconfigurable add&drop functionalityin Optical Cross Connects based on the concept of common add&drop in the node (OXC with per-Node Add&Drop, OXC-NAD). The proposed architecture constrains usable transponderswavelengths, but the analysis on significant mesh topologies demonstrates that this does not affectthe behavior of the transparent networks. The OXC-NAD is shown to also significantly reduce thenode cost irrespective of the nodal degree.©2006 Optical Society of AmericaOCIS codes: (060.4250) Networks; (060.4510) Optical communications

1. IntroductionThe continuous increase in traffic demand, the requirement for cost reduction, and the need of flexibility in networktopology are driving the development of transparent and reconfigurable optical mesh networks in core and metrosegments. In these networks Optical Cross Connect (OXC) nodes having the ability to optically switch the trafficto/from any direction with the granularity of the single WDM channel, are preferred. The recent development oftunable switches with wavelength granularity, namely the Wavelength Selective Switches (WSSs) [1],[2],integrating different functions in a single low-loss device, is allowing the realization of reconfigurable nodes withhigh nodal degree, thus permitting the design of the required OXCs for mesh networks.

Besides transparently switching the optical traffic, the OXCs are also needed to connect the local subnetworks tothe mesh network for adding and dropping channels to/from the WDM line streams crossing the node. In the OXCarchitectures proposed to-date [3]-[5], the Add&Drop (A&D) functionality is implemented at every bidirectionalport of the node (Optical Cross Connect with per-Port Add&Drop, OXC-PAD). This flexibility also requires eachbidirectional port to be able to add or drop all the channels assigned at the subnetworks. Therefore in the nodesbased on the OXC-PAD architecture the cost of the A&D functionality increases with the nodal degree.

In this paper we propose a cost effective approach for the implementation of the A&D functionality in theWSS-based OXCs. This approach is based on the concept of common Add&Drop in the node (OXC with per-NodeAdd&Drop, OXC-NAD), and considers the A&D channels as “local traffic” exchanged between the OXC and the“adjacent local node” constituted by the subnetworks. In this way, the A&D functionality is performed by anadditional bidirectional port only, therefore its cost is independent on the nodal degree. After describing the ordinaryOXC-PAD architecture and the proposed OXC-NAD scheme in detail, we compare their costs as a function of nodaldegree, demonstrating significant cost benefit of our proposed solution. We also show that the proposed OXC-NADarchitecture introduces a constraint on the A&D channels: a specific wavelength cannot be used by more than onesubnetwork of the same OXC to add and drop traffic in the transparent network. Finally, we demonstrate throughnumerical analysis that this constraint does not affect the behavior of the transparent network.

2. Optical Cross Connect with per-Port Add&Drop (OXC-PAD)Optical Cross Connects architectures reported in the literature can be schematized as shown in Fig. 1a [3]-[5]. Thenode comprises of a number of ports, each of which is connected to the neighboring OXC by means of an opticalfiber pairs for bidirectional communication. All the optical ports in the node are then connected together inside theOXC. Each optical port is able to demultiplex the WDM traffic coming from its adjacent node, and to switch it toother ports in the OXC. Similarly, it can collect all the channels directed towards the neighboring node, andmultiplex them in a single WDM stream. Two optical amplifiers are placed at the input and output fibers of everynode port to compensate for the losses induced in crossing the node, and for boosting the power of the WDM trafficbefore transmission.

Every OXC node is generally connected with one or more subnetworks which add and drop traffic in the WDMnetwork through a number of transponders. In the OXC structures proposed so far, the Add&Drop functionality is

a1636_1.pdf NTuC3.pdf

©OSA 1-55752-830-6

implemented directly at every bidirectional port by means of an A&D interface (Fig. 1a), therefore we can refer tothese architectures as OXCs with per-Port Add&Drop (OXC-PAD).

The implementation of a reliable and reconfigurable transparent network requires every subnetwork to use anyavailable WDM channel, therefore the transponders should be tunable. Moreover the subnetworks must have thepossibility to reconfigurably exchange traffic with the network through any of the optical ports in the node, in orderto ensure the network reliability in case of link failures. So the A&D traffic at every node port must be centrallycontrolled in the node by means of a Centralized Transponder Manager (CTM) [3], that can correctly connect anyadded or dropped channel with the desired subnetwork, irrespective of the used wavelength or node port. Moreover,in order to ensure the maximum reliability of the transparent network, the OXC should be able to add and drop allthe required channels from any port., Typically the maximum number of A&D channels at a node is about 20% ofthe total number of channels supported by a given deployment: thus in a 40-channel system every port of the OXC-PAD should be able to add and drop up to 8 channels.

Looking at this architecture in more details, the bidirectional ports exploits a splitter as a demultiplexer and aWSS as the multiplexer (Fig. 1b). This port scheme permits the correct functionality of the node with the best trade-off between cost effectiveness and channels impairments with respect to other port configurations [6], allowing alsothe broadcasting capability and the network protection (redundancy by path or link protection).

One of the output fibers of the demultiplexing splitter in the bidirectional port is devoted to forwarding thetraffic to the A&D interface, where the channels can be selected and dropped using a WSS. Similarly, an input fiberof the multiplexing WSS is used for adding the channels coming from the A&D interface, where a passive couplercollects the signals to be added coming from different tunable transmitters. (Fig. 1b).

All the added and dropped channels in every node port are connected to the Centralized Transponder Manager.The most cost effective and power saving implementation of the CTM proposed so far [3] is realized connecting allthe optical inputs and outputs of the A&D interfaces with optical switches, and connecting the switches to thetransponders (Fig. 1b). In this way any channel can always be sent to a specific transponder no matter which port itis dropped from, and no matter which wavelength has been assigned to the transponder.

3. Optical Cross Connect with per-Node Add&Drop (OXC-NAD)Fig. 2 shows the alternative proposed solution for the implementation of the Add&Drop functionality in the OXC.This architecture considers the A&D channels as “local traffic” that comes from - or is delivered to - an “adjacentlocal node”, which is constituted by the subnetworks connected to the OXC through the set of transponders. Theport which is devoted to the communication with this “local node” is the A&D port. In this way the Add&Dropfunctionality is implemented only once in the node, and this architecture can therefore be referred to as OXC withper-Node Add&Drop (OXC-NAD). The channels entering/exiting the A&D port are treated by the rest of the nodelike all the other WDM channels. The A&D port can then reconfigurably connect the node to the set of transpondersby means of one A&D interface, without the need of any CTM.

a)

Por

t2

Port 3

Port 1

OXC-PAD

A&Dinterface

A&Dinterface

CTM

A&

Dinterface

A&

Dinterface

A&

Dinterface

A&

Dinterface

subnetworkssubnetworks

b)

A&Dinterfaces

WSS ……

coup

ler

WSS ……

coup

ler

WSS ……

coup

ler

…

transponder

Centralized TransponderManeger (CTM)

WSS

WSS

WSS

Por

t1

split

ter

splitter

splitter

Port 2

Por

t3

OXC-PAD

subnetworksubnetwork

subnetworksubnetwork

…

transponder

Fig. 1 – a) Node architecture of an Optical Cross Connect with per-Port Add&Drop (OXC-PAD). Added and dropped channels must becentrally controlled by the Centralized Transponders Manager (CTM), which connects the Add&Drop interface of every port to thesubnetworks. Triangles on the transmitting and receiving fibers represent the optical amplifiers. Here the case of Nodal Degree (ND) equal to 3is considered. b) Detailed internal structure of the OXC-PAD architecture showing A&D interfaces and CTM.

a1636_1.pdf NTuC3.pdf

©OSA 1-55752-830-6

The port architecture for the proposed OXC-NAD unlike OXC-PAD has an additional port, which is dedicatedto the A&D operation and has the same structure as all the other ports. In this way, the total number of ports in theOXC-NAD is equal to the nodal degree plus one (one more than in the OXC-PAD structure). In the A&D interfacethe dropped channels are distributed to the correct receiver by means of a WSS, and the added channels are collectedwith a coupler.

4. Comparison between the OXC-NAD and OXC-PAD architecturesComparing Fig. 1 and Fig. 2 it is evident that the proposed OXC-NAD, besides being more compact than thesolution with per-port A&D, is also more cost-effective. The cost of the Add&Drop functionality in the OXC-PADin fact clearly increases with the nodal degree of the OXC. In more detail all the ports should be equipped with aWSS for dropping the channels, and a number of optical switches are needed for connecting the transponder to theA&D interfaces in the CTM. On the other hand, the proposed solution with per-node A&D dedicates an entire portto the A&D operation, but in this way the cost of the reconfigurable Add&Drop functionality is constant whatever isthe nodal degree.

Looking at the impairments induced on the optical channels in the OXC, the insertion loss experienced by thepass-trough traffic in both the cases is the same, and is compensated by the optical amplifiers placed at the input andoutput fiber ports. However, the attenuation induced on the added or dropped channels is different in the twoarchitectures: in the case of OXC-PAD, the small additional insertion loss due to the switches in the optical CTM(see Fig. 1b) can be easily accounted for slightly increasing the gain of the optical amplifiers at the ports input; inthe OXC-NAD architecture, the cascade of the A&D port and the A&D interface sensibly increases the attenuation,and two amplifiers must be inserted in the A&D interface to avoid spoiling the quality of the added and droppedsignals (see Fig. 2).

Fig. 3 shows cost comparison between the OXC-PAD and OXC-NAD architectures, as a functional of the nodaldegree ranging from ND=3 to ND=8. The number of transponders has been set to 8, and the use of wavelengthselective switches with 1x9 configuration has been considered. The total cost also includes the two optical amplifiers

WSS

WSS

WSS

A&DPort

WSS

Por

t1

….

….

split

ter

splitter

splitter

splitter

OXC-NAD

WSS

coupler

Port 2

Port3

A&DInterface

transponder

…

subnetworksubnetwork

subnetworksubnetwork

…

transponder

Fig. 2 – Structure of the proposed Optical Cross Connect with per-Node Add&Drop (OXC-NAD), in the case of ND=3. The A&D functionalityis carried out by an additional port, which communicate with the transponders and subnetworks through a single Add&Drop interface.

0

1

2

3

OX

CC

ost

s

3 4 5 6 7 8

Nodal Degree

OXC-NAD

OXC-PAD

Fig. 3 – Comparison between the cost of a node based on the proposed OXC-NAD architecture, and the cost of a node based on the OXC-PADscheme, varying the nodal degree from ND=3 to ND=8. The costs are calculated considering the structures reported in Fig. 2 and in Fig. 1brespectively, and including manufacturing costs. Costs are then normalized to the cost of the OXC-NAD with ND=3.

a1636_1.pdf NTuC3.pdf

©OSA 1-55752-830-6

at the input and output fibers of every port. Dispersion compensating modules have also been taken into account inthe cost analysis. These are accommodated in the mid-stage of two stage amplifiers at the node input ports. In thecase of OXC-NAD two amplifiers have been considered between the A&D port and the A&D interface tocompensate for the losses experienced by added and dropped channels. The manufacturing costs have been takeninto account too. The values are normalized to the cost of the OXC-NAD node with ND=3.

Fig. 3 also clearly shows that the OXC-NAD solution is more cost effective than the OXC-PAD for any NDvalue, even if the solution with per-port A&D requires an added port and two additional optical amplifiers. TheOXC-PAD is 21% more expensive with ND=3, and the cost difference increases with the nodal degree: at ND=8 theOXC-PAD costs 35% more than the OXC-NAD solution.

5. Network behaviorThe OXC-NAD architecture described in the previous paragraphs is clearly more cost effective than the OXC withper-port A&D. But it introduces a constraint on the wavelengths which the transponders can use. In case two ormore channels at the same wavelength are dropped at the same node from more than one port, then the multiplexingWSS at the A&D port would only deliver one of them to the A&D interface, whilst the others would be rejected.Similarly, it is not possible to add two or more channels at the same wavelength from separate transponders, sincethey would interfere in the multiplexing coupler of the A&D interface. In other words, in the OXC-NADarchitecture, a specific wavelength cannot be used by more than one subnetwork to add and drop traffic in thetransparent network. This A&D wavelength limitation is analyzed in the following in terms of network resourceallocation.

The performance of different networks based on both OXC-PAD and OXC-NAD architectures has beencompared using the Integer Linear Programming (ILP) model [7]. ILP flow-based formulations are utilized to solvethe Routing and Wavelength Assignment (RWA) problem. In particular, the problem of allocating all the lightpathconnections of a given traffic matrix using the least number of wavelengths on the most loaded link, namely theMin-RWA problem [7], has been considered. Wavelength continuity constraint has been applied (i.e., no wavelengthconversion has been considered in the network). Flow constraints have been introduced in the ILP to consider only asuitable set of routes, i.e., only the shortest paths has been considered in this analysis. In the case of multiple equalcost paths, the objective function guarantees the minimization of the maximum number of utilized wavelengths onthe most loaded link. A further flow constraint has been introduced in the case of OXC-NAD architectures to takeinto account the A&D wavelength limitation described above.

The performance comparison between OXC-PAD and OXC-NAD architectures is based on the evaluation ofthe amount of utilized wavelengths per fiber link, and on the total amount of required wavelengths in the network.The maximum number of utilized wavelengths per link represents the maximum number of channels effectivelyprovisioned on the most loaded link. The total number of required wavelengths in the network represents the numberof different required “colors” to accommodate the given traffic matrix. The maximum number of utilizedwavelengths per link corresponds also to the lower bound of the total number of required wavelengths.

The worst case scenario for the evaluation of the A&D wavelength limitation is represented by a N-nodes fullmesh network topology, accommodating a traffic matrix requiring one wavelength between each pair of nodes. Infact in this scenario both the networks realized with OXC-PAD and OXC-NAD architectures utilize just onewavelength to directly connect any node pairs; but in the case of OXC-PAD (i.e., without any constraint on theA&D wavelengths) the total amount of required wavelengths is equal to one (the same color in all links), whileusing the OXC-NAD architecture (i.e., in presence of the A&D wavelength limitation) the number of required

1 2 3

4 5 6

7 8 9

1 2 311 22 33

4 5 644 55 66

7 8 977 88 99

1 2 4

3 5 6

8 9 10

12 13 15

7

11

14

16

1 2 411 22 44

3 5 633 55 66

8 9 1088 99 1010

12 13 151212 1313 1515

77

1111

1414

1616

1

2

3

4

56

7

8

911

22

33

44

5566

77

88

99

12 3

54

76

89

10

1211

14

13

1516

17

12 3

54

76

89

10

1211

14

13

1516

17A B C D

1 2 3

4 5 6

7 8 9

1 2 311 22 33

4 5 644 55 66

7 8 977 88 99

1 2 4

3 5 6

8 9 10

12 13 15

7

11

14

16

1 2 411 22 44

3 5 633 55 66

8 9 1088 99 1010

12 13 151212 1313 1515

77

1111

1414

1616

1

2

3

4

56

7

8

911

22

33

44

5566

77

88

99

12 3

54

76

89

10

1211

14

13

1516

17

12 3

54

76

89

10

1211

14

13

1516

17

1 2 3

4 5 6

7 8 9

1 2 311 22 33

4 5 644 55 66

7 8 977 88 99

1 2 4

3 5 6

8 9 10

12 13 15

7

11

14

16

1 2 411 22 44

3 5 633 55 66

8 9 1088 99 1010

12 13 151212 1313 1515

77

1111

1414

1616

1

2

3

4

56

7

8

911

22

33

44

5566

77

88

99

12 3

54

76

89

10

1211

14

13

1516

17

12 3

54

76

89

10

1211

14

13

1516

17A B C D

Fig. 4 – Network topologies considered for the analysis of the resource allocation, in case of presence or absence of the A&D wavelengthlimitation. A) 9-node ring; B) 3x3 Manhattan; C) 4x4 Manhattan; D) Central-Northern Italy. In each case a regular traffic matrix has beenconsidered, requiring a wavelength between every node pairs. An additional scenario has been taken into account, using the topology D with arealistic traffic matrix.

a1636_1.pdf NTuC3.pdf

©OSA 1-55752-830-6

wavelengths is equal to N-1, even if just one wavelength is effectively utilized in every link. However full meshtopologies are practically not feasible, and we will not treat this particular case any longer.

Instead, we consider the four backbone networks shown in Fig. 4. They have been selected since they cover thecases of both regular and irregular topologies, with various numbers of links and nodes, and different average nodaldegrees. The considered topologies also take into account that realistic backbone networks must have a limitednumber of nodes in order to allow the transparent operation and to avoid expensive electronic regenerators.

For each network scenario a traffic matrix requiring one bidirectional lightpath between every node pairs hasbeen considered. Moreover an additional fifth scenario has been evaluated considering the topology D in Fig. 4(Central-Northern Italy) with a realistic traffic matrix. The main characteristics of the five scenarios are shown inTable 1.

The ILP computation results, in terms of the amount of utilized and required wavelengths considering bothOXC-PAD and OXC-NAD architectures, are shown in Table 2. In particular, the maximum number of utilizedwavelengths is exactly the same for both the node schemes. Therefore the A&D wavelength limitation occurringusing the proposed OXC-NAD architecture does not affect the overall amount of required network resources. As atypical example of wavelength utilization, Fig. 5 shows the utilized wavelength per link in the 4th scenario. It shouldbe noted that the maximum number of utilized wavelength is imposed by few links only, while most of the otherconnections are much less loaded.

Table 2 also reports the total number of required wavelengths, graphically reported in Fig. 6. As expected, theresults show that the two OXC architectures perform differently. In particular, OXC-PAD is able to accommodatethe input traffic matrix requiring a total number of wavelength that is exactly equal to the maximum number ofutilized wavelengths per link (i.e., the lower bound for the required wavelengths), whereas OXC-NAD needs ahigher amount of required wavelengths. However, for both architectures the maximum number of requiredwavelengths never exceeds the value of 20, which is significantly lower than the typical limit of 40 availablewavelengths that characterizes the ITU-T grid adopted in current optical networks. Moreover, the number ofadditional required wavelengths never exceeds the value of 4 in all considered scenarios. Other backbone topologieswith different traffic matrices have been evaluated and they all confirm the presented results. Therefore the A&Dwavelength limitation occurring with the OXC-NAD architecture does not introduce any significant performancedrawback in reasonable-sized backbone networks.

Table 1: Main network characteristics

Scenario Network n° of nodes n° of linkAvg. nodal

degreen° of lightpaths

1 A 9 9 2 72

2 B 9 12 2.66 72

3 C 16 24 3 240

4 D 17 35 4.11 272

5 D 17 35 4.11 188

Table 2: Max number of utilized and required wavelengths for the considered five scenarios

Max n° of utilized wavelengths Total n° of required wavelengthsScenario

OXC-PAD OXC-NAD OXC-PAD OXC-NAD

1 10 10 10 14

2 6 6 6 10

3 16 16 16 20

4 17 17 17 20

5 11 11 11 15

a1636_1.pdf NTuC3.pdf

©OSA 1-55752-830-6

6. ConclusionsWe have proposed a cost-effective architecture for the implementation of the Add&Drop functionality in OpticalCross Connects for transparent optical mesh networks. The proposed architecture is based on the basic idea ofcommon Add&Drop in the node (OXC with per-Node Add&Drop, OXC-NAD).

We have demonstrated that the use of our proposed architecture reduces the overall cost of the node with respectto the ordinary node architecture with per-Port A&D: the OXC-PAD turns out to cost 21% more than the new OXC-NAD at ND=3, and the difference increases up to 35% for ND=8. We have also shown that the use of the proposedOXC-NAD architecture introduces a constraint on usable transponders wavelengths. Nevertheless a numericalanalysis on different significant mesh topologies demonstrates that this constraint does not affect the behavior of thetransparent network in terms of resource allocation, resulting in the need of only few extra wavelengths along thenetwork.

The proposed OXC architecture with per-Node A&D thus appears to be a cost effective scheme for theimplementation of the A&D functionality in the transparent optical mesh networks.

References[1] D.M. Marom, D. T. Nielson, D. S. Greywall, C. S. Pai, N. R. Basavanhally, V. A. Aksyuk, D. O. Lopez, F. Pardo, M. E. Simon, Y. Low, P.

Kolodner, and C. A. Bolle, “Wavelength Selective 1xK Switches Using Free-Space Optics and MEMS Micromirrors: Theory, Design, andImplementation”, IEEE J. Lightwave Technol., v.23, n.4, pp. 1620-1630, 2005.

[2] J. Tsai, S. T. Huang, D. Hah, and M. C. Wu, “1xN2 Wavelength Selective Switch with Two Cross-Scanning One-Axis Micromirros Arraysina a 4-f Optical System”, IEEE J. Lightwave Technol., v.24, n.2, pp. 897-903, 2006.

[3] L. Zong, P. Ji, T. Wang, O. Matsuda, M. Cvijetic, “Study on Wavelength Cross-Connect Realized with Wavelength Selective Switches”,Proc. OFC 2006, NThC3, 2006.

[4] E. B. Basch, R. Egorov, S. Gringeri, and S. Elby, “Architectural Trade-Offs for Reconfigurable Dense Wavelength-Division MultiplexingSystems”, IEEE J. Sel. Topics Quantum Elec., vol 12, n. 4, pp 615-626, 2006

[5] C. R. Doerr, G. Wilfong, and S. Chandrasekhar, “Reducing the Complexity of Mesh Nodes by Using Reflective Wavelength SelectiveSwitches”, IEEE J. Sel.Topics Quantum Elec., vol 12, n. 4, pp 627-634, 2006.

[6] K. Ennser, T. Rogowski, P. Ghelfi, F. Cugini, P.Castoldi, “Reconfigurable Add/Drop Multiplexer Design to Implement Flexibility in OpticalNetworks”, in Proc. ICTON2006, 2006.

[7] D. Banerjee and B. Mukherjee, “A Practical Approach for Routing and Wavelength Assignment in Large Wavelength-Routed OpticalNetworks”, IEEE J. Sel. Ares in Communications, vol. 14, n. 5, June 1996.

0

4

8

12

16

20

Util

ized

wav

ele

ngth

s

1-2

1-6

1-8

2-3

2-6

2-7

2-8

6-3

6-7

6-8

8-12 3-4

3-7

3-9

7-4

7-9

7-12 4-5

4-9

9-5

9-1

09

-11

9-12

5-10

10-1

312

-11

12-1

711

-14

11-

1613

-14

13-1

514

-16

16-1

51

6-17

17-1

5

L ink id

Fig. 5 – Utilized wavelengths per link in the 4th considered scenario (topology D in Fig. 4, with a traffic matrix requiring one wavelength betweeneach pair of node). The links are identified based on the nodes they connect (e.g., link 1-2 connects node 1 and node 2).

0

4

8

12

16

20

Req

uire

dw

avel

engt

hs

1 2 3 4 5

Scenario

OXC-PAD

OXC-NAD

Fig. 6 – Total amount of required wavelengths in the considered five scenarios.

a1636_1.pdf NTuC3.pdf

©OSA 1-55752-830-6