Abstract: Basic switching elements areinterconnected in various architectures to form large photonic switches. To curtail cost and boost switchingperformance, the large photonic switch architecture should consider all issueswhich limit its maximum capacity.
These include the number of switchingelements on a single substrate, the total insertion loss, the number ofcrossovers, the power loss, the crosstalk, the worst-case SNR, and the internalblocking probability. This paper addresses these issues applied on an 8×8 space divisionmultistage photonic switch and discusses the tradeoffs involved. The performance of the switch compared withother well-known switches is also presented. Keywords: opticals switches; nonblocking; SNR; IL;waveguide crossovers 1.
INTRODUCTION Optical switches have demonstrated many advantagesover their electronic counterparts and therefore hold great promises in opticalnetworks which are being realized worldwide at an increasing rapidity tosatisfy the rushing demand for bandwidth.Optical switches, however, also announces newchallenges associated with the switching elements (SE) used to implement them.If the optical switch is based on guided-wave devices (like directionalcouplers), for example, dealing with the unique problem of crosstalk in the SEsand minimizing the number of waveguide crossovers are main concerns 1. Thesenew challenges add to the common defies of switching systems design whichinclude system architecture, blocking/nonblocking property, control complexity,unicast/multicast capability, number of SEs, number of drivers, systemcrossovers, attenuation, and signal to noise ratio (SNR). These importantdesign parameters are accentuated briefly in the following subsections. 1.
1 System Architecture Basic switching elements are interconnected invarious architectures to form large photonic switches. To curtail cost andboost switching performance, the large photonic switch architecture shouldconsider all issues which limit its maximum capacity. These include the numberof switching elements on a single substrate, the total insertion loss, thenumber of crossovers, the power loss, the crosstalk, the worst-case SNR, andthe internal blocking probability. Numerous architectures have been suggested foroptical switching 2-12.
These architectureshave been constructed using various optical SEs. Many of the classicalswitching architectures that found in the electronic and communication domainscould also be implemented with photonic SEs in the optical domain 1. Theseclassical architectures include crossbar, Clos, tree-type, Baseline, and Benes,to name a few.
1.2 Blocking/Nonblocking Property Internal blocking occurs in a given switchingarchitecture if no route can be found for a required connection even when thedestination output port is permitted. To have a nonblocking switching system,the internal blocking probability should be abridged to zero. There are threeconditions of nonblocking. If some or more existing connections in a switchingnetwork may need to be rearranged to allow the new connection to be added, theswitching network is called rearrangeably nonblocking 13.
In wide sensenonblocking switching network, an algorithm occurs for setting up the paths ina way which guarantees that any future connection can always be made withoutnecessitating rearrangement of current paths. A switching network is strictlynonblocking if any input to any free output connection can always be completedwithout rearrangement irrespective of the connecting algorithm used 13. 1.3 Control Complexity Complex connecting algorithms (i.e.
control)require more computational and arrangement time. The tendency for algorithmiccomplexity generally moves in the contrasting direction where no algorithms areneeded for strictly nonblocking architectures and major control andcomputational complexity are obligatory for the rearrangeably nonblockingarchitectures 14. There is a direct association between the complexity of therouting algorithm and the number of SEs used in constructing the switcharchitecture. 1.4 Unicast/Multicast Capability Switching system architectures can be classifiedinto two types; unicast and multicast. Unicast is the term used to describecommunication from one point/sender to only another one point/receiver.Multicast is the term used to describe communication from one or morepoints/senders to a set of other points/receivers 1, 5.
Multicastarchitectures typically have larger attenuations than unicast architecturesbecause the input optical power must be divided among several output channels.In addition to SEs, other devices such as splitters and combiners are neededfor constructing multicast architectures. 1.5 Number of SEs The total number of SEs (crosspoints) directlyreveals the system design cost. Different designs involve diverse numbers ofSEs for the same switch dimensionality. The number of stages crossed by thesignal (i.e.
the number of SEs on route) determines how the signal will beattenuated because of the loss inserted by each of the SEs in the path 15.Thus, the total number of SEs and the number of stages in a system should beminimized as possible. 1.
6 Number of Drivers Most of the architectures require an electronicdriver for each SE. Some of the architectures can, however, bond several SEs tothe same driver circuit (e.g.
the tree-type architectures) provided that groupsof SEs may always change their state coherently at the same time. The number ofdrivers is therefore less than or equal the number of SEs in any architecture.The number of drivers becomes a problem if the budget, power dissipation, orspace occupation related to each driver is massive 16.1.7 System Crossovers Considerable architecture necessitates the signalpaths to cross through one another on the optical substrate between the SEs inorder to set a particular topology. This crossover between waveguides inintegrated optics is more costly than its equivalent between two wires in VLSIelectronic chips. Although these passive integrated optical waveguidecrossovers appear practical, they can cause crosstalk, signal attenuation, andincrease the manufacturing complexity. Crossovers should therefore be minimizedwhen designing large photonic switching systems 17.
1.8 System Attenuation The number of SEs, cusps, stoops, substrates andfibers that a given signal path passes through governs the signal attenuation(or insertion loss) 18. Signal attenuation can be compensated for with theaddition of optical amplifiers (if the SE used is other than semiconductoroptical amplifier), repeaters or regenerators but this increases the systemcost and causes additional noise. It is for that reason better to retain thesignal attenuation to the least and keep it uniform for all possible signalpaths. 1.
9 System Signal-to-Noise Ratio Every SE and crossover that the signal path passesthrough may leak some optical power (noise or crosstalk) into the desired channel.This undesired power should be reduced. In other words, the worst-case SNRshould be as high as conceivable for a good bit-error-rate (BER) performance19. 2. PHOTONIC SWITCHING TECHNOLOGIES The presently operational optical switches have an electroniccentral level and scale to sizes in the order of 500×500 port counts with ahuge expense of electronic apparatus. They are protocol and bit rate dependentand establish a tailback for future communication systems. Photonic switcheswith no O/E signal conversion are transparent to all protocols, bit rates andchannel counts.
These switches use diverse switching principles and differentmaterials. This section will review some of the probable photonics technologiesthat could become important modules of future telecommunications systems. A photonic switch may function by mechanical means,by electro-optic effects, magneto-optic effects, or other applicablemethods. Slow optical switches, such as those using mechanically moving fibers,may be used for alternate routing of an opticalswitch transmission path, such as routing around a fault. Fastoptical switches, like electro absorption modulators, interferometric devicesemploying the electro optical effect, and semiconductor optical amplifiers(SOA) controlled by optical signals and integrated with discrete or integratedmicroelectronic circuits may be used to perform logic operations 20.Fast switches are relatively complex and unlikely to be employed in largerswitching fabrics. The following subsections gives an overview of themost prominent examples of photonic switches which do scale to larger matricesbut are usually slow (millisecond switching time). 2.
1 Thermo optic interference switch Thermo optic interference switches depend on theheat induced change of refractive index in the waveguide materials which inturn changes the spatial power distribution and phase relations between thesystem modes in coupled waveguide systems or multimode guides which allowswavelength dependent light switching between several output ports 2. TheMach-Zehnder-Interferometer (MZI) and the directional coupler are examples ofthis type. 2.2 Thermo optic digital switch The digitaloptical switches are based on a single system mode instead of interferencebetween several modes (in contrast to interferometric devices). Slowly varyingy-structures that allow for slow mode conversion to avoid power coupling intohigher order modes are the most common form of realization for these type 21. 2.3 MEMS switch Switchesbased on MEMS in silicon technology can be divided in two categories: twodimensional devices in a single plane consisting of arrayed mirrors that pop upinto beam paths to redirect the light in a cross bar architecture, and threedimensional devices which also consist of mirror arrays where the mirrors arecontinuously movable in two dimensions 22.
An added matter of concern forthis type is lifetime due to the involvement of moveable parts. 2.4 TIR switch (Total Internal Reflection) Similar to2D MEMS the light in TIR switches is reflected in its path. The mirror is asmall air or gas bubble created in a trench at waveguide crossings by microheating a fluid that is index matched to the waveguides. The bubble can also bedestroyed by micro heating and due to the index matching the light thentraverses each intersection as though no trench was there 21.
2.5 Liquid crystal switch Liquid crystal switches utilize polarization forswitching. A birefringent liquid crystal rotates the polarization governed by abias voltage. A passive beam displacer routes the signal to the output portsaccording to the polarization. By cascading two liquid crystals slow relaxationeffects can be overcome and switching speed is in the order of milliseconds.
Due to the nature of the switch the polarization at the input port must beexactly known. Matrix sizes of 8×8 have been demonstrated 24. 2.6 Acousto optic switch Depending on the frequency of a sound wavetraveling along a crystal, the light can be deflected at certain angles. Bycombining two crystals for the horizontal and the vertical direction, the lightbeams can be switched in free space. The switches are relatively fast(microseconds) but are narrow banded (20 nm) 25.
3. PROBLEM FORMULATION AND OBJECTIVES All proposed switcharchitectures in the literature try to improve one or more of the issuesdiscussed in section I at the expense of more hardware. Each issue may singleout a different architecture as being better than others. However, the overalloptimal architecture depends on the relative weighting a designer would assignto the various issues. Designing nonblocking optical switching systems with improvedcrosstalk, nonblocking, and attenuation properties is still a scope forresearch. The task is to develop new switch architectures and to examine thecompromises which can be prepared between the performance concerns. Thus, this paperis a theoretical study of a proposed 8×8 optical switching architecture thataddresses the nonblocking, SNR, and attenuation issues and investigates howthese design parameters can be traded for optical switching systems.
In thispaper a wide-sense nonblocking architecture is proposed for a modular 8x8photonic switch. The idea behind the proposed design is presented. Theperformance of the switch is analyzed, compared with other switches, anddiscussed for diverse possible SEs.
The paper is organizedas follows; Section II briefly introduces the present photonic switchingtechnologies. Section III formulates the research problem and sets the researchobjectives and scope. Section IV explains the architecture of the proposed modular8x8.
The performance of the designed switch is analyzed and discussed insection V. Section VI concludes the work. 4. THE PROPOSED SWITCH Two planar switches are shown in Fig. 1. One is a2x3 switch and the other is a 3×2 switch. Based on theorems stated in 13,both switches were proved to be nonblocking in the wide-sense. They will becalled 2W3 and 3W2 elements, respectively.
The two elements are shown in Fig. 1. The 2w3and 3W2 elements are used recursively to build the proposed modular 8x8switching architecture with the basic 2×2 SE representing the smallest possiblesub network. The analysis of the modular 8×8 switch can be made in thefollowing manner.
Each switch in the first stage of the network and of all sub networksis assumed to have two inputs and each switch in the last stage of the networkand of all sub networks is assumed to have two outputs. For this network andfor all sub networks, all input switches are 2W3 elements and all outputswitches are 3W2 elements. Connections between a given input switch and a givenoutput switch is made via three sub networks each consisting of 4×4 similararchitectures. The analysis is recursive and can be repeatedly applied on thesub networks themselves until we reach the middle stage where the basic 2×2 SErepresents the smallest possible sub network.
Since every 2W3 and 3W2 elementhas three stages itself, the 8W8 switch network has consequently an overallnumber of 13 stages (from S1 to S13). Themodular 8×8 switch is illustrated in Fig. 2. Fig.1. The basic elements, (a) the 2W3 element and (b) the 3W2 element Fig.2. The modular 8W8 switch Fig.
3. The number of SEs and crossovers alongdifferent connection paths in the 8W8 switch5. PERFORMANCEANALYSIS AND COMPARISON This section analyzesand compares the performance of the proposed switch with switchingarchitectures with the same size and nonblocking features. Hence, the 8x8wide-sense nonblocking switching networks chosen for comparison are, thecrossbar, the double crossbar, and the modified double crossbar.
For each ofthese architectures, we have reported six main features; the number of SEs, thenumber of drivers (Ds), the number of waveguide crossovers (COs), and the SNR.Two additional features; the insertion loss difference (? IL) and the SNRdifference (? SNR) were also reported to reflect the eccentricity between bestand worst cases. Firstly, we need to introduce key parametersspecific to the formulas used. The total insertion loss (IL) in dB along a pathdepends on the number of SEs crossed, the number of couplings required, and thenumber of crossovers encountered. If s is the loss of each SE, c is the losscaused by coupling and w is the loss experienced due to each crossover (COs)then: IL (dB) = s x SEs + c x couplings + w x COsX, which is used in the SNR formula, represents the inverse of theextinction ratio in dB of a single SE. Thus, if CC is the number ofcrosstalk-contributing SEs along a connection path, SNR is given by: SNR (dB)= ?X ? 10 log10 CCwhere ? = 1or ? = 2if the evaluated crosstalk is of the first or thesecond order, respectively. Only first order crosstalk is considered in SNRcalculations whenever applicable.
The values for X, s, c, and w highly rely onthe photonic switching technology used and can be collected from data sheets ofcommercially available components. However, in this paper the worst case valueswhich represent relatively poor performance are used. The characteristics ofthe considered architectures are calculated based on formulas given in 12.Typically, we will take X = 20 dB, s = 1dB, c = 2dB, and w = 0.45 dB. The characteristics of various 8×8 architectures aretabulated in Table 1. The IL and SNR values shown in the table for the 8W8switch are calculated for the worst-case path presented in Fig. 3 by bold redlines while the best-case values of IL and SNR are calculated for the pathpresented by bold green lines.
The difference between the best and the worst casevalues gives ? IL and ? SNR for each parameter respectively. Table 1. Characteristics of various 8×8 switches Switch No. of SEs No. of Drivers No. of Crossovers IL ? IL SNR ? SNR 8W8 69 49 16 20.2 11.2 13 7 CB 64 64 0 19 14 11.
2 8.5 DCB 128 128 56 38.2 25.2 31.
5* 0 MDCB 128 96 24 22.8 10.2 37* 3 * due to second order crosstalk Sincethe 8W8 switch is constructed from 10 2W3 elements at the left half, 10 3W2elements at the right half, and 9 2×2 elements in the middle, thus the totalcount of SEs for the switch is 69. The number of SEs needed to design each ofthe compared architectures is plotted in Fig. 4. The proposed switch shows thesecond best performance after the crossbar. It is worth noting that for N ? 8,the switches constructed based on the 8W8 design idea require less SEs thantheir counterpart CB switches did.
For example the 16W16 and 32W32 switchesrequire 255 and 861 SEs respectively while CB switches with similar dimensionsneed 256 and 1024 SEs respectively. Regarding the number of drivers requiredfor each switch, the 8W8 switch holds the best performance as shown in Fig. 5.For the number of COs, the 8W8 switch comes second after the CB as given in Fig.6. The insertion loss experienced along the worst-case connection path isplotted in Fig.
7. The crossbar reflects the best performance followed by the8W8 switch with less than 2 dB difference between them. The SNR performance isillustrated in Fig.
8. The MDCB switch shows the highest value of SNR succeededby the DCB switch. This is because both switches suffer only from second ordercrosstalk. The MDCB is the best since a proper control function is used forsetting up switches 11.
The performance pf the 8W8 switch is however betterthan the performance of the CB due to the characteristics of the 2W3 and 3W2elements used. Observing Fig. 3, first order crosstalk is encountered only in 5stages;S2 or S3, S5 or S6, S7, S8 or S9, and S11 or S12. Fig. 4. The number of SEs Fig.5.
The number of drivers Fig. 9 presents the differential attenuation amonginput-output connections. The DCB architecture has the worst behavior underthis aspect. The insertion loss difference for the CB increases rapidly withthe switch dimension and is almost equal to the insertion loss value. This isdue to the fact that the shortest path inside the matrix always crosses just oneSE at the corner and hence it is independent of the network dimension.
Rankingsecond after the MDCB switch, the 8W8 switch has an acceptable value ofdifferential loss.Fig. 6. The number of COs Fig. 7. The IL Fig.
8. The differential IL Fig. 9. The SNR Fig.
10. The differentialSNR Dissimilarity in the IL may impose extra stress onthe optical receiver amplitude-dynamic and consequently the use of large switcharchitectures become unrealizable. The 8W8 switch stands the third place compared toother switches based on the differential SNR performance as plotted in Fig.
10.The worst-case SNR route is when the signal crosses 9 SEs with 5 of themcontributing to first order crosstalk (i.e. one SE in each 2W3/3W2 elementcrossed plus the middle stage 2×2 SE) while the best-case SNR route is when thesignal crosses 5 SEs with only the middle stage 2×2 SE contributing to firstorder crosstalk. The DCB shows no difference in SNR calculations because allconnection pairs always cross a fixed number of 7 SEs. 6.
CONCLUSION Various parameters need be optimized when designingphotonic switching networks. These parameters include system architecture,blocking/nonblocking property, control complexity, unicast/multicastcapability, number of SEs, number of drivers, system crossovers, attenuation,and signal to noise ratio (SNR).system attenuation, number of crossovers, andsignal-to-noise ratio.
This paper addresses the considerations involved inselecting a specific architecture to be used as an optical space switch fabric.A wide-sense nonblocking modular 8×8 switch has been proposed. The performanceof this switch has been discussed and compared with three other competingdesigns. The proposed switch comes first in the number of drivers required,second in the number of SEs, number of COs, IL, and differential IL measures.It comes third regarding the SNR and differential SNR measures. Generally, noarchitecture can be selected as absolutely best.
Rather, the selection isapplication dependent and constrains imposed by the requirements of eachapplication should be addressed relatively for a successful selection. Theresults obtained reflect that the proposed design can probably be adopted inthe next generation optical cross connects specially for the selected switchsize.