|Views: 551|Likes: 1

Published by ijcsis

The Four Wave Mixing (FWM) crosstalk with equally spaced channels from each other is the dominant nonlinear effect in long haul, repeaterless, wavelength division multiplexing (WDM) lightwave fiber optical communication systems. To reduce FWM crosstalk in optical communication systems, unequally spaced channel allocation is used. One of the unequal bandwidth channel allocation technique is designed by using the concept of Golomb Ruler. It allows the gradual computation of a channel allocation set to result in an optimal point where degradation caused by inter–channel interference (ICI) and FWM is minimal. In this paper a new Soft Computing approach called Biogeography Based Optimization (BBO) for the generation and optimization of Golomb Ruler sequences is applied. It has been observed that BBO approach perform better than the two other existing classical methods i.e. Extended Quadratic Congruence (EQC) and Search Algorithm (SA).

The Four Wave Mixing (FWM) crosstalk with equally spaced channels from each other is the dominant nonlinear effect in long haul, repeaterless, wavelength division multiplexing (WDM) lightwave fiber optical communication systems. To reduce FWM crosstalk in optical communication systems, unequally spaced channel allocation is used. One of the unequal bandwidth channel allocation technique is designed by using the concept of Golomb Ruler. It allows the gradual computation of a channel allocation set to result in an optimal point where degradation caused by inter–channel interference (ICI) and FWM is minimal. In this paper a new Soft Computing approach called Biogeography Based Optimization (BBO) for the generation and optimization of Golomb Ruler sequences is applied. It has been observed that BBO approach perform better than the two other existing classical methods i.e. Extended Quadratic Congruence (EQC) and Search Algorithm (SA).

See More

See less

(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 5, May 2011

Golomb Ruler Sequences Optimization: A BBOApproach

Shonak Bansal

Department of Electronics& Communications,Maharishi MarkandeshwarUniversity, Mullana,Haryana, INDIAshonakk@gmail.com

Shakti Kumar

Computational IntelligenceLaboratoryInstitute of Science andTechnology, Klawad,Haryana, INDIAshaktik@gmail.com

Himanshu Sharma

Department of Electronics& Communications,Maharishi MarkandeshwarUniversity, Mullana,Haryana, INDIAhimanshu.zte@gmail.com

Parvinder Bhalla

Computational IntelligenceLaboratoryInstitute of Science andTechnology, Klawad,Haryana, INDIAparvinderbhalla@gmail.com

Abstract

—

The Four Wave Mixing (FWM) crosstalk withequally spaced channels from each other is the dominantnonlinear effect in long haul, repeaterless, wavelength divisionmultiplexing (WDM) lightwave fiber optical communicationsystems. To reduce FWM crosstalk in optical communicationsystems, unequally spaced channel allocation is used. One of theunequal bandwidth channel allocation technique is designed byusing the concept of Golomb Ruler. It allows the gradualcomputation of a channel allocation set to result in an optimalpoint where degradation caused by inter

–

channel interference(ICI) and FWM is minimal. In this paper a new SoftComputing approach called Biogeography Based Optimization(BBO) for the generation and optimization of Golomb Rulersequences is applied. It has been observed that BBO approachperform better than the two other existing classical methods i.e.Extended Quadratic Congruence (EQC) and Search Algorithm(SA).

Keywords

—

Four wave mixing, Optimal Golomb Ruler, SoftComputing, Biogeography Based Optimization.

I.

INTRODUCTIONIn conventional wavelength division multiplexing systems,channels are usually assigned with center frequencies (orwavelength) equally spaced from each other. Due to equalspacing among the channels there is very high probabilitythat noise signals (such as FWM signals) may fall into theWDM channels, resulting in severe crosstalk [1].FWM crosstalk is the main source of performancedegradation in all WDM systems. Performance can besubstantially improved if FWM generation at the channelfrequencies is avoided. It is therefore important to developalgorithms to allocate the channel frequencies in order tominimize the FWM effect. The efficiency of FWM dependson the channel spacing and fiber dispersion [2], [3]. If thefrequency separation of any two channels of a WDM systemis different from that of any other pair of channels, no FWMsignals will be generated at any of the channel frequencies.This suppresses FWM crosstalk [4]

–

[7]. Thus, the use of proper unequal channel spacing keeps FWM signals fromcoherently interfering with the desired signals.In order to reduce the FWM crosstalk effects in WDMsystems, several unequally spaced channel allocation(USCA) techniques have been studied in literature [1], [8]

–

[14]. An optimum

USCA (O

–

USCA) technique ensures thatno FWM signals will ever be generated at any of the channelfrequencies if the frequency separation of any two channels isdifferent from any other pair of channels in a minimumoperating bandwidth [11].Forghieri et al. [6

] treated the ―channel–allocation‖ design

as an integer linear programming (ILP) problem by dividingthe total available bandwidth into equal frequency slots. Butthe ILP problem was NP

–

complete and no general orefficient method was known to solve the problem. Sooptimum solutions (i.e., channel locations) were obtainedonly with an exhaustive computer search [1].However, the techniques [8]

–

[14] have the drawback of increased bandwidth requirement as compared to equallyspaced channel allocation. This is due to the constraint of theminimum channel spacing between each channel and that thedifference in the channel spacing between any two channelsis assigned to be distinct. As the number of channel increases,the bandwidth for the unequally spaced channel allocationmethods increases in proportion [4].This paper proposes a method for finding the solutions tochannel allocation problem by using the concept of OptimalGolomb Rulers (OGR) [7], [15]

–

[17]. This method forchannel allocation achieves reduction in FWM effect with theWDM systems without inducing additional cost in terms of bandwidth. This technique allows the gradual computation of a channel allocation set to result in an optimal point wheredegradation caused by inter

–

channel interference (ICI) andFWM is minimal [4], [16].Much effort has been made to compute short or dense RUsand to prove them optimal. Golomb Rulers represent a classof problems known as NP

–

complete [18]. Unlike thetraveling salesman problem (TSP), which may be classifiedas a

complete ordered set

, the Golomb Ruler may beclassified as an

incomplete ordered set

. The exhaustivesearch [19], [20] of such problems is impossible for higherorder models. As another mark is added to the ruler, the timerequired to search the permutations and to test the rulerbecomes exponentially greater. The success of SoftComputing approaches such as Genetic Algorithms (GAs)[21]

–

[23] in finding relatively good solutions to NP

–

complete problems provides a good starting point formethods of finding Optimal Golomb Ruler sequences. Hence,soft computing approaches seem to be very effectivesolutions for the NP

–

complete problems. No doubt, theseapproaches do not give the exact or best solutions butreasonably good solutions are available at given cost. In thispaper, a novel optimization algorithm based on the theory of biogeography of species called Biogeography Based

63http://sites.google.com/site/ijcsis/ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 5, May 2011

Optimization (BBO) is being applied to generate the optimalGolomb Ruler sequences for various marks.The remainder of this paper is organized as follows:Section II introduces the concept of Golomb Rulers. SectionIII presents the problem formulation. Section IV describes abrief introduction about BBO and steps to generate theGolomb Ruler sequences by using this soft computingapproach. Section V provides simulation results comparingwith conventional classical approaches of generating unequalchannel spacing i.e. Extended Quadratic Congruence (EQC)and Search Algorithm (SA). Section VI presents someconcluding remarks.II.

GOLOMB

RULERSThe idea

of ‗Golomb

R

ulers‘ was first introduced by W.C.

Babcock [7] in 1952, and further derived in 1977 from therelevant work by Professor Solomon W. Golomb [15], aprofessor of Mathematics and Electrical Engineering at theUniversity of Southern California. According to Colannino[24] and Dimitromanolakis [25], W. C. Babcock [7] firstdiscovered Golomb Rulers up to 10

–

marks, while analyzingpositioning of radio channels in the frequency spectrum. Heinvestigated inter

–

modulation distortion appearing inconsecutive radio bands and observed that when positioningeach pair of channels at a distinct distance, then third orderdistortion was eliminated and fifth order distortion waslessened greatly. According to William T. Rankin [26], all of

rulers‘ upto eight are optimum, the nine and ten mark rulers

that W. C. Babcock presents are near optimum.

The term ‗Golomb Ruler‘ refers to a set of non–

negativeintegers such that no distinct pairs of numbers from the sethave the same difference [27]. These numbers are referred toas marks [15], [21], [28] and correspond to positions on alinear scale. The difference between the values of any twomarks is called the distance between those marks. Thedifference between the largest and smallest number isreferred to as the length of the ruler. The number of marks ona ruler is sometimes referred to as the size of the ruler. Unlikeusual rulers, Golomb Rulers measure more discrete lengthsthan the number of marks they carry. Normally the first mark of the ruler [15], [16], [29] is set on position 0. Since thedifference between any two numbers is distinct, the newFWM frequencies generated would not fall into the onealready assigned for the carrier channels. Golomb Rulers arenot redundant as they do not measure the same distance twice[29].Figure 1 shows an example of Golomb Ruler. The distancebetween each pair of marks is also shown in the figure [21].

Figure 1. A Golomb Ruler with 4 Marks and Length 6

The particularity of Golomb Rulers is that all differencesbetween pairs of marks are unique [29], [30]. Although thedefinition of a Golomb Ruler does not place any restrictionon the length of the ruler, researchers are usually interested inrulers with minimum length.A perfect Golomb Ruler measures all the integer distancesfrom 0 to L, where L is the length of the ruler [18], [21], [22].In other words, the difference triangle of a perfect GolombRuler contains all numbers between one and the length of theruler. The length [31] of an n

–

mark perfect Golomb Ruleris

.For example, as shown in Figure 2 the set (0, 1, 3, 7) is anon optimal 4

–

mark Golomb Ruler since its differences are(1 = 1

–

0, 2 = 3

–

1, 3 = 3

–

0, 4 = 7

–

3, 6 = 7

–

1, 7 =7

–

0),all of which are distinct. As from the differences it is clearthat the number 5 is missing so it is not a perfect GolombRuler sequence.

Figure 2. A Non Optimal Golomb Ruler of 4

–

Marks and Length 7

However, the unique optimal Golomb 4

–

mark ruler is (0,1, 4, 6), which measures the distances (1, 2, 3, 4, 5, 6) (and istherefore also a perfect ruler) as shown in Figure 1.An Optimal Golomb Ruler is defined as the shortest lengthruler for a given number of marks [21], [32]. There can bemultiple different OGRs for a specific number of marks.The OGRs are used in a variety of real

–

worldapplications including Communications and RadioAstronomy, X

–

Ray Crystallography, Coding Theory, LinearArrays, Computer Communication Network, PPMCommunications, circuit layout, geographical mapping andSelf

–

Orthogonal Codes [7], [15], [21], [22], [26].An n

–

mark Golomb Ruler is a set of n distinctnonnegative integers

, called "marks," suchthat the positive differences

–

, computed over allpossible pairs of different integers

with

are distinct [20]. Let

be the largest integer in an n

–

mark Golomb Ruler [33]. Then an n

–

mark Golomb Ruler

is said to be optimal if and only if 1.

There exists no other n

–

mark Golomb Rulers havingsmaller largest mark a

n

, and2.

The ruler is written in canonical form as the "smaller"of the equivalent rulers

and

–

, where "smaller" means the firstdiffering entry is less than the corresponding entry inthe other ruler.In such a case,

is the called the

length

of the optimal n

–

mark ruler.Various classical methods are proposed in [1], [8]

–

[14] togenerate the OGRs. The soft computing methods that employgenetic algorithm (GA) based methods [21]

–

[23] could befound in literature. This paper proposes a new soft computingtechnique based on the mathematics of biogeography to

64http://sites.google.com/site/ijcsis/ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 5, May 2011

generate Golomb Ruler sequences, i.e., biogeography basedoptimization algorithm and its performance comparison withexisting classical methods that employ EQC and SA [1], [13],[21].III.

PROBLEM

FORMULATIONIf the spacing between any pair of channels is denoted as

and the total number of channels is

N

, then the objectiveis to minimize the length of the ruler denoted as

, which isgiven by the equation (1):

(1)subject to

where

with

are distinct.If each individual element is a Golomb Ruler, the sum of all elements of an individual forms the bandwidth of thechannels. Thus, if an individual element is denoted as

andthe total number of elements is

M

, then the second objectiveis to minimize the bandwidth (

), which is given by theequation (2):

(2)subject to

≠

where

with

are distinct.IV.

SOFT

COMPUTING

APPROACHIn this section, the capabilities of a new technique based onthe mathematics of biogeography called BBO for thegeneration of optimal Golomb Ruler sequences will bediscussed.

A.

Biogeography Based Optimization

Biogeography Based Optimization is a population

–

basedevolutionary algorithm (EA) developed for globaloptimization. It is based on the mathematics of biogeography.It is a new kind of optimization algorithm which is inspiredby the science of Biogeography. It mimics the migrationstrategy of animals to solve the problem of optimization [34]

–

[39]. Biogeography is the study of the geographicaldistribution of biological organisms. Biogeography theoryproposes that the number of species found on habitat ismainly determined by immigration and emigration.Immigration is the arrival of new species into a habitat, whileemigration is

the act of leaving one‘s native region.

Thescience of biogeography can be traced to the work of nineteenth century naturalists such as Alfred Wallace [40]and Charles Darwin [41].In BBO, problem solutions are represented as islands andthe sharing of features between solutions is represented asemigration and immigration. An island is any habitat that isgeographically isolated from other habitats [42].The idea of BBO was first presented by Dan Simon inDecember 2008 and is an example of how a natural processcan be modeled to solve general optimization problems [43].This is similar to what has occurred in the past few decadeswith Genetic Algorithms (GAs), Artificial Neural Networks(ANNs), Ant Colony Optimization (ACO), Particle SwarmOptimization (PSO), and other areas of computer

intelligence. Biogeography is nature‘s way of distributing

species, and is analogous to general problem solving.Suppose that there are some problems and that a certainnumber of candidate solutions are there. A good solution isanalogous to an island with a high HSI (Habitat suitabilityindex), and a poor solution is like an island with a low HSI.Features that correlate with HSI include factors such asdistance to the nearest neighboring habitat, climate, rainfall,plant and animal diversity, diversity of topographic features,land area, human activity, and temperature [39]. Thevariables that characterize habitability are called suitabilityindex variables (SIVs). High HSI solutions are more likely toshare their features with other solutions, and HSI solutionsare more likely to accept shared features from other solutions[43]

–

[45]. As with every other evolutionary algorithm, eachsolution might also have some probability of mutation,although mutation is not an essential feature of BBO theimprovement of solutions is obtained by perturbing thesolution after the migration operation [46].

1)

BBO Algorithm to Generate Optimal Golomb Ruler Sequences

The basic structure of BBO algorithm to generate OGRsequences is as follows:1.

Initialize the BBO parameters: maximum speciescount i.e. population size

S

max

, the maximummigration rates

E

and

I

, the maximum mutation rate

m

max

, an elitism parameter and the number of iterations.2.

Initialize the number of channels (or marks) ‗N‘ and

the upper bound on the length of the ruler.3.

Initialize a random set of habitats (integerpopulation), each habitat corresponding to apotential solution to the given problem. The numberof integers in each habitat being equal to the numberof channels or mark input by the user.4.

Check the golombness of each habitat. If it satisfiesthe conditions for Golomb Ruler sequence, retainthat habitat; if it does not, delete that particularhabitat from the population generated from the step3.5.

For each habitat, map the HSI (Total Bandwidth) tothe number of species

S,

the immigration rate

λ

, andthe emigration rate

μ

.6.

Probabilistically use immigration and emigration tomodify each non

–

elite habitat, then recompute eachHSI.7.

For each habitat, update the probability of its speciescount given by equation (3). Then, mutate eachnon

–

elite habitat based on its probability, check golombness of each habitat again and thenrecompute each HSI.

1 1,1 1 1 1,1 1,

s s s s ss s s s s s s ss s s s s

P PP P P PP P

(3)where

λ

s

and

μ

s

are the immigration and emigrationrates, when there are

S

species in the habitat.8.

Is acceptable solution found? If yes then go to Step10.9.

Number of iterations over? If no then go to Step 3for the next iteration.10.

Stop

maxmax

01 1

SS SS S

65http://sites.google.com/site/ijcsis/ISSN 1947-5500