Research Article A Novel Model on Curve Fitting and ...

Research ArticleA Novel Model on Curve Fitting and Particle SwarmOptimization for Vertical Handover in HeterogeneousWireless Networks

Shidrokh Goudarzi1 Wan Haslina Hassan1 Mohammad Hossein Anisi2

Seyed Ahmad Soleymani3 and Parvaneh Shabanzadeh4

1Communication System and Network (iKohza) Research Group Malaysia-Japan International Institute of Technology (MJIIT)Universiti Teknologi Malaysia Jalan Semarak 54100 Kuala Lumpur Malaysia2Department of Computer System amp Technology Faculty of Computer Science amp Information Technology University of Malaya50603 Kuala Lumpur Malaysia3Faculty of Computing Universiti Teknologi Malaysia UTM Johor Bahru 81310 Johor Malaysia4Centre for Artificial Intelligence and Robotics Universiti Teknologi Malaysia 54100 Kuala Lumpur Malaysia

Correspondence should be addressed to Mohammad Hossein Anisi anisiumedumy

Received 18 June 2015 Accepted 6 September 2015

Academic Editor Yihua Lan

Copyright copy 2015 Shidrokh Goudarzi et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

The vertical handover mechanism is an essential issue in the heterogeneous wireless environments where selection of an efficientnetwork that provides seamless connectivity involves complex scenarios This study uses two modules that utilize the particleswarm optimization (PSO) algorithm to predict and make an intelligent vertical handover decision In this paper we predict thereceived signal strength indicator parameter using the curve fitting based particle swarmoptimization (CF-PSO) and theRBFneuralnetworks The results of the proposed methodology compare the predictive capabilities in terms of coefficient determination (R2)and mean square error (MSE) based on the validation dataset The results show that the effect of the model based on the CF-PSOis better than that of the model based on the RBF neural network in predicting the received signal strength indicator situationIn addition we present a novel network selection algorithm to select the best candidate access point among the various accesstechnologies based on the PSO Simulation results indicate that using CF-PSO algorithm can decrease the number of unnecessaryhandovers and prevent the ldquoPing-Pongrdquo effect Moreover it is demonstrated that the multiobjective particle swarm optimizationbased method finds an optimal network selection in a heterogeneous wireless environment

1 Introduction

An efficient algorithm to determine the best network amongthe available ones is significant for wireless networks Themerits of each available network should be realized todiscover the best network Several multicriteria schemesbased on artificial intelligence methods such as fuzzy logicneural networks and genetic algorithms by [1 2] haveweaknesses on scalability and modularity issuesThey simplycannot manage based on the high numbers of radio accesstechnologies (RATs) and the criteria of the heterogeneouswireless networks (HWN) These algorithms have severalweaknesses in terms of scalability and complexity because

they enter all inputs from the different RATs to one fuzzy logicblock simultaneously rather than an exponential increasebased on the number of inference rules

The field of future wireless networks is one of themost attractive areas among researchers [3ndash6] The proposedalgorithms in this area of research are classified into differentgroups based on the analyses studies and tutorials foundin the related literature [3ndash6] These algorithms are classi-fied into different groups based on the expended decisiontechnique Reference [7] proposed a novel method for RATselection namely the hopfield neural network RAT selectionmechanism (HRM) that utilizes the hopfield neural networksas a strong decision-making tool A new approach using

Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2015 Article ID 620658 16 pageshttpdxdoiorg1011552015620658

2 Mathematical Problems in Engineering

information about data rate monetary cost and receivedsignal strength as different parameters to make a handoverdecision has been reported by [8 9] The main weaknessesare related to the computation of the error function andthe Jacobian inversion for acquiring a matrix in which thesizes are equal to the whole of all the weights in the neuralnetwork (NN) Therefore the necessity for memory is veryhigh Existing algorithms [8 9] consider the service fee thereceived signal strength information (RSSI) user preferenceand so forth The proposed algorithm in comparison withthe traditional RSSI-based algorithm enhances the outcomesmainly for both the user and the network as a result ofthe offered fuzzy based handover systems In terms ofhybrid categories [10] proposed a PSO-FNN-based verticalhandover decision scheme that could make an intelligenthandover decision based on the analysis of the networkrsquosposition The authors of [11] mainly dealt with a novelvertical handover decision algorithm built on fuzzy logicwith the assistance of the Grey theory and the dynamicweights adaptation A neurofuzzy multiparameter-based ver-tical handover decision algorithm (VHDA) was proposedby [12] The results of the performance evaluation carriedout by the handover quality indicator (used to quantifyQoS) which is related to the ldquoPing-Pongrdquo effect ESA andthroughput proved that the proposed VHDA offered betterQoS than the existing vertical handover methods Pahlavanet al [13] method is a good representation of applying afuzzy logic-based normalized quantitative decision algorithmand a differential prediction algorithm with a high level ofaccuracy

The vertical handover schemes stated above have theirown benefits but they do not consider the complexity ofthe network selection and the allowance of lower compu-tation cost function is unreasonable Clearly the decisionprocess should focus on a steadfast intelligent algorithm toexecute an accurate decision and to shift to the best networkcandidate quickly The goal of this study is to propose anovel network selection optimization algorithm that takesadvantage of the prediction model using the CF-PSO tomeet the requirements stated above The usage of the CF-PSO algorithm for prediction with the PSO algorithm forselection has three purposes (1) to serve as a validationalgorithm for the outcome of the MOPSO (2) to decreasethe number of unnecessary handovers and prevent the ldquoPing-Pongrdquo effect and (3) to select the best candidate accesspoint among various access technologiesThis paper providesa comparison based on two prediction methods namelythe CF-PSO and the RBF neural network to predict theRSSI More importantly this comparison evaluates the twomethods from different aspects for example time coefficientdetermination and mean squared error The results showthat the CF-PSO has better performance which will bepresented in-depth in the followingThe proposed method isa step towards future computer-based optimization methodswhere huge uncertainties by the optimization algorithmmustbe avoided To do this the prediction algorithm is com-bined with the particle swarm optimization The proposedMOPSO-based vertical handover decision algorithm canmake an intelligent handover decision based on the network

Set RSSI for APi

Prediction of RSSI by CS-PSO algorithm

QoS factors observation

Network selection by MO-PSO

Figure 1 Proposed network selection model based on the Mul-tiobjective Particle Swarm Optimization (MOPSO) in a wirelessheterogeneous environment

positionThe proposed network selection model is presentedin Figure 1

The outline of this paper is as follows First the relatedworks are described in Section 2 In Section 3 the proposedprediction methods based on CF-PSO and RBF networkare introduced Then the results of comparison based onthe proposed models are illustrated Section 4 illustratesthe MOPSO-based vertical handover decision algorithmSection 5 analyzes the performance of the algorithm throughthe simulation results Finally Section 6 concludes the paper

2 Related Work

Many network selection algorithms have been proposed inliterature in which PSO has been used for handover deci-sions It is initializedwith a set of randomparticles (solutions)that finds an optimal result by updating the generations Theoptimum results are called particles which fly throughoutthe problem space by following the current optimal particlesIn addition a GSM-like [14] hard handover algorithm hasbeen proposed where countermeasures prevent the Ping-Pong effect by providing a baseline handover thresholdagainst power fluctuations due to channel variations Theauthors defined four rules where two rules were for pre-venting unnecessary handovers (Ping-Pong effects) and onerule was aimed at helping the loaded eNB j by delayinghandovers from eNB i Moreover the other rule was aimed atalleviating the loaded eNB i by advancing handovers towardseNB j The PSO algorithm denoted that the neighbor withwhich the eNB has the largest handover traffic exchangewas the best neighbor Prior knowledge is required on theoptimization process that provided the parameterized formof the controller which was optimized by the MOPSO It hasbeen shown that the dynamic optimization outperformed thestatic optimization and produced a better mobility load bal-ancing self-organizing network controller which improvedthe throughput and access probability by a few percents withrespect to the planning solution

Wang et al [15] proposed an always best connected (ABC)maintained QoS handover decision scheme based on theniche PSO algorithm This literature reported the access

Mathematical Problems in Engineering 3

network in terms of its current load and considered theterminal in terms of its current velocity and residual electriccapacity The authors also studied the application of QoSrequirement preference of the user over the access networkcoding system and preference of the user over the accessnetwork provider and so forth

In addition Venkatachalaiah et al [16] presented atechnique for forecasting the signal strength value that aidsin offering efficient handovers in wireless networks and thePSO was expended to fine-tune the weighting function of thehandover decision The proposed technique minimized thenumber of handovers and was shown to have a very shortcalculation time and better prediction accuracy comparedto hysteresis based decisions In this paper they describedthe use of the Grey model in combination with the fuzzylogic and PSO algorithms Since prediction error is inevitablethe output from the Grey model could be compensated withthe use of a fuzzy controller and then fine-tuned using PSOalgorithms

The method by Liu and Jiang [11] is a good symbolicexample of using a fuzzy logic-based normalized quantitativedecision algorithm and a differential prediction algorithmwith a high level of accuracy This scheme tried to controlhandovers between WLANs and UMTS A predecision sec-tion was used in this algorithm The forward differentialprediction algorithm was used to acquire the predictiveRSS which could trigger a handover in advance Moreoverthe predecision method was applied before the handoverdecision module which is able to filter the unnecessary dataand to provide a correct handover trigger Optimizing theselection method is an important subject of research whichcan lead to the decline of network signaling andmobile devicepower loss and also boost network QoS and grade of service(GoS)

Themain focus of this paper is to achieve betterQoSwhileselecting the ABC network using a novel RSSI predictionalgorithm based on curve fitting which uses an enhancedPSO and a network selection algorithm based on MOPSOfor vertical handover in heterogeneous wireless networksThe network selection algorithm utilizes dynamic optimizedweights using the AHP method to select the ABC networkin the heterogeneous environment of the cellular networksDynamic weights of the parameters used in the cost functionfor network selection are optimized using an improvedMOPSO algorithm

3 Prediction of RSSI Using the ProposedPrediction Methods

In this section we intend to show that the CF-PSO has betterperformance compared to the RBF as a prediction modelReceived signal strength indicator or RSSI is a commonmetric used in handover decision-making [17] The qualityand distance for each access point (AP) in the range canbe analyzed by scanning the RSSI via a mobile node (MN)When the MN changes its location to an AP the level of RSSIfor that AP will change A real-tested measurement is usedto realize the RSSI level through the MN movement This

123456789101112131415

CICT minus20minus27minus30minus33minus38minus45minus50minus57minus57minus60minus66minus70minus80minus94minus98

L9 minus29minus36minus37minus37minus38minus40minus45minus50minus60minus72minus79minus80minus89minus94minus95


NSE2 minus98minus97minus96minus94minus93minus91minus90minus86minus79minus66minus52minus46minus39minus33minus26

Figure 2 Collected RSSI for four APs through the scanning phaseby the Chanalyzer software

scenario comprises one laptop as theMNwithin theAPs usedto offer the UMTS and WLAN services Data is collected viaa MetaGeek Wi-Spy spectrum analyzer and the Chanalyzersoftware The most necessary L2 information was collectedusing the Chanalyzer software as the information assistsin scheming the proposed prediction algorithm Figure 2displays the varying RSSI values during the MN movementduring 15 Sec The RSSI is gathered by the Chanalyzersoftware which is installed in one laptop moving across thefour APs

31 Prediction Using the Curve Fitting-Based Particle SwarmOptimization A prediction technique for vertical handoverusing the CF-PSO algorithm is presented in this sectionCurve fitting is based on the procedure of creating a curveor mathematical function which has the best fit to the datapoints subject to constraints In addition the PSO is a kindof swarm intelligence which utilizes a group of particles thatform a swarm moving across the search space observingthe best solution Particles are treated as a point in an 119873-dimensional space which modifies their ldquoflyingrdquo based ontheir own flying experience as well as the flying experienceof other particles Particles can change their positions basedon various types of information namely existing positionsrecent velocities the distance between the current positionand the personal best position (pbest) and the distancebetween the current position and the global best position(gbest) Particles can route their positions in the solutionspace which are related to the best solution (fitness) that hasbeen realized thus far by these particles This value is calledthe personal best pbest Another best value which is trackedby the PSO is the best value distance obtained subsequentlyby the particles in the region of these particles This item iscalled gbestThe basic concept of the PSO focuses on the fast-tracking particles based on their pbest and gbest locationseach time The searching concept of the PSO algorithm isshown in Figure 3 The change of the particlersquos position canbe mathematically formed based on the following

119881119898+1

119894= 120596 lowast 119881

119898

119894+ 1198881rand1times (pbest

119894minus 119904119898

119894) + 1198882rand2

times (gbest minus 119904119898

119894)

(1)


Vmi

Vm+1

i

Vgbe

st

i

Vpbesti

Xmi

Xm+1i

pbesti

gbesti

Figure 3 Searching concept of the PSO

where 119881119898

119894is velocity of particle i in iteration m 120596 is inertia

factor 119888119894is weighting factor rand is uniformly distributed

random number between 0 and 1 119904119898119894is current position of

particle 119894 in iteration119898 pbest119894is pbest of particle 119894 and gbest

is gbest of the groupA small inertia weight facilitates a local search and a large

inertia weight (120596) assists a global search Linearly decliningthe inertia weight from a large value to a small value throughthe course of the PSO run provides the best PSO performancein comparison to the stable inertia weight settings Inertiafactor (120596) is increased by the velocity of a particle at 119899thposition to achieve the changed velocity update as shownin (1) 120596 is initialized to 10 and is steadily decreased overtime The CF-PSO program was created and executed in theMATLAB

In the scanning stage the RSSI data for each AP will beentered as the input values to the CF-PSO to compute eachpoint of data (RSSI) The process is done by building upperand lower bounds covering each RSSI value point collectedfor the period of scanning time Thus the CF-PSO achievesthe next predicted RSSI value depending on how far the realRSSI point is from the curve of fitting This permits the MNto forecast the next RSSI value of available APs and its currentattached AP and then obtains a handover decision with a lowlatency

The RSSI data for each APwill be entered as the input val-ues to the CF-PSO to compute each point of data throughoutthe scanning step Creating higher and lower limits coveringeach RSSI value point collected during the scanning timecompletes the procedureTherefore the CF-PSO gets the nextpredicted RSSI rate depending on the distance of the realdata with the curve of fitting Through this method the MNcan predict the next RSSI value of the available APs and theexisting attached AP to the available AP Then it makes thebest handover decision with the lowest latency

In this study the polynomial model has been selected foruse in the curve fitting model to obtain the degree of fitnessfor each set of RSSI data that is related to a specific AP Thepolynomial model has enough flexibility for data with a highlevel of complexity such as the RSSI data In addition it islinearwhich shows that the fitting process applies in plain andeffective ways The CF-PSO works on collecting the RSSI foreach AP to find a close fit between the points of RSSI valuesand then obtains the expected curve between them In other

Input layer Hidden layer Output layer

Real Predictedvalue value

middotmiddotmiddot

Figure 4 RBF neural network structure

words the curve fitting usually works to get close values tomaintain the curve in the production region Equation (2)presents the Polynomial model in the CF-PSO

119865 (119909) = [11987511198834+ 11987521198833+ 11987531198832+ 11987541198831+ 1198755] (2)

Further details about the RSSI prediction using the CF-PSOare presented in Section 6

32 Prediction Using the RBF Neural Network Method Theartificial neural network (ANN) is used to develop optimizeestimate predict and monitor complicated systems A newand effective feed forward neural network with three layerscalled the RBF neural network has fine characteristics ofapproximation performance and global optimum [18] Ingeneral the RBF network consists of the input layer thehidden layer and the output layer Each neuron in the inputlayer is responsible for transferring the recorded signal to thehidden layer In the hidden layer the radial basis functionis often used as the transfer function while a simple linearfunction is usually adopted in the output layer The RBFprogram was implemented in the MATLAB The RBF neuralnetwork with three layers as used in the paper is displayed inFigure 4

33 Comparative Analysis After the scanning step the RSSIdata of each AP will be inserted into the CF-PSO as the maininputs for evaluating the best prediction achieved from theRSSI variations throughout the time of the MNrsquos movementHence the MN will acquire an exact method that firstlyanalyzes and then expects the next incoming RSSI for APsin the scanning region

In order to assess the performance of fit in the CF-PSO as shown in Figure 5 the residual analysis has beenchanged and used This is to justify the ways in which theCF-PSO can predict new RSSI values with a great degreeof certainty resulting from extremely variable RSSI datacollected from the APs Three statistical estimators namelythe mean squared error (MSE) in (3) the coefficient ofdetermination (1198772) in (4) and the root mean square error(RMSE) in (5) where if the RMSE is zero then the methodhas outstanding performance were used to evaluate theperformance of the CF-PSO


minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP1

(dBm

)

Time

RSSI AP1 versus timeFitted curve AP1

(a) The prediction bound for AP1 RSSI

minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP2

(dBm

)

Time


(b) The prediction bound for AP2 RSSI

minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP3

(dBm

)

Time


(c) The prediction bound for AP3 RSSI

minus120

minus100

minus80

minus60

minus40

minus20

0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

RSSI

AP4

(dBm

)

Time


(d) The prediction bound for AP4 RSSI

Figure 5 Process of CF-PSO prediction for RSSI of AP1 AP2 AP3 and AP4

MSE =1

119903sum119903

119894=1(V119901119894

minus V119886119894)2 (3)

1198772= 1 minus

sum119903

119894=1(V119901119894

minus V119886119894)2

sum119903

119894=1(V119901119894

minus V119886V)2

(4)

RMSE = radic

1

119903sum119903

119894=1(V119901119894

minus V119886119894)2 (5)

where 119903 is the number of points V119901119894is the estimated value V

119886119894

is the actual value and Vav is the average of the actual valuesThe coefficient of determination 1198772 of the linear regressionline between the estimated values of the neural networkmodel and the required output was also used as a measure ofperformance [19]The closer the1198772 value is to 1 the better themodel fits to the actual data [20]Thismeasurement processeshow successful the fit is in describing the change in the dataIn other words 119877-square is the square of the correlationbetween the response values and the predicted responsevalues It is also named the square of the multiple correlationcoefficients and the coefficient of multiple determinations1198772rsquos for AP1 AP2 AP3 and AP4 are 09973 09818 099278

and 09938 respectively The correlation coefficient squared(1198772) in four sets of calibration curves ranges from 098 to

100 indicating an almost perfect linearity of all the curvesFigure 6 shows further details of AP1 AP2 AP3 and AP4

The artificial neural network with RBF was used toestimate the RSSI data In this section the explanation ofthe experiment using the RBF neural network is presentedThe RSSI data will be inserted as inputs into the RBF inorder to examine the best prediction using this method Therelated results for the four APs are stated in Figure 7 Insummary the RBF network consists of three layers includingthe input layer the hidden layer and the output layer TheANN executes a nominal computation to offer an outputComputation comprises the one-pass arithmetic steps Theiterative and nonlinear computations are not complicated inoffering an output The RBF networks were chosen becausethismethod involves a simple design that has just three layersIn this study the number of neurons in the hidden layer is setto 15 the mean squared error (MSE) is 01 according to theactual training process and the 120590 (sigma) parameter is thewidth of RBF by 002

The main advantage is that the RBF has a hidden layerthat includes nodes known as the RBF units Each RBF hasmain factors that designate the location deviation or widthof the functionrsquos center The hidden component processes thedistance from the input data vector and the center of its RBF Ifthe distance from the specific center to the input data vectoris zero then the RBF has its own peak and if the distanceincreases then the peak of RBF will decline steadily


minus20minus27

minus30minus33

minus38minus45

minus50minus57minus57

minus60minus66

minus70

minus80

minus94minus98

minus140

minus120

minus100

minus80

minus60

minus40

minus20

00 5 10 15 20

minus29minus36minus37minus37minus38

minus40minus45

minus50

minus60

minus72

minus79minus80


minus100

minus90

minus80

minus70

minus60

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20

minus27minus31

minus40

minus60

minus68



minus82minus76

minus68

minus56

minus100

minus90

minus80

minus70

minus60

minus98minus97minus96minus94minus93minus91minus90minus86

minus79

minus66

minus52minus46

minus39minus33

minus26

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20 0 5 10 15 20

minus120

minus100

minus80

minus60

minus40

minus20

0

y = minus00037x4 + 00868x3 minus 06516x2

minus

y = minus00116x4 + 03419x3 minus 27611x2

y = minus00079x4 + 02758x3 minus 22402x2

minus

29826x minus 17137

44995x minus 17932

R2 = 099107

R2 = 09937

R2 = 098325

R2 = 0991

y = 00097x4 minus 026

minus 7316

82x3 + 20218x2

2x minus 2529

+ 88404x minus 10505

Figure 6 1198772 for AP1 AP2 AP3 and AP4

In RBF the hidden layer has different sets of weightsthat are divided into two sets These weights can connect thehidden layer to the input and output layers as linkages Thesubjects of the basis functions are fixed into the weights thatare connected to the input layer The issues of the networkoutputs are fixed into the weights that connect the hiddenlayer to the output layer Since the hidden units are nonlinearthe outputs of the hidden layer can be merged linearly andconsequently the processing is fastTheoutput of the networkis retrieved using the following [21]

119910119896(119909) =

119873

sum

119895=1

1199081198961198950119895(119909) + 119908

1198960 (6)

where 119873 is the number of basis functions 119908119896119895

representsa weighted connection between the basis function and theoutput layer 119909 is the input data vector and 0

119895is the nonlinear

function of unit 119895 which is typically a Gaussian form asshown in the following [21]

0119895(119909) = exp(minus

1003817100381710038171003817119909 minus 1205831003817100381710038171003817

2

21205902

119895

) (7)

where 119909 and 120583 are the input and the center of the RBFunit respectively and 120590

119895is the spread of the Gaussian based

function [21] The weights can be optimized by the leastmean square (LMS) algorithm once the centers of the RBFunits are determined The centers are selected randomlyor by clustering the algorithms In this paper the centerswere nominated from the dataset randomly The radial basisartificial neural network model was trained to minimize theMSE with the parameter (RSSI) as input and the desiredoutput (predicted RSSI) To design and verify the reliabilityof the proposed model the dataset was divided into twodifferent sets including the training and test data that are80 and 20 of the total data respectively The test data arenot presented to the network in the training process When


0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets

minus100 minus80 minus60 minus40 minus20minus100

minus80

minus60

minus40

minus20R = 093302

0 5 10 15minus10

0

10

20

30MSE = 877518

Error

Error

minus40 minus20 0 20 400

5

10

15

0 5 10 15minus100

minus80

minus60

minus40

minus20Train data

Outputs Targets

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

2 MSE = 96384e minus 29


1

2

3

4

1 2 3 4minus60

minus50

minus40

minus30

minus20Test data


minus50

minus40

minus30

minus20R = 063879

1 2 3 40

10

20

30 MSE = 3290692

Error

minus20 0 20 40 600

02040608

1

120583 = 41778

times10minus14

times10minus14

RMSE = 98176e minus 15120583 = minus64595e minus 16

10274e minus 14

120583 = 156668

RMSE = 93676

RMSE = 181403

120590 = 86787

120590 =

120590 = 105591

(a) The RBF prediction values for AP1

1 2 3 4minus100

minus80

minus60

minus40

minus20 Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 088922

1 2 3 4minus40

minus20

0

20MSE = 2640048

Error

minus100 minus50 0 500

02040608

1RMSE = 162482

120583 = minus93368 120590 = 153549

0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 094015

0 5 10 15minus40

minus20

0

20MSE = 704013

Error


5

10

15

RMSE = 83905120583 = minus24898

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

4

2MSE = 16179e minus 28

Error

minus5 0 50

2

4

6times10minus14

times10minus14

RMSE = 1272e minus 14120583 = 16149e minus 15120590 = 13232e minus 14

120590 = 82939

(b) The RBF prediction values for AP2

Figure 7 Continued


1 2 3 4minus80

minus60

minus40

minus20Test data

Outputs Targets

minus80 minus60 minus40 minus20minus80

minus60

minus40

minus20R = 052196

1 2 3 4minus50

minus40

minus30

minus20

minus10 MSE = 8757558

Error


02040608

1RMSE = 295932 120583 = minus274699

120590 = 127102

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 08234

0 5 10 15minus60

minus40

minus20

0

20 MSE = 2335349

Error

minus50 0 500

5

10

15

Outputs Targets

RMSE = 152818120583 = minus73253120590 = 138825

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20

0 5 10 15minus2

minus1

0

1

2 MSE = 97532e minus 29

Error


1

2

3

4R = 1

times10minus14

times10minus14

120583 = minus32297e minus 16120590 = 10352e minus 14RMSE = 98758e minus 15

(c) The RBF prediction values for AP3 RSSI

1 2 3 4minus100

minus80

minus60

minus40

minus20Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 080882

1 2 3 4minus60

minus40

minus20

0MSE = 4860424

Error


05

1

15

2

RMSE = 220464120583 = minus167689

120590 = 165267

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 091205

0 5 10 15minus60

minus40

minus20

0

20 MSE = 1296113

ErrorError

minus50 0 500

5

10

15

Outputs Targets

RMSE = 113847120583 = minus44717 120590 = 108372

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data


minus80

minus60

minus40

minus20

0 5 10 15minus4

minus2

0

2 MSE = 21113e minus 28

minus5 0 50

1

2

3

4R = 1

times10minus14

times10minus14

ErrorOutputs Targets

RMSE = 1453e minus 14 120583 = minus77514e minus 15120590 = 1289e minus 14

(d) The RBF prediction values for AP4 RSSI

Figure 7 Overall performance of RBF for AP1 AP2 AP3 and AP4


Table 1 Real and predicted RSSI values

Time Real RSSI Predicted RSSI using CF-PSO Predicted RSSI using RBFAP1 AP2 AP3 AP4 AP1 AP2 AP3 AP4 AP1 AP2 AP3 AP4

1 minus20 minus29 minus27 minus9817 minus235 minus331 minus296 minus964 minus382 minus438 minus27 minus982 minus27 minus36 minus31 minus9725 minus285 minus348 minus374 minus953 minus345 minus33 minus31 minus973 minus30 minus37 minus40 minus9632 minus325 minus358 minus452 minus95 minus30 minus37 minus40 minus964 minus33 minus37 minus60 minus9444 minus383 minus375 minus591 minus948 minus33 minus467 minus60 minus945 minus38 minus38 minus68 minus9356 minus433 minus404 minus719 minus935 minus38 minus456 minus68 minus936 minus45 minus40 minus80 minus9168 minus479 minus451 minus822 minus899 minus45 minus40 minus646 minus9297 minus50 minus45 minus81 minus9075 minus506 minus489 minus865 minus866 minus49 minus45 minus81 minus908 minus57 minus50 minus87 minus8686 minus549 minus562 minus905 minus793 minus57 minus50 minus87 minus869 minus57 minus60 minus93 minus7992 minus575 minus608 minus912 minus744 minus57 minus60 minus802 minus80410 minus60 minus72 minus87 minus66104 minus633 minus708 minus893 minus63 minus60 minus715 minus87 minus6611 minus66 minus79 minus85 minus52116 minus703 minus809 minus837 minus506 minus655 minus72 minus645 minus5212 minus70 minus80 minus82 minus46128 minus789 minus895 minus754 minus39 minus656 minus708 minus592 minus4613 minus80 minus89 minus76 minus39134 minus839 minus926 minus706 minus341 minus709 minus80 minus76 minus48714 minus94 minus94 minus68 minus33146 minus956 minus946 minus611 minus281 minus90 minus89 minus68 minus47515 minus98 minus95 minus56 minus26

the training process was done the reliability and overfittingof the network were verified with the test data The overallperformance of the proposed models in estimating the RSSIof four APs has been graphically depicted in Figure 7

The real and predicted RSSI values for four APs during 15seconds have been stated in Table 1 By looking at Table 1 it isobserved that the CF-PSOmodel can estimate the RSSI valueabout 800ms before the actual time

It must be mentioned that an unsuitable selection ofthe initial weights may cause local minimum data In orderto prevent this unfavorable phenomenon 30 runs for eachmethod are applied and in each run different random valuesof initial weights are measured Finally in the RBF the best-trained network which has minimum MSE of validateddata is selected as the trained network The estimationperformance of CF-PSO and RBF networks is assessed by1198772 and MSE The output values are stated in Table 2 This

table shows the results of the 30 different running times withiteration = 100

Table 2 shows the 1198772 values of all datasets for the CF-

PSO and RBFN It is clear that the fit is rationally suitable

for all datasets with 119877 values of 1 for the CF-PSO The RBFNwas found to be sufficient for estimation of the RSSI whereasthe CF-PSO model showed a significantly high degree ofaccuracy in the estimation of 119877

2 between 098 and 0993In addition the root of the MSE found that the smaller theRMSEof the test dataset the higher the predictive qualityTheassessment of the CF-PSO and RBF network models showsthe suitable predictive capabilities of the CF-PSO model

4 Network Selection Using the MultiobjectiveParticle Swarm Optimization (MOPSO)

Given the complexity of different access network technolo-gies anAI-based architecturewould provide an efficient solu-tion with high accuracy stability and faster convergence forvertical handover in heterogeneous wireless environmentsOne of the most important aspects of modern communi-cations deals with access to wireless networks by mobiledevices looking for a good quality of service under the userrsquospreferences Nevertheless a mobile terminal can discover


Table 2 1198772 and MSE values

Run numberCF-PSO RBFN

1198772 RMSE 119877

2 RMSEAP1 AP2 AP3 AP4 AP1 AP2 AP3 AP4 AP1 AP2 AP3 AP4 AP1 AP2 AP3 AP4

1 099 098 099 099 069 078 069 063 093 072 078 072 136 198 198 1982 098 098 098 099 078 078 078 069 091 072 066 082 098 198 156 1903 097 098 099 099 079 059 069 069 080 069 067 072 090 102 098 1024 099 099 099 099 069 069 069 069 092 072 068 080 098 198 143 1025 099 098 098 099 069 079 059 059 088 070 078 086 099 136 198 1986 097 098 099 099 079 069 069 069 086 070 064 082 102 098 199 1987 099 098 099 099 069 069 069 069 093 068 077 082 136 136 136 0988 099 098 098 099 069 059 059 069 093 072 068 074 136 198 098 0999 097 099 099 098 079 059 069 089 091 072 076 074 098 198 143 10210 099 098 099 099 069 069 069 069 090 069 067 072 143 133 136 13611 099 098 099 099 069 069 069 069 072 072 068 070 135 198 136 19812 099 098 099 099 069 069 069 069 098 070 068 076 099 133 189 19813 099 099 099 098 059 069 059 080 096 070 064 072 133 136 098 18914 099 099 099 099 069 069 069 059 093 068 067 072 136 098 136 13615 099 098 099 098 069 069 069 079 093 072 058 072 136 198 098 13616 099 098 099 099 058 069 079 069 093 072 056 083 136 198 099 18917 099 098 099 099 069 059 069 069 090 069 085 082 189 189 102 09818 098 099 099 097 078 069 069 079 082 072 058 080 098 198 136 18919 099 099 097 099 069 069 069 069 088 070 058 086 099 136 136 09820 096 098 099 099 080 069 079 059 086 070 054 087 102 136 198 13621 099 098 099 099 069 069 069 069 093 068 057 082 136 098 198 09822 098 099 099 099 069 069 059 069 093 072 058 072 136 198 189 19823 099 099 099 098 069 069 069 079 091 073 056 087 098 098 098 19824 098 099 099 099 078 059 069 069 090 079 057 082 189 143 136 18925 097 098 099 099 079 069 069 079 091 072 058 080 098 198 098 18926 090 098 099 098 088 069 069 079 093 070 058 087 136 098 135 09827 099 099 097 099 069 079 079 069 092 070 054 088 098 136 098 13628 098 099 099 099 069 069 069 059 097 070 047 088 135 098 099 09829 099 099 099 098 069 069 069 079 095 072 058 082 101 198 102 13530 098 098 099 099 069 069 069 069 091 072 056 089 098 198 136 189

more than one network of different technology along its routein heterogeneous scenarios that are capable of connectingto other wireless access points according to their qualityof service values The characteristics of the current mobiledevices are designed to use fast and efficient algorithmsthat provide solutions close to real-time systems Theseconstraints have moved us to develop intelligent algorithmsthat avoid the slow and massive computations associatedwith direct search techniques thus reducing the computationtime

Initially the RSSI is used to identify the existence ofwireless networks In heterogeneous wireless environmentswhen the mobile device senses more than one wirelessnetwork at the same time the network selection with the bestQoS becomes the main problem In the proposed model in

the scanning phase first the physical layer metric (RSSI) ofthe available networks in the heterogeneous environmentsthat perform an intelligence prediction is observed

For this purpose the CF-PSO model has been designedin this paper as a novel prediction model to construct amathematical function which has the best fit to a series ofdata points for RSSI value in UMTS and WLAN networksIn the scanning phase the received signal strength indicatorparameter is measured as the main input to the CF-PSOmodel to predict the received signal strength value for thefour access points and to select the best AP among themusingthe MOPSO algorithm in an intelligent manner

The proposed network selection model evaluates a setof cellular networks in the heterogeneous environmentsThe cost function measures the quality of each network at


every location while moving from one network to anotherIn the vertical handover the cost function can measurethe cost utilized by handing off to a particular network Itis calculated for each network 119899 which covers the servicearea of a user The wireless network technology has theability to support high-rate data services with high spec-trum efficiency In the heterogeneous wireless networksstreaming of multimedia data over heterogeneous wirelessnetworks has increased dramatically and video streamingservices have become more popular among different types ofmultimedia services such as video conferencing and voiceover internet protocol Consideration of video streamingon vertical handover where the capacity of the link is notstable is vital because the network status has an effect on thequality of video streaming During the vertical handover thebandwidth changes suddenly and also the handover latencyleads to packet loss and disconnection of the video In thissection we explain the network selection during the verticalhandover in heterogeneous wireless networks to select thebest network to offer seamless video streaming We can seethat the transferring rate and the quality of transmitted dataare strongly related to the bandwidth variation

The proposed network selection scheme presents amulti-objective decision-making (MODM) function that evaluatesdifferent cellular networks in order to select the best onein the heterogeneous networks This function is defined byusing physical layer metrics such as RSS ABR bit error rate(BER) SNR and throughput of networks in heterogeneousenvironments

In the next stage the ranking of the available networksis performed based on the interface priority scores and theapplication priority scores assigned in stage 1 Consider aset of candidate networks 119878 = 119904

1 1199042 119904

119873 and a set of

quality of service factors 119876 = 1199021 1199022 119902

119872 where 119873 is

the number of candidate networks and 119872 is the numberof quality of service factors In addition each QoS factoris considered to have its own weight and this weight showsthe effect of the factor on the network or user Thus eachnetworkrsquos cost function can be calculated using (8) where119882119873is calculated using the MOPSO (multiobjective particle

swarm optimization) MOPSO [24] It is selected based on itsability to change its weighting among each factor accordingto network conditions and user preferences Therefore

119862119873

= 119882Interface times119872

sum

119895=1

119902119895times 119882119895 (8)

It is known that various types of services need variouspatterns of reliability latency and data rate Therefore weconsider the service type as the main metric In termsof network conditions network-related parameters such asavailable bandwidth and network latency must be measuredfor efficient network usage

In addition the use of network information in theselection for handover can be useful for load balancingamong various types of networks Moreover to maintain thesystemrsquos performance a range of parameters can be used inthe handover decision such as the BERThemobile terminal

condition also includes dynamic factors such as velocitymoving pattern moving histories and location information

QoS is another important parameter that ensures usersatisfaction to the fullest extent The user can establishpreferences for the different QoS parameters depending onthe services required The experimentation is more realisticconsidering different sets of the userrsquos preferences wheremore importantly the QoS parameters have higher valuesof weights It must be mentioned that dynamic factors suchas available bandwidth velocity and network latency havedifferent values in various conditions The set of significantparameters is as follows RSS ABR BER SNR and through-put (119879) which is shown by vectorQN in

QN = [RSS ABR BER SNR 119879] (9)

At the end based on (8) the cost values of each userrsquosrequested service fromnetwork n can be computedThe com-putation time constraint of MOPSO algorithm is relativelylow and eventually conforms to the multimedia time require-ment for reliable communication over wireless networksSeveral algorithms related to theMOPSOhave been proposedin the literature [25] When using the MOPSO it is possiblefor the degree of the velocities to become very large Two tech-niqueswere advanced to control the increase of velocitiesThefirst is to modify the inertia factor in a dynamic manner andthe second is to use the constriction coefficient In order toimprove the inertia and convergence of the particle over timea variant ofMOPSO called theModifiedMOPSO [24] is usedwhere the inertia factor (120596) and constriction coefficient (120575)have been introduced in the velocity update as shown in (1)respectively The network having the lowest cost function isselected as the ldquoalways best connectedrdquo network or optimumaccess network in heterogeneous environments Based on theMOPSO a new handover decision algorithm is proposed toseek the best network to connect the mobile node to theaccess network The proposed model has a cost functionto measure the quality of networks at each situation Ourschemepredicted theRSSI each time to enhance the selectionand the outputs of the prediction unit should be sent to theselection unit Since the selection of the best network is amultiobjective problem hence the MOPSO can be efficientbecause access network selection has conflicting objectives interms of minimization and maximization For example thevertical handover decision algorithm should select the accessnetwork with high RSSI ABR SNR and throughput and thisalgorithm also needs to select the best network with low BERhandover rate and outage probability The operation of ourstrategy is shown in Pseudocode 1

5 Computational Experiments on theMultiobjective Particle Swarm Optimization(MOPSO)

In Section 33 the efficacy of the CF-PSO to predict thereceived signal strength value for the four access points inthe neighborhood is shown We explain the MOPSO as anovel model the best choice for the candidate access pointand as a smart approach in the designed scenario In the


MOPSO Parameters

MaxIt=200 Maximum Number of Iterations

nPop=50 Population Size

nRep=50 Repository Size

w=05 Inertia Weight

wdamp=099 Inertia Weight Damping Rate

c1=1 Personal Learning Coefficient

c2=2 Global Learning Coefficient

nGrid=5 Number of Grids per Dimension

alpha=01 Inflation Rate

beta=2 Leader Selection Pressure

gamma=2 Deletion Selection Pressure

mu=01 Mutation Rate

Initialization

pop=repmat(empty particlenPop1)

for i=1nPop

pop(i)Position=unifrnd(VarMinVarMaxVarSize)

pop(i)Velocity=zeros(VarSize)

pop(i)Cost=CostFunction(pop(i)Position)

Update Personal Best

pop(i)BestPosition=pop(i)Position

pop(i)BestCost=pop(i)Cost

end

Determine Domination

pop=DetermineDomination(pop)

rep=pop(sim[popIsDominated])

Grid=CreateGrid(repnGridalpha)

for i=1numel(rep)

rep(i)=FindGridIndex(rep(i)Grid)

end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop

leader=SelectLeader(repbeta)

pop(i)Velocity = wpop(i)Velocity

+c1rand(VarSize)(pop(i)BestPosition-pop(i)Position)

+c2rand(VarSize)(leaderPosition-pop(i)Position)

pop(i)Position = pop(i)Position + pop(i)Velocity

pop(i)Cost = CostFunction(pop(i)Position)

Apply Mutation

pm=(1-(it-1)(MaxIt-1)) and (1mu)

NewSolPosition=Mutate(pop(i)PositionpmVarMinVarMax)

NewSolCost=CostFunction(NewSolPosition)

rep=[rep pop(sim[popIsDominated])] Add Non-Dominated Particles to REPOSITORY

rep=DetermineDomination(rep) Determine Domination of New Resository Members

rep=rep(sim[repIsDominated]) Keep only Non-Dminated Memebrs in the Repository

Grid=CreateGrid(repnGridalpha) Update Grid

for i=1numel(rep) Update Grid Indices


end

if numel(rep)gtnRep Check if Repository is Full

Extra=numel(rep)-nRep

for e=1Extra

rep=DeleteOneRepMemebr(repgamma)

end

end

Pseudocode 1 Network selection by MOPSO pseudo-code


Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B

Figure 8 Simulation scenario

MOPSO algorithm we have performed simulations to showthe feasibility of the proposed scheme using the simulationsoftware QualNet 70 This experiment is for the case ofa vertical handover between a 3G network and WLANsThe scenario simulated in the QualNet is composed of theUMTS BS and IEEE 80211 APs The WLAN is based on theIEEE 80211b standard and the physical data rate is 11MbpsThe capacity of a 3G Universal Mobile TelecommunicationsSystem is 384 kbps All links except for the wireless links havea capacity of 100Mbps each Ad hoc on-demand distancevector (AODV) protocol is used as the reactive routingprotocol [26] This protocol offers quick convergence whenthe ad hoc network topology changes (typically when a nodemoves in the network)

As shown in Figure 8 the mobile node can be at a fixedtime in the coverage area of the UMTS Nevertheless dueto movement it can travel into the areas that cover morethan one access network that is simultaneously within thecoverage areas of for example a UMTS BS and an IEEE80211 AP Multiple IEEE 80211 WLAN coverage areas areusually comprised within a UMTS coverage area Since theWLAN1 has a lower coverage range when the mobile user ismoving out of aWLAN1 area the existence of an accurate andtimely handoff decision to maintain the connectivity beforethe loss of the WLAN access is necessary Subsequently theuser couldmove into the regions covered by aUMTSnetworkand then the user couldmove into aWLAN2 area to achieve ahigher QoS at the lowest costTherefore the user changes theconnection to the WLAN2The mobile node associated withthe UMTS or WLANs monitors and measures 119863RSS whichis the diversity of the received signal strength between thenetworks of the nearby WLANsUMTS to check whether anaccess network with high data rate is offered

The performance of the proposed algorithm has beenassessed in a scenario when the mobile node moves witha constant speed along a straight line path from the areacovered by WLAN1 to the one covered by UMTS and

then roams to the area covered by WLAN2 Clearly withthe increase of distance the average of RSS ABR SNRand throughput will be reduced and the BER will also beincreased By calculating the cost value based on the numberof iterations an optimal network can be selected as presentedin Figure 9

The MOPSO algorithm using the CF-PSO predictionis compared to the RBF network based on the number ofvertical handoffs The scenario of a heterogeneous wire-less network consisting of WLANs and Universal MobileTelecommunications System (UMTS)B3G is demonstratedin Figure 8 In the simulations the evaluated heterogeneouswireless networks contain 4 to 10 mobile terminals and10 WLANs The mobility of random way point is adoptedfor each mobile node with random direction and randomvelocity from 1 to 25ms the number of UMTSs equals1 arrival rate of Poisson distribution is 6 to 16 and thebandwidths of B3GUMTS and WLAN are 384 kbs and54Mbs respectively The topology covers an area of 2000min length and 2000m in width The simulation considerstwo classes of traffic that is constant bit rate (CBR) andvariable bit rate (VBR) Bursty applications produce VBR(variable bit rate) traffic streams while constant applicationsproduce CBR traffic streams The CBR traffic stream is easyto model and to predict its impact on the performance ofthe network Data rates of CBR of B3G and WLAN are 50and 200 (Kbps) and low and high level data rates of VBR forB3G and WLAN are 10 (Kbps) and 16 (Mbps) respectively[27]

Figure 10 shows the evaluation of the proposed approachby comparing the number of vertical handoffs Clearly theproposed prediction approach considerably improves thenumber of vertical handoffs The 95 confidence intervalsof the simulation results in Figure 10 are created from 30independent runs The number of vertical handoffs of allthe approaches increases when the arrival rate increases Theproposed prediction model the CF-PSO approach resultsin the fewest vertical handoffs Consequently the CF-PSOapproach outperforms other approaches in the number ofvertical handoffs in heterogeneous wireless networks

Figure 11 shows the variation of state between the cellularheterogeneous networkswith respect to the distance based onthe proposed algorithm Clearly when the distance increasesthe cost value increases in the cellular networks The effectof increasing the number of iterations on the cost value hasbeen examined to find the number of iterations essential toreach optima Iterations are different from 10 to 100 basedon the fact that no significant variation was detected after100 successive iterations for WLANs and UMTS cellularnetworks respectively

6 Conclusion

We have proposed a new MOPSO-based vertical handoverdecision algorithm Our goal was to achieve an optimizedMOPSO in a small cost function and for this reason theMOPSO rules were proposed for the heterogeneous wirelessnetworks The CF-PSO prediction model is able to reduce


minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)

Figure 9 (a) Average RSS versus distance (b) ABR versus distance (c) SNR versus distance and (d) throughput versus distance in cellularheterogeneous environment of WLANs and UMTS where ABR ABR = bandwidth times log

2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900

[23] 119864119887is the received energy per bit and 119873

119900is the noise power of the channel

the number of unnecessary handovers and avoid the ldquoPing-Pongrdquo effect in the heterogeneous wireless environmentsOur novel section scheme was aimed at making mobilenodes smart enough to be able to autonomously determinethe best network using data prediction The experimentalresults show that in the prediction of the received signalstrength indicator parameter the model based on CF-PSOis more effective than the RBF neural network model Theselection decision function is defined as a cost functionconsisting of five parameters namely RSS ABR BER SNRand throughput (119879) The proposed approach performed wellwith several test problems in terms of the number of costfunction evaluations required the quality of the solutionsfound the rate of successful minimizations and the averagecost functions Several experiments were performed for

the optimization of the vertical handover decision-makingalgorithms in an intelligent manner in the heterogeneouswireless networksTheMOPSO-VHOhas a low cost functioncompared to previous works Simulation results indicatedthat our proposed vertical handover decision algorithm wasable to minimize the cost function reduce the number ofunnecessary handovers avoid the ldquoPing-Pongrdquo effect andselect the best access network that is optimized to networkconditions quality of service requirements mobile terminalconditions and user preferences

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper


01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate

Prediction by CF-PSORSS-based thresholdPrediction by RBF

Figure 10 Comparing the number of vertical handoffs undervarious arrival rates

0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance

Figure 11 Network selection betweenWLANs and UMTS based onthe proposed algorithm

Acknowledgments

The authors would like to thank the Ministry of Educa-tion Malaysia for funding this research project through aResearch University Grant of Universiti Teknologi Malaysia(UTM) Malaysian Japan International Institute of Technol-ogy (MJIIT) and i-Kohza Computer System and Network(CSN) Also the authors thank University of Malaya for thefinancial support (BKP Grant BK060-2014) and facilities tocarry out the work

References

[1] LGiupponi RAgusti J Perez-Romero andO Sallent ldquoAnoveljoint radio resource management approach with reinforce-ment learning mechanismsrdquo in Proceedings of the 24th IEEEInternational Performance Computing and CommunicationsConference (IPCCC rsquo05) pp 621ndash626 IEEE Phoenix ArizUSA April 2005

[2] A Wilson A Lenaghan and R Malyan ldquoOptimising wirelessaccess network selection to maintain QoS in heterogeneouswireless environmentsrdquo in Proceedings of the 8th InternationalSymposium on Wireless Personal Multimedia Communications(WPMC rsquo05) Aalborg Denmark September 2005

[3] A Ahmed L M Boulahia and D Gaıti ldquoEnabling verticalhandover decisions in heterogeneous wireless networks a state-of-the-art and a classificationrdquo IEEECommunications SurveysampTutorials vol 16 no 2 pp 776ndash811 2014

[4] S Paul J Pan and R Jain ldquoArchitectures for the futurenetworks and the next generation internet a surveyrdquo ComputerCommunications vol 34 no 1 pp 2ndash42 2011

[5] P TalebiFard T Wong and V C M Leung ldquoAccess and serviceconvergence over the mobile internetmdasha surveyrdquo ComputerNetworks vol 54 no 4 pp 545ndash557 2010

[6] Z Movahedi M Ayari R Langar and G Pujolle ldquoA survey ofautonomic network architectures and evaluation criteriardquo IEEECommunications Surveys and Tutorials vol 14 no 2 pp 464ndash490 2012

[7] V Rakovic and L Gavrilovska ldquoNovel RAT selection mecha-nism based on Hopfield neural networksrdquo in Paper Presented atthe International Congress onUltraModern Telecommunicationsand Control Systems and Workshops (ICUMT rsquo10) pp 210ndash217Moscow Russia October 2010

[8] A Calhan and C Ceken ldquoArtificial neural network based ver-tical handoff algorithm for reducing handoff latencyrdquo WirelessPersonal Communications vol 71 no 4 pp 2399ndash2415 2013

[9] A Calhan and C Ceken ldquoCase study on handoff strategies forwireless overlay networksrdquo Computer Standards and Interfacesvol 35 no 1 pp 170ndash178 2013

[10] N Wang W Shi S Fan and S Liu ldquoPSO-FNN-based ver-tical handoff decision algorithm in heterogeneous wirelessnetworksrdquo Procedia Environmental Sciences vol 11 part A pp55ndash62 2011

[11] X Liu and L-G Jiang ldquoA novel vertical handoff algorithmbased on fuzzy logic in aid of grey prediction theory inwireless heterogeneous networksrdquo Journal of Shanghai JiaotongUniversity (Science) vol 17 no 1 pp 25ndash30 2012

[12] A Singhrova and N Prakash ldquoVertical handoff decision algo-rithm for improved quality of service in heterogeneous wirelessnetworksrdquo IET Communications vol 6 no 2 pp 211ndash223 2012

[13] K Pahlavan P Krishnamurthy A Hatami et al ldquoHandoff inhybrid mobile data networksrdquo IEEE Personal Communicationsvol 7 no 2 pp 34ndash47 2000

[14] Z Altman S Sallem R Nasri B Sayrac andM Clerc ldquoParticleswarm optimization for mobility load balancing SON in LTEnetworksrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference Workshops (WCNCW rsquo14) pp 172ndash177 IEEE Istanbul Turkey April 2014

[15] X-W Wang P-Y Qin M Huang and H Cheng ldquoNiche PSObased QoS handoff decision scheme with ABC supportedrdquo inProceedings of the IEEE International Conference on IntelligentComputing and Intelligent Systems (ICIS rsquo09) pp 423ndash427Shanghai China November 2009

[16] S Venkatachalaiah R J Harris and J Murphy ldquoImprovinghandoff in wireless networks using Grey and particle swarmoptimisationrdquo in Proceedings of the 2nd International Confer-ence on Computing Communication and Control Technologies(CCCT rsquo04) vol 5 pp 368ndash373 Austin Tex USA August 2004

[17] M Rehan M Yousaf A Qayyum and S Malik ldquoA cross-layeruser centric vertical handover decision approach based onMIHlocal triggersrdquo in Wireless and Mobile Networking vol 308 pp359ndash369 Springer Berlin Germany 2009

[18] M O Ansong H-X Yao and J S Huang ldquoRadial and sigmoidbasis function neural networks in wireless sensor routingtopology control in underground mine rescue operation based


on particle swarm optimizationrdquo International Journal of Dis-tributed Sensor Networks vol 2013 Article ID 376931 14 pages2013

[19] C-H Kuo T-A Liu J-H Chen C-M J Chang and C-J Shieh ldquoResponse surface methodology and artificial neu-ral network optimized synthesis of enzymatic 2-phenylethylacetate in a solvent-free systemrdquo Biocatalysis and AgriculturalBiotechnology vol 3 no 3 pp 1ndash6 2014

[20] P Shabanzadeh R Yusof and K Shameli ldquoArtificial neuralnetwork for modeling the size of silver nanoparticlesrsquo pre-pared in montmorillonitestarch bionanocompositesrdquo Journalof Industrial and Engineering Chemistry vol 24 pp 42ndash50 2015

[21] G M Foody ldquoSupervised image classification byMLP and RBFneural networks with and without an exhaustively defined setof classesrdquo International Journal of Remote Sensing vol 25 no15 pp 3091ndash3104 2004

[22] H D Pfister J B Soriaga and P H Siegel ldquoOn the achievableinformation rates of finite state ISI channelsrdquo in Proceedings ofthe IEEE Global Telecommunications Conference (GLOBECOMrsquo01) vol 5 pp 2992ndash2996 IEEE San Antonio Tex USANovember 2001

[23] F R Reza ldquoOptimum ranges for data transmission in mobilecommunicationsrdquo International Journal of ScientificampEngineer-ing Research vol 3 no 4 pp 481ndash489 2012

[24] C A Coello Coello andM S Lechuga ldquoMOPSO a proposal formultiple objective particle swarm optimizationrdquo in Proceedingsof the Congress on Evolutionary Computation (CEC rsquo02) vol 2pp 1051ndash1056 IEEE Honolulu Hawaii USA May 2002

[25] M Reyes-Sierra and C A Coello Coello ldquoMulti-objectiveparticle swarm optimizers a survey of the state-of-the-artrdquoInternational Journal of Computational Intelligence Researchvol 2 no 3 pp 287ndash308 2006

[26] C E Perkins E M Royer S R Das and M K MarinaldquoPerformance comparison of two on-demand routing protocolsfor ad hoc networksrdquo IEEE Personal Communications vol 8 no1 pp 16ndash28 2001

[27] B-J Chang and J-F Chen ldquoCross-layer-based adaptive ver-tical handoff with predictive RSS in heterogeneous wirelessnetworksrdquo IEEE Transactions on Vehicular Technology vol 57no 6 pp 3679ndash3692 2008

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MathematicsJournal of


Mathematical Problems in Engineering

Hindawi Publishing Corporationhttpwwwhindawicom

Differential EquationsInternational Journal of

Volume 2014

Applied MathematicsJournal of


Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of


Mathematical PhysicsAdvances in

Complex AnalysisJournal of


OptimizationJournal of


CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of


Operations ResearchAdvances in

Journal of


Function Spaces

Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of Mathematics and Mathematical Sciences


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Algebra

Discrete Dynamics in Nature and Society



Decision SciencesAdvances in

Discrete MathematicsJournal of


Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Stochastic AnalysisInternational Journal of


information about data rate monetary cost and receivedsignal strength as different parameters to make a handoverdecision has been reported by [8 9] The main weaknessesare related to the computation of the error function andthe Jacobian inversion for acquiring a matrix in which thesizes are equal to the whole of all the weights in the neuralnetwork (NN) Therefore the necessity for memory is veryhigh Existing algorithms [8 9] consider the service fee thereceived signal strength information (RSSI) user preferenceand so forth The proposed algorithm in comparison withthe traditional RSSI-based algorithm enhances the outcomesmainly for both the user and the network as a result ofthe offered fuzzy based handover systems In terms ofhybrid categories [10] proposed a PSO-FNN-based verticalhandover decision scheme that could make an intelligenthandover decision based on the analysis of the networkrsquosposition The authors of [11] mainly dealt with a novelvertical handover decision algorithm built on fuzzy logicwith the assistance of the Grey theory and the dynamicweights adaptation A neurofuzzy multiparameter-based ver-tical handover decision algorithm (VHDA) was proposedby [12] The results of the performance evaluation carriedout by the handover quality indicator (used to quantifyQoS) which is related to the ldquoPing-Pongrdquo effect ESA andthroughput proved that the proposed VHDA offered betterQoS than the existing vertical handover methods Pahlavanet al [13] method is a good representation of applying afuzzy logic-based normalized quantitative decision algorithmand a differential prediction algorithm with a high level ofaccuracy

The vertical handover schemes stated above have theirown benefits but they do not consider the complexity ofthe network selection and the allowance of lower compu-tation cost function is unreasonable Clearly the decisionprocess should focus on a steadfast intelligent algorithm toexecute an accurate decision and to shift to the best networkcandidate quickly The goal of this study is to propose anovel network selection optimization algorithm that takesadvantage of the prediction model using the CF-PSO tomeet the requirements stated above The usage of the CF-PSO algorithm for prediction with the PSO algorithm forselection has three purposes (1) to serve as a validationalgorithm for the outcome of the MOPSO (2) to decreasethe number of unnecessary handovers and prevent the ldquoPing-Pongrdquo effect and (3) to select the best candidate accesspoint among various access technologiesThis paper providesa comparison based on two prediction methods namelythe CF-PSO and the RBF neural network to predict theRSSI More importantly this comparison evaluates the twomethods from different aspects for example time coefficientdetermination and mean squared error The results showthat the CF-PSO has better performance which will bepresented in-depth in the followingThe proposed method isa step towards future computer-based optimization methodswhere huge uncertainties by the optimization algorithmmustbe avoided To do this the prediction algorithm is com-bined with the particle swarm optimization The proposedMOPSO-based vertical handover decision algorithm canmake an intelligent handover decision based on the network

Set RSSI for APi

Prediction of RSSI by CS-PSO algorithm

QoS factors observation

Network selection by MO-PSO

Figure 1 Proposed network selection model based on the Mul-tiobjective Particle Swarm Optimization (MOPSO) in a wirelessheterogeneous environment

positionThe proposed network selection model is presentedin Figure 1

The outline of this paper is as follows First the relatedworks are described in Section 2 In Section 3 the proposedprediction methods based on CF-PSO and RBF networkare introduced Then the results of comparison based onthe proposed models are illustrated Section 4 illustratesthe MOPSO-based vertical handover decision algorithmSection 5 analyzes the performance of the algorithm throughthe simulation results Finally Section 6 concludes the paper

2 Related Work

Many network selection algorithms have been proposed inliterature in which PSO has been used for handover deci-sions It is initializedwith a set of randomparticles (solutions)that finds an optimal result by updating the generations Theoptimum results are called particles which fly throughoutthe problem space by following the current optimal particlesIn addition a GSM-like [14] hard handover algorithm hasbeen proposed where countermeasures prevent the Ping-Pong effect by providing a baseline handover thresholdagainst power fluctuations due to channel variations Theauthors defined four rules where two rules were for pre-venting unnecessary handovers (Ping-Pong effects) and onerule was aimed at helping the loaded eNB j by delayinghandovers from eNB i Moreover the other rule was aimed atalleviating the loaded eNB i by advancing handovers towardseNB j The PSO algorithm denoted that the neighbor withwhich the eNB has the largest handover traffic exchangewas the best neighbor Prior knowledge is required on theoptimization process that provided the parameterized formof the controller which was optimized by the MOPSO It hasbeen shown that the dynamic optimization outperformed thestatic optimization and produced a better mobility load bal-ancing self-organizing network controller which improvedthe throughput and access probability by a few percents withrespect to the planning solution

Wang et al [15] proposed an always best connected (ABC)maintained QoS handover decision scheme based on theniche PSO algorithm This literature reported the access








123456789101112131415








119881119898+1

119894= 120596 lowast 119881

119898


119894minus 119904119898

119894) + 1198882rand2


119894)

(1)


Vmi

Vm+1

i

Vgbe

st

i

Vpbesti

Xmi

Xm+1i

pbesti

gbesti


where 119881119898












middotmiddotmiddot



119865 (119909) = [11987511198834+ 11987521198833+ 11987531198832+ 11987541198831+ 1198755] (2)






minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP1

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP2

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP3

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

RSSI

AP4

(dBm

)

Time




MSE =1

119903sum119903

119894=1(V119901119894

minus V119886119894)2 (3)

1198772= 1 minus

sum119903

119894=1(V119901119894

minus V119886119894)2

sum119903

119894=1(V119901119894

minus V119886V)2

(4)

RMSE = radic

1

119903sum119903

119894=1(V119901119894

minus V119886119894)2 (5)


119886119894







minus20minus27

minus30minus33

minus38minus45


minus60minus66

minus70

minus80

minus94minus98

minus140

minus120

minus100

minus80

minus60

minus40

minus20

00 5 10 15 20


minus40minus45

minus50

minus60

minus72

minus79minus80


minus100

minus90

minus80

minus70

minus60

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20

minus27minus31

minus40

minus60

minus68



minus82minus76

minus68

minus56

minus100

minus90

minus80

minus70

minus60


minus79

minus66

minus52minus46

minus39minus33

minus26

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20 0 5 10 15 20

minus120

minus100

minus80

minus60

minus40

minus20

0

y = minus00037x4 + 00868x3 minus 06516x2

minus

y = minus00116x4 + 03419x3 minus 27611x2

y = minus00079x4 + 02758x3 minus 22402x2

minus

29826x minus 17137

44995x minus 17932

R2 = 099107

R2 = 09937

R2 = 098325

R2 = 0991

y = 00097x4 minus 026

minus 7316

82x3 + 20218x2

2x minus 2529

+ 88404x minus 10505



119910119896(119909) =

119873

sum

119895=1

1199081198961198950119895(119909) + 119908

1198960 (6)





0119895(119909) = exp(minus

1003817100381710038171003817119909 minus 1205831003817100381710038171003817

2

21205902

119895

) (7)





0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 093302

0 5 10 15minus10

0

10

20

30MSE = 877518

Error

Error


5

10

15

0 5 10 15minus100

minus80

minus60

minus40

minus20Train data

Outputs Targets

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

2 MSE = 96384e minus 29


1

2

3

4

1 2 3 4minus60

minus50

minus40

minus30

minus20Test data


minus50

minus40

minus30

minus20R = 063879

1 2 3 40

10

20

30 MSE = 3290692

Error

minus20 0 20 40 600

02040608

1

120583 = 41778

times10minus14

times10minus14


10274e minus 14

120583 = 156668

RMSE = 93676

RMSE = 181403

120590 = 86787

120590 =

120590 = 105591


1 2 3 4minus100

minus80

minus60

minus40

minus20 Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 088922

1 2 3 4minus40

minus20

0

20MSE = 2640048

Error


02040608

1RMSE = 162482

120583 = minus93368 120590 = 153549

0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 094015

0 5 10 15minus40

minus20

0

20MSE = 704013

Error


5

10

15

RMSE = 83905120583 = minus24898

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

4


Error

minus5 0 50

2

4

6times10minus14

times10minus14


120590 = 82939


Figure 7 Continued


1 2 3 4minus80

minus60

minus40

minus20Test data

Outputs Targets


minus60

minus40

minus20R = 052196

1 2 3 4minus50

minus40

minus30

minus20

minus10 MSE = 8757558

Error


02040608

1RMSE = 295932 120583 = minus274699

120590 = 127102

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 08234

0 5 10 15minus60

minus40

minus20

0

20 MSE = 2335349

Error

minus50 0 500

5

10

15

Outputs Targets

RMSE = 152818120583 = minus73253120590 = 138825

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20

0 5 10 15minus2

minus1

0

1

2 MSE = 97532e minus 29

Error


1

2

3

4R = 1

times10minus14

times10minus14



1 2 3 4minus100

minus80

minus60

minus40

minus20Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 080882

1 2 3 4minus60

minus40

minus20

0MSE = 4860424

Error


05

1

15

2

RMSE = 220464120583 = minus167689

120590 = 165267

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 091205

0 5 10 15minus60

minus40

minus20

0

20 MSE = 1296113

ErrorError

minus50 0 500

5

10

15

Outputs Targets

RMSE = 113847120583 = minus44717 120590 = 108372

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data


minus80

minus60

minus40

minus20

0 5 10 15minus4

minus2

0

2 MSE = 21113e minus 28

minus5 0 50

1

2

3

4R = 1

times10minus14

times10minus14






















1198772 RMSE 119877


1 099 098 099 099 069 078 069 063 093 072 078 072 136 198 198 1982 098 098 098 099 078 078 078 069 091 072 066 082 098 198 156 1903 097 098 099 099 079 059 069 069 080 069 067 072 090 102 098 1024 099 099 099 099 069 069 069 069 092 072 068 080 098 198 143 1025 099 098 098 099 069 079 059 059 088 070 078 086 099 136 198 1986 097 098 099 099 079 069 069 069 086 070 064 082 102 098 199 1987 099 098 099 099 069 069 069 069 093 068 077 082 136 136 136 0988 099 098 098 099 069 059 059 069 093 072 068 074 136 198 098 0999 097 099 099 098 079 059 069 089 091 072 076 074 098 198 143 10210 099 098 099 099 069 069 069 069 090 069 067 072 143 133 136 13611 099 098 099 099 069 069 069 069 072 072 068 070 135 198 136 19812 099 098 099 099 069 069 069 069 098 070 068 076 099 133 189 19813 099 099 099 098 059 069 059 080 096 070 064 072 133 136 098 18914 099 099 099 099 069 069 069 059 093 068 067 072 136 098 136 13615 099 098 099 098 069 069 069 079 093 072 058 072 136 198 098 13616 099 098 099 099 058 069 079 069 093 072 056 083 136 198 099 18917 099 098 099 099 069 059 069 069 090 069 085 082 189 189 102 09818 098 099 099 097 078 069 069 079 082 072 058 080 098 198 136 18919 099 099 097 099 069 069 069 069 088 070 058 086 099 136 136 09820 096 098 099 099 080 069 079 059 086 070 054 087 102 136 198 13621 099 098 099 099 069 069 069 069 093 068 057 082 136 098 198 09822 098 099 099 099 069 069 059 069 093 072 058 072 136 198 189 19823 099 099 099 098 069 069 069 079 091 073 056 087 098 098 098 19824 098 099 099 099 078 059 069 069 090 079 057 082 189 143 136 18925 097 098 099 099 079 069 069 079 091 072 058 080 098 198 098 18926 090 098 099 098 088 069 069 079 093 070 058 087 136 098 135 09827 099 099 097 099 069 079 079 069 092 070 054 088 098 136 098 13628 098 099 099 099 069 069 069 059 097 070 047 088 135 098 099 09829 099 099 099 098 069 069 069 079 095 072 058 082 101 198 102 13530 098 098 099 099 069 069 069 069 091 072 056 089 098 198 136 189










1 1199042 119904

119873 and a set of


119872 where 119873 is



119862119873


sum

119895=1

119902119895times 119882119895 (8)










MOPSO Parameters




w=05 Inertia Weight








mu=01 Mutation Rate

Initialization


for i=1nPop







end





for i=1numel(rep)


end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop







Apply Mutation










end



for e=1Extra


end

end



Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra
















123456789101112131415








119881119898+1

119894= 120596 lowast 119881

119898


119894minus 119904119898

119894) + 1198882rand2


119894)

(1)


Vmi

Vm+1

i

Vgbe

st

i

Vpbesti

Xmi

Xm+1i

pbesti

gbesti


where 119881119898












middotmiddotmiddot



119865 (119909) = [11987511198834+ 11987521198833+ 11987531198832+ 11987541198831+ 1198755] (2)






minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP1

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP2

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP3

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

RSSI

AP4

(dBm

)

Time




MSE =1

119903sum119903

119894=1(V119901119894

minus V119886119894)2 (3)

1198772= 1 minus

sum119903

119894=1(V119901119894

minus V119886119894)2

sum119903

119894=1(V119901119894

minus V119886V)2

(4)

RMSE = radic

1

119903sum119903

119894=1(V119901119894

minus V119886119894)2 (5)


119886119894







minus20minus27

minus30minus33

minus38minus45


minus60minus66

minus70

minus80

minus94minus98

minus140

minus120

minus100

minus80

minus60

minus40

minus20

00 5 10 15 20


minus40minus45

minus50

minus60

minus72

minus79minus80


minus100

minus90

minus80

minus70

minus60

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20

minus27minus31

minus40

minus60

minus68



minus82minus76

minus68

minus56

minus100

minus90

minus80

minus70

minus60


minus79

minus66

minus52minus46

minus39minus33

minus26

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20 0 5 10 15 20

minus120

minus100

minus80

minus60

minus40

minus20

0

y = minus00037x4 + 00868x3 minus 06516x2

minus

y = minus00116x4 + 03419x3 minus 27611x2

y = minus00079x4 + 02758x3 minus 22402x2

minus

29826x minus 17137

44995x minus 17932

R2 = 099107

R2 = 09937

R2 = 098325

R2 = 0991

y = 00097x4 minus 026

minus 7316

82x3 + 20218x2

2x minus 2529

+ 88404x minus 10505



119910119896(119909) =

119873

sum

119895=1

1199081198961198950119895(119909) + 119908

1198960 (6)





0119895(119909) = exp(minus

1003817100381710038171003817119909 minus 1205831003817100381710038171003817

2

21205902

119895

) (7)





0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 093302

0 5 10 15minus10

0

10

20

30MSE = 877518

Error

Error


5

10

15

0 5 10 15minus100

minus80

minus60

minus40

minus20Train data

Outputs Targets

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

2 MSE = 96384e minus 29


1

2

3

4

1 2 3 4minus60

minus50

minus40

minus30

minus20Test data


minus50

minus40

minus30

minus20R = 063879

1 2 3 40

10

20

30 MSE = 3290692

Error

minus20 0 20 40 600

02040608

1

120583 = 41778

times10minus14

times10minus14


10274e minus 14

120583 = 156668

RMSE = 93676

RMSE = 181403

120590 = 86787

120590 =

120590 = 105591


1 2 3 4minus100

minus80

minus60

minus40

minus20 Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 088922

1 2 3 4minus40

minus20

0

20MSE = 2640048

Error


02040608

1RMSE = 162482

120583 = minus93368 120590 = 153549

0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 094015

0 5 10 15minus40

minus20

0

20MSE = 704013

Error


5

10

15

RMSE = 83905120583 = minus24898

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

4


Error

minus5 0 50

2

4

6times10minus14

times10minus14


120590 = 82939


Figure 7 Continued


1 2 3 4minus80

minus60

minus40

minus20Test data

Outputs Targets


minus60

minus40

minus20R = 052196

1 2 3 4minus50

minus40

minus30

minus20

minus10 MSE = 8757558

Error


02040608

1RMSE = 295932 120583 = minus274699

120590 = 127102

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 08234

0 5 10 15minus60

minus40

minus20

0

20 MSE = 2335349

Error

minus50 0 500

5

10

15

Outputs Targets

RMSE = 152818120583 = minus73253120590 = 138825

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20

0 5 10 15minus2

minus1

0

1

2 MSE = 97532e minus 29

Error


1

2

3

4R = 1

times10minus14

times10minus14



1 2 3 4minus100

minus80

minus60

minus40

minus20Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 080882

1 2 3 4minus60

minus40

minus20

0MSE = 4860424

Error


05

1

15

2

RMSE = 220464120583 = minus167689

120590 = 165267

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 091205

0 5 10 15minus60

minus40

minus20

0

20 MSE = 1296113

ErrorError

minus50 0 500

5

10

15

Outputs Targets

RMSE = 113847120583 = minus44717 120590 = 108372

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data


minus80

minus60

minus40

minus20

0 5 10 15minus4

minus2

0

2 MSE = 21113e minus 28

minus5 0 50

1

2

3

4R = 1

times10minus14

times10minus14






















1198772 RMSE 119877


1 099 098 099 099 069 078 069 063 093 072 078 072 136 198 198 1982 098 098 098 099 078 078 078 069 091 072 066 082 098 198 156 1903 097 098 099 099 079 059 069 069 080 069 067 072 090 102 098 1024 099 099 099 099 069 069 069 069 092 072 068 080 098 198 143 1025 099 098 098 099 069 079 059 059 088 070 078 086 099 136 198 1986 097 098 099 099 079 069 069 069 086 070 064 082 102 098 199 1987 099 098 099 099 069 069 069 069 093 068 077 082 136 136 136 0988 099 098 098 099 069 059 059 069 093 072 068 074 136 198 098 0999 097 099 099 098 079 059 069 089 091 072 076 074 098 198 143 10210 099 098 099 099 069 069 069 069 090 069 067 072 143 133 136 13611 099 098 099 099 069 069 069 069 072 072 068 070 135 198 136 19812 099 098 099 099 069 069 069 069 098 070 068 076 099 133 189 19813 099 099 099 098 059 069 059 080 096 070 064 072 133 136 098 18914 099 099 099 099 069 069 069 059 093 068 067 072 136 098 136 13615 099 098 099 098 069 069 069 079 093 072 058 072 136 198 098 13616 099 098 099 099 058 069 079 069 093 072 056 083 136 198 099 18917 099 098 099 099 069 059 069 069 090 069 085 082 189 189 102 09818 098 099 099 097 078 069 069 079 082 072 058 080 098 198 136 18919 099 099 097 099 069 069 069 069 088 070 058 086 099 136 136 09820 096 098 099 099 080 069 079 059 086 070 054 087 102 136 198 13621 099 098 099 099 069 069 069 069 093 068 057 082 136 098 198 09822 098 099 099 099 069 069 059 069 093 072 058 072 136 198 189 19823 099 099 099 098 069 069 069 079 091 073 056 087 098 098 098 19824 098 099 099 099 078 059 069 069 090 079 057 082 189 143 136 18925 097 098 099 099 079 069 069 079 091 072 058 080 098 198 098 18926 090 098 099 098 088 069 069 079 093 070 058 087 136 098 135 09827 099 099 097 099 069 079 079 069 092 070 054 088 098 136 098 13628 098 099 099 099 069 069 069 059 097 070 047 088 135 098 099 09829 099 099 099 098 069 069 069 079 095 072 058 082 101 198 102 13530 098 098 099 099 069 069 069 069 091 072 056 089 098 198 136 189










1 1199042 119904

119873 and a set of


119872 where 119873 is



119862119873


sum

119895=1

119902119895times 119882119895 (8)










MOPSO Parameters




w=05 Inertia Weight








mu=01 Mutation Rate

Initialization


for i=1nPop







end





for i=1numel(rep)


end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop







Apply Mutation










end



for e=1Extra


end

end



Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra










Vmi

Vm+1

i

Vgbe

st

i

Vpbesti

Xmi

Xm+1i

pbesti

gbesti


where 119881119898












middotmiddotmiddot



119865 (119909) = [11987511198834+ 11987521198833+ 11987531198832+ 11987541198831+ 1198755] (2)






minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP1

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP2

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP3

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

RSSI

AP4

(dBm

)

Time




MSE =1

119903sum119903

119894=1(V119901119894

minus V119886119894)2 (3)

1198772= 1 minus

sum119903

119894=1(V119901119894

minus V119886119894)2

sum119903

119894=1(V119901119894

minus V119886V)2

(4)

RMSE = radic

1

119903sum119903

119894=1(V119901119894

minus V119886119894)2 (5)


119886119894







minus20minus27

minus30minus33

minus38minus45


minus60minus66

minus70

minus80

minus94minus98

minus140

minus120

minus100

minus80

minus60

minus40

minus20

00 5 10 15 20


minus40minus45

minus50

minus60

minus72

minus79minus80


minus100

minus90

minus80

minus70

minus60

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20

minus27minus31

minus40

minus60

minus68



minus82minus76

minus68

minus56

minus100

minus90

minus80

minus70

minus60


minus79

minus66

minus52minus46

minus39minus33

minus26

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20 0 5 10 15 20

minus120

minus100

minus80

minus60

minus40

minus20

0

y = minus00037x4 + 00868x3 minus 06516x2

minus

y = minus00116x4 + 03419x3 minus 27611x2

y = minus00079x4 + 02758x3 minus 22402x2

minus

29826x minus 17137

44995x minus 17932

R2 = 099107

R2 = 09937

R2 = 098325

R2 = 0991

y = 00097x4 minus 026

minus 7316

82x3 + 20218x2

2x minus 2529

+ 88404x minus 10505



119910119896(119909) =

119873

sum

119895=1

1199081198961198950119895(119909) + 119908

1198960 (6)





0119895(119909) = exp(minus

1003817100381710038171003817119909 minus 1205831003817100381710038171003817

2

21205902

119895

) (7)





0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 093302

0 5 10 15minus10

0

10

20

30MSE = 877518

Error

Error


5

10

15

0 5 10 15minus100

minus80

minus60

minus40

minus20Train data

Outputs Targets

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

2 MSE = 96384e minus 29


1

2

3

4

1 2 3 4minus60

minus50

minus40

minus30

minus20Test data


minus50

minus40

minus30

minus20R = 063879

1 2 3 40

10

20

30 MSE = 3290692

Error

minus20 0 20 40 600

02040608

1

120583 = 41778

times10minus14

times10minus14


10274e minus 14

120583 = 156668

RMSE = 93676

RMSE = 181403

120590 = 86787

120590 =

120590 = 105591


1 2 3 4minus100

minus80

minus60

minus40

minus20 Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 088922

1 2 3 4minus40

minus20

0

20MSE = 2640048

Error


02040608

1RMSE = 162482

120583 = minus93368 120590 = 153549

0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 094015

0 5 10 15minus40

minus20

0

20MSE = 704013

Error


5

10

15

RMSE = 83905120583 = minus24898

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

4


Error

minus5 0 50

2

4

6times10minus14

times10minus14


120590 = 82939


Figure 7 Continued


1 2 3 4minus80

minus60

minus40

minus20Test data

Outputs Targets


minus60

minus40

minus20R = 052196

1 2 3 4minus50

minus40

minus30

minus20

minus10 MSE = 8757558

Error


02040608

1RMSE = 295932 120583 = minus274699

120590 = 127102

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 08234

0 5 10 15minus60

minus40

minus20

0

20 MSE = 2335349

Error

minus50 0 500

5

10

15

Outputs Targets

RMSE = 152818120583 = minus73253120590 = 138825

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20

0 5 10 15minus2

minus1

0

1

2 MSE = 97532e minus 29

Error


1

2

3

4R = 1

times10minus14

times10minus14



1 2 3 4minus100

minus80

minus60

minus40

minus20Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 080882

1 2 3 4minus60

minus40

minus20

0MSE = 4860424

Error


05

1

15

2

RMSE = 220464120583 = minus167689

120590 = 165267

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 091205

0 5 10 15minus60

minus40

minus20

0

20 MSE = 1296113

ErrorError

minus50 0 500

5

10

15

Outputs Targets

RMSE = 113847120583 = minus44717 120590 = 108372

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data


minus80

minus60

minus40

minus20

0 5 10 15minus4

minus2

0

2 MSE = 21113e minus 28

minus5 0 50

1

2

3

4R = 1

times10minus14

times10minus14






















1198772 RMSE 119877


1 099 098 099 099 069 078 069 063 093 072 078 072 136 198 198 1982 098 098 098 099 078 078 078 069 091 072 066 082 098 198 156 1903 097 098 099 099 079 059 069 069 080 069 067 072 090 102 098 1024 099 099 099 099 069 069 069 069 092 072 068 080 098 198 143 1025 099 098 098 099 069 079 059 059 088 070 078 086 099 136 198 1986 097 098 099 099 079 069 069 069 086 070 064 082 102 098 199 1987 099 098 099 099 069 069 069 069 093 068 077 082 136 136 136 0988 099 098 098 099 069 059 059 069 093 072 068 074 136 198 098 0999 097 099 099 098 079 059 069 089 091 072 076 074 098 198 143 10210 099 098 099 099 069 069 069 069 090 069 067 072 143 133 136 13611 099 098 099 099 069 069 069 069 072 072 068 070 135 198 136 19812 099 098 099 099 069 069 069 069 098 070 068 076 099 133 189 19813 099 099 099 098 059 069 059 080 096 070 064 072 133 136 098 18914 099 099 099 099 069 069 069 059 093 068 067 072 136 098 136 13615 099 098 099 098 069 069 069 079 093 072 058 072 136 198 098 13616 099 098 099 099 058 069 079 069 093 072 056 083 136 198 099 18917 099 098 099 099 069 059 069 069 090 069 085 082 189 189 102 09818 098 099 099 097 078 069 069 079 082 072 058 080 098 198 136 18919 099 099 097 099 069 069 069 069 088 070 058 086 099 136 136 09820 096 098 099 099 080 069 079 059 086 070 054 087 102 136 198 13621 099 098 099 099 069 069 069 069 093 068 057 082 136 098 198 09822 098 099 099 099 069 069 059 069 093 072 058 072 136 198 189 19823 099 099 099 098 069 069 069 079 091 073 056 087 098 098 098 19824 098 099 099 099 078 059 069 069 090 079 057 082 189 143 136 18925 097 098 099 099 079 069 069 079 091 072 058 080 098 198 098 18926 090 098 099 098 088 069 069 079 093 070 058 087 136 098 135 09827 099 099 097 099 069 079 079 069 092 070 054 088 098 136 098 13628 098 099 099 099 069 069 069 059 097 070 047 088 135 098 099 09829 099 099 099 098 069 069 069 079 095 072 058 082 101 198 102 13530 098 098 099 099 069 069 069 069 091 072 056 089 098 198 136 189










1 1199042 119904

119873 and a set of


119872 where 119873 is



119862119873


sum

119895=1

119902119895times 119882119895 (8)










MOPSO Parameters




w=05 Inertia Weight








mu=01 Mutation Rate

Initialization


for i=1nPop







end





for i=1numel(rep)


end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop







Apply Mutation










end



for e=1Extra


end

end



Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra










minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP1

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP2

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 3 5 7 9 11 13 15

RSSI

AP3

(dBm

)

Time



minus120

minus100

minus80

minus60

minus40

minus20

0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

RSSI

AP4

(dBm

)

Time




MSE =1

119903sum119903

119894=1(V119901119894

minus V119886119894)2 (3)

1198772= 1 minus

sum119903

119894=1(V119901119894

minus V119886119894)2

sum119903

119894=1(V119901119894

minus V119886V)2

(4)

RMSE = radic

1

119903sum119903

119894=1(V119901119894

minus V119886119894)2 (5)


119886119894







minus20minus27

minus30minus33

minus38minus45


minus60minus66

minus70

minus80

minus94minus98

minus140

minus120

minus100

minus80

minus60

minus40

minus20

00 5 10 15 20


minus40minus45

minus50

minus60

minus72

minus79minus80


minus100

minus90

minus80

minus70

minus60

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20

minus27minus31

minus40

minus60

minus68



minus82minus76

minus68

minus56

minus100

minus90

minus80

minus70

minus60


minus79

minus66

minus52minus46

minus39minus33

minus26

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20 0 5 10 15 20

minus120

minus100

minus80

minus60

minus40

minus20

0

y = minus00037x4 + 00868x3 minus 06516x2

minus

y = minus00116x4 + 03419x3 minus 27611x2

y = minus00079x4 + 02758x3 minus 22402x2

minus

29826x minus 17137

44995x minus 17932

R2 = 099107

R2 = 09937

R2 = 098325

R2 = 0991

y = 00097x4 minus 026

minus 7316

82x3 + 20218x2

2x minus 2529

+ 88404x minus 10505



119910119896(119909) =

119873

sum

119895=1

1199081198961198950119895(119909) + 119908

1198960 (6)





0119895(119909) = exp(minus

1003817100381710038171003817119909 minus 1205831003817100381710038171003817

2

21205902

119895

) (7)





0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 093302

0 5 10 15minus10

0

10

20

30MSE = 877518

Error

Error


5

10

15

0 5 10 15minus100

minus80

minus60

minus40

minus20Train data

Outputs Targets

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

2 MSE = 96384e minus 29


1

2

3

4

1 2 3 4minus60

minus50

minus40

minus30

minus20Test data


minus50

minus40

minus30

minus20R = 063879

1 2 3 40

10

20

30 MSE = 3290692

Error

minus20 0 20 40 600

02040608

1

120583 = 41778

times10minus14

times10minus14


10274e minus 14

120583 = 156668

RMSE = 93676

RMSE = 181403

120590 = 86787

120590 =

120590 = 105591


1 2 3 4minus100

minus80

minus60

minus40

minus20 Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 088922

1 2 3 4minus40

minus20

0

20MSE = 2640048

Error


02040608

1RMSE = 162482

120583 = minus93368 120590 = 153549

0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 094015

0 5 10 15minus40

minus20

0

20MSE = 704013

Error


5

10

15

RMSE = 83905120583 = minus24898

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

4


Error

minus5 0 50

2

4

6times10minus14

times10minus14


120590 = 82939


Figure 7 Continued


1 2 3 4minus80

minus60

minus40

minus20Test data

Outputs Targets


minus60

minus40

minus20R = 052196

1 2 3 4minus50

minus40

minus30

minus20

minus10 MSE = 8757558

Error


02040608

1RMSE = 295932 120583 = minus274699

120590 = 127102

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 08234

0 5 10 15minus60

minus40

minus20

0

20 MSE = 2335349

Error

minus50 0 500

5

10

15

Outputs Targets

RMSE = 152818120583 = minus73253120590 = 138825

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20

0 5 10 15minus2

minus1

0

1

2 MSE = 97532e minus 29

Error


1

2

3

4R = 1

times10minus14

times10minus14



1 2 3 4minus100

minus80

minus60

minus40

minus20Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 080882

1 2 3 4minus60

minus40

minus20

0MSE = 4860424

Error


05

1

15

2

RMSE = 220464120583 = minus167689

120590 = 165267

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 091205

0 5 10 15minus60

minus40

minus20

0

20 MSE = 1296113

ErrorError

minus50 0 500

5

10

15

Outputs Targets

RMSE = 113847120583 = minus44717 120590 = 108372

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data


minus80

minus60

minus40

minus20

0 5 10 15minus4

minus2

0

2 MSE = 21113e minus 28

minus5 0 50

1

2

3

4R = 1

times10minus14

times10minus14






















1198772 RMSE 119877


1 099 098 099 099 069 078 069 063 093 072 078 072 136 198 198 1982 098 098 098 099 078 078 078 069 091 072 066 082 098 198 156 1903 097 098 099 099 079 059 069 069 080 069 067 072 090 102 098 1024 099 099 099 099 069 069 069 069 092 072 068 080 098 198 143 1025 099 098 098 099 069 079 059 059 088 070 078 086 099 136 198 1986 097 098 099 099 079 069 069 069 086 070 064 082 102 098 199 1987 099 098 099 099 069 069 069 069 093 068 077 082 136 136 136 0988 099 098 098 099 069 059 059 069 093 072 068 074 136 198 098 0999 097 099 099 098 079 059 069 089 091 072 076 074 098 198 143 10210 099 098 099 099 069 069 069 069 090 069 067 072 143 133 136 13611 099 098 099 099 069 069 069 069 072 072 068 070 135 198 136 19812 099 098 099 099 069 069 069 069 098 070 068 076 099 133 189 19813 099 099 099 098 059 069 059 080 096 070 064 072 133 136 098 18914 099 099 099 099 069 069 069 059 093 068 067 072 136 098 136 13615 099 098 099 098 069 069 069 079 093 072 058 072 136 198 098 13616 099 098 099 099 058 069 079 069 093 072 056 083 136 198 099 18917 099 098 099 099 069 059 069 069 090 069 085 082 189 189 102 09818 098 099 099 097 078 069 069 079 082 072 058 080 098 198 136 18919 099 099 097 099 069 069 069 069 088 070 058 086 099 136 136 09820 096 098 099 099 080 069 079 059 086 070 054 087 102 136 198 13621 099 098 099 099 069 069 069 069 093 068 057 082 136 098 198 09822 098 099 099 099 069 069 059 069 093 072 058 072 136 198 189 19823 099 099 099 098 069 069 069 079 091 073 056 087 098 098 098 19824 098 099 099 099 078 059 069 069 090 079 057 082 189 143 136 18925 097 098 099 099 079 069 069 079 091 072 058 080 098 198 098 18926 090 098 099 098 088 069 069 079 093 070 058 087 136 098 135 09827 099 099 097 099 069 079 079 069 092 070 054 088 098 136 098 13628 098 099 099 099 069 069 069 059 097 070 047 088 135 098 099 09829 099 099 099 098 069 069 069 079 095 072 058 082 101 198 102 13530 098 098 099 099 069 069 069 069 091 072 056 089 098 198 136 189










1 1199042 119904

119873 and a set of


119872 where 119873 is



119862119873


sum

119895=1

119902119895times 119882119895 (8)










MOPSO Parameters




w=05 Inertia Weight








mu=01 Mutation Rate

Initialization


for i=1nPop







end





for i=1numel(rep)


end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop







Apply Mutation










end



for e=1Extra


end

end



Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra










minus20minus27

minus30minus33

minus38minus45


minus60minus66

minus70

minus80

minus94minus98

minus140

minus120

minus100

minus80

minus60

minus40

minus20

00 5 10 15 20


minus40minus45

minus50

minus60

minus72

minus79minus80


minus100

minus90

minus80

minus70

minus60

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20

minus27minus31

minus40

minus60

minus68



minus82minus76

minus68

minus56

minus100

minus90

minus80

minus70

minus60


minus79

minus66

minus52minus46

minus39minus33

minus26

minus50

minus40

minus30

minus20

minus10

00 5 10 15 20 0 5 10 15 20

minus120

minus100

minus80

minus60

minus40

minus20

0

y = minus00037x4 + 00868x3 minus 06516x2

minus

y = minus00116x4 + 03419x3 minus 27611x2

y = minus00079x4 + 02758x3 minus 22402x2

minus

29826x minus 17137

44995x minus 17932

R2 = 099107

R2 = 09937

R2 = 098325

R2 = 0991

y = 00097x4 minus 026

minus 7316

82x3 + 20218x2

2x minus 2529

+ 88404x minus 10505



119910119896(119909) =

119873

sum

119895=1

1199081198961198950119895(119909) + 119908

1198960 (6)





0119895(119909) = exp(minus

1003817100381710038171003817119909 minus 1205831003817100381710038171003817

2

21205902

119895

) (7)





0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 093302

0 5 10 15minus10

0

10

20

30MSE = 877518

Error

Error


5

10

15

0 5 10 15minus100

minus80

minus60

minus40

minus20Train data

Outputs Targets

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

2 MSE = 96384e minus 29


1

2

3

4

1 2 3 4minus60

minus50

minus40

minus30

minus20Test data


minus50

minus40

minus30

minus20R = 063879

1 2 3 40

10

20

30 MSE = 3290692

Error

minus20 0 20 40 600

02040608

1

120583 = 41778

times10minus14

times10minus14


10274e minus 14

120583 = 156668

RMSE = 93676

RMSE = 181403

120590 = 86787

120590 =

120590 = 105591


1 2 3 4minus100

minus80

minus60

minus40

minus20 Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 088922

1 2 3 4minus40

minus20

0

20MSE = 2640048

Error


02040608

1RMSE = 162482

120583 = minus93368 120590 = 153549

0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 094015

0 5 10 15minus40

minus20

0

20MSE = 704013

Error


5

10

15

RMSE = 83905120583 = minus24898

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

4


Error

minus5 0 50

2

4

6times10minus14

times10minus14


120590 = 82939


Figure 7 Continued


1 2 3 4minus80

minus60

minus40

minus20Test data

Outputs Targets


minus60

minus40

minus20R = 052196

1 2 3 4minus50

minus40

minus30

minus20

minus10 MSE = 8757558

Error


02040608

1RMSE = 295932 120583 = minus274699

120590 = 127102

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 08234

0 5 10 15minus60

minus40

minus20

0

20 MSE = 2335349

Error

minus50 0 500

5

10

15

Outputs Targets

RMSE = 152818120583 = minus73253120590 = 138825

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20

0 5 10 15minus2

minus1

0

1

2 MSE = 97532e minus 29

Error


1

2

3

4R = 1

times10minus14

times10minus14



1 2 3 4minus100

minus80

minus60

minus40

minus20Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 080882

1 2 3 4minus60

minus40

minus20

0MSE = 4860424

Error


05

1

15

2

RMSE = 220464120583 = minus167689

120590 = 165267

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 091205

0 5 10 15minus60

minus40

minus20

0

20 MSE = 1296113

ErrorError

minus50 0 500

5

10

15

Outputs Targets

RMSE = 113847120583 = minus44717 120590 = 108372

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data


minus80

minus60

minus40

minus20

0 5 10 15minus4

minus2

0

2 MSE = 21113e minus 28

minus5 0 50

1

2

3

4R = 1

times10minus14

times10minus14






















1198772 RMSE 119877


1 099 098 099 099 069 078 069 063 093 072 078 072 136 198 198 1982 098 098 098 099 078 078 078 069 091 072 066 082 098 198 156 1903 097 098 099 099 079 059 069 069 080 069 067 072 090 102 098 1024 099 099 099 099 069 069 069 069 092 072 068 080 098 198 143 1025 099 098 098 099 069 079 059 059 088 070 078 086 099 136 198 1986 097 098 099 099 079 069 069 069 086 070 064 082 102 098 199 1987 099 098 099 099 069 069 069 069 093 068 077 082 136 136 136 0988 099 098 098 099 069 059 059 069 093 072 068 074 136 198 098 0999 097 099 099 098 079 059 069 089 091 072 076 074 098 198 143 10210 099 098 099 099 069 069 069 069 090 069 067 072 143 133 136 13611 099 098 099 099 069 069 069 069 072 072 068 070 135 198 136 19812 099 098 099 099 069 069 069 069 098 070 068 076 099 133 189 19813 099 099 099 098 059 069 059 080 096 070 064 072 133 136 098 18914 099 099 099 099 069 069 069 059 093 068 067 072 136 098 136 13615 099 098 099 098 069 069 069 079 093 072 058 072 136 198 098 13616 099 098 099 099 058 069 079 069 093 072 056 083 136 198 099 18917 099 098 099 099 069 059 069 069 090 069 085 082 189 189 102 09818 098 099 099 097 078 069 069 079 082 072 058 080 098 198 136 18919 099 099 097 099 069 069 069 069 088 070 058 086 099 136 136 09820 096 098 099 099 080 069 079 059 086 070 054 087 102 136 198 13621 099 098 099 099 069 069 069 069 093 068 057 082 136 098 198 09822 098 099 099 099 069 069 059 069 093 072 058 072 136 198 189 19823 099 099 099 098 069 069 069 079 091 073 056 087 098 098 098 19824 098 099 099 099 078 059 069 069 090 079 057 082 189 143 136 18925 097 098 099 099 079 069 069 079 091 072 058 080 098 198 098 18926 090 098 099 098 088 069 069 079 093 070 058 087 136 098 135 09827 099 099 097 099 069 079 079 069 092 070 054 088 098 136 098 13628 098 099 099 099 069 069 069 059 097 070 047 088 135 098 099 09829 099 099 099 098 069 069 069 079 095 072 058 082 101 198 102 13530 098 098 099 099 069 069 069 069 091 072 056 089 098 198 136 189










1 1199042 119904

119873 and a set of


119872 where 119873 is



119862119873


sum

119895=1

119902119895times 119882119895 (8)










MOPSO Parameters




w=05 Inertia Weight








mu=01 Mutation Rate

Initialization


for i=1nPop







end





for i=1numel(rep)


end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop







Apply Mutation










end



for e=1Extra


end

end



Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra










0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 093302

0 5 10 15minus10

0

10

20

30MSE = 877518

Error

Error


5

10

15

0 5 10 15minus100

minus80

minus60

minus40

minus20Train data

Outputs Targets

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

2 MSE = 96384e minus 29


1

2

3

4

1 2 3 4minus60

minus50

minus40

minus30

minus20Test data


minus50

minus40

minus30

minus20R = 063879

1 2 3 40

10

20

30 MSE = 3290692

Error

minus20 0 20 40 600

02040608

1

120583 = 41778

times10minus14

times10minus14


10274e minus 14

120583 = 156668

RMSE = 93676

RMSE = 181403

120590 = 86787

120590 =

120590 = 105591


1 2 3 4minus100

minus80

minus60

minus40

minus20 Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 088922

1 2 3 4minus40

minus20

0

20MSE = 2640048

Error


02040608

1RMSE = 162482

120583 = minus93368 120590 = 153549

0 5 10 15minus100

minus80

minus60

minus40

minus20All data

Outputs Targets


minus80

minus60

minus40

minus20R = 094015

0 5 10 15minus40

minus20

0

20MSE = 704013

Error


5

10

15

RMSE = 83905120583 = minus24898

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20R = 1

0 5 10 15minus4

minus2

0

4


Error

minus5 0 50

2

4

6times10minus14

times10minus14


120590 = 82939


Figure 7 Continued


1 2 3 4minus80

minus60

minus40

minus20Test data

Outputs Targets


minus60

minus40

minus20R = 052196

1 2 3 4minus50

minus40

minus30

minus20

minus10 MSE = 8757558

Error


02040608

1RMSE = 295932 120583 = minus274699

120590 = 127102

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 08234

0 5 10 15minus60

minus40

minus20

0

20 MSE = 2335349

Error

minus50 0 500

5

10

15

Outputs Targets

RMSE = 152818120583 = minus73253120590 = 138825

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20

0 5 10 15minus2

minus1

0

1

2 MSE = 97532e minus 29

Error


1

2

3

4R = 1

times10minus14

times10minus14



1 2 3 4minus100

minus80

minus60

minus40

minus20Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 080882

1 2 3 4minus60

minus40

minus20

0MSE = 4860424

Error


05

1

15

2

RMSE = 220464120583 = minus167689

120590 = 165267

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 091205

0 5 10 15minus60

minus40

minus20

0

20 MSE = 1296113

ErrorError

minus50 0 500

5

10

15

Outputs Targets

RMSE = 113847120583 = minus44717 120590 = 108372

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data


minus80

minus60

minus40

minus20

0 5 10 15minus4

minus2

0

2 MSE = 21113e minus 28

minus5 0 50

1

2

3

4R = 1

times10minus14

times10minus14






















1198772 RMSE 119877


1 099 098 099 099 069 078 069 063 093 072 078 072 136 198 198 1982 098 098 098 099 078 078 078 069 091 072 066 082 098 198 156 1903 097 098 099 099 079 059 069 069 080 069 067 072 090 102 098 1024 099 099 099 099 069 069 069 069 092 072 068 080 098 198 143 1025 099 098 098 099 069 079 059 059 088 070 078 086 099 136 198 1986 097 098 099 099 079 069 069 069 086 070 064 082 102 098 199 1987 099 098 099 099 069 069 069 069 093 068 077 082 136 136 136 0988 099 098 098 099 069 059 059 069 093 072 068 074 136 198 098 0999 097 099 099 098 079 059 069 089 091 072 076 074 098 198 143 10210 099 098 099 099 069 069 069 069 090 069 067 072 143 133 136 13611 099 098 099 099 069 069 069 069 072 072 068 070 135 198 136 19812 099 098 099 099 069 069 069 069 098 070 068 076 099 133 189 19813 099 099 099 098 059 069 059 080 096 070 064 072 133 136 098 18914 099 099 099 099 069 069 069 059 093 068 067 072 136 098 136 13615 099 098 099 098 069 069 069 079 093 072 058 072 136 198 098 13616 099 098 099 099 058 069 079 069 093 072 056 083 136 198 099 18917 099 098 099 099 069 059 069 069 090 069 085 082 189 189 102 09818 098 099 099 097 078 069 069 079 082 072 058 080 098 198 136 18919 099 099 097 099 069 069 069 069 088 070 058 086 099 136 136 09820 096 098 099 099 080 069 079 059 086 070 054 087 102 136 198 13621 099 098 099 099 069 069 069 069 093 068 057 082 136 098 198 09822 098 099 099 099 069 069 059 069 093 072 058 072 136 198 189 19823 099 099 099 098 069 069 069 079 091 073 056 087 098 098 098 19824 098 099 099 099 078 059 069 069 090 079 057 082 189 143 136 18925 097 098 099 099 079 069 069 079 091 072 058 080 098 198 098 18926 090 098 099 098 088 069 069 079 093 070 058 087 136 098 135 09827 099 099 097 099 069 079 079 069 092 070 054 088 098 136 098 13628 098 099 099 099 069 069 069 059 097 070 047 088 135 098 099 09829 099 099 099 098 069 069 069 079 095 072 058 082 101 198 102 13530 098 098 099 099 069 069 069 069 091 072 056 089 098 198 136 189










1 1199042 119904

119873 and a set of


119872 where 119873 is



119862119873


sum

119895=1

119902119895times 119882119895 (8)










MOPSO Parameters




w=05 Inertia Weight








mu=01 Mutation Rate

Initialization


for i=1nPop







end





for i=1numel(rep)


end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop







Apply Mutation










end



for e=1Extra


end

end



Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra










1 2 3 4minus80

minus60

minus40

minus20Test data

Outputs Targets


minus60

minus40

minus20R = 052196

1 2 3 4minus50

minus40

minus30

minus20

minus10 MSE = 8757558

Error


02040608

1RMSE = 295932 120583 = minus274699

120590 = 127102

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 08234

0 5 10 15minus60

minus40

minus20

0

20 MSE = 2335349

Error

minus50 0 500

5

10

15

Outputs Targets

RMSE = 152818120583 = minus73253120590 = 138825

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data

Outputs Targets


minus80

minus60

minus40

minus20

0 5 10 15minus2

minus1

0

1

2 MSE = 97532e minus 29

Error


1

2

3

4R = 1

times10minus14

times10minus14



1 2 3 4minus100

minus80

minus60

minus40

minus20Test data

Outputs Targets


minus80

minus60

minus40

minus20R = 080882

1 2 3 4minus60

minus40

minus20

0MSE = 4860424

Error


05

1

15

2

RMSE = 220464120583 = minus167689

120590 = 165267

0 5 10 15minus100

minus80

minus60

minus40

minus20All data


minus80

minus60

minus40

minus20R = 091205

0 5 10 15minus60

minus40

minus20

0

20 MSE = 1296113

ErrorError

minus50 0 500

5

10

15

Outputs Targets

RMSE = 113847120583 = minus44717 120590 = 108372

0 5 10 15minus100

minus80

minus60

minus40

minus20 Train data


minus80

minus60

minus40

minus20

0 5 10 15minus4

minus2

0

2 MSE = 21113e minus 28

minus5 0 50

1

2

3

4R = 1

times10minus14

times10minus14






















1198772 RMSE 119877


1 099 098 099 099 069 078 069 063 093 072 078 072 136 198 198 1982 098 098 098 099 078 078 078 069 091 072 066 082 098 198 156 1903 097 098 099 099 079 059 069 069 080 069 067 072 090 102 098 1024 099 099 099 099 069 069 069 069 092 072 068 080 098 198 143 1025 099 098 098 099 069 079 059 059 088 070 078 086 099 136 198 1986 097 098 099 099 079 069 069 069 086 070 064 082 102 098 199 1987 099 098 099 099 069 069 069 069 093 068 077 082 136 136 136 0988 099 098 098 099 069 059 059 069 093 072 068 074 136 198 098 0999 097 099 099 098 079 059 069 089 091 072 076 074 098 198 143 10210 099 098 099 099 069 069 069 069 090 069 067 072 143 133 136 13611 099 098 099 099 069 069 069 069 072 072 068 070 135 198 136 19812 099 098 099 099 069 069 069 069 098 070 068 076 099 133 189 19813 099 099 099 098 059 069 059 080 096 070 064 072 133 136 098 18914 099 099 099 099 069 069 069 059 093 068 067 072 136 098 136 13615 099 098 099 098 069 069 069 079 093 072 058 072 136 198 098 13616 099 098 099 099 058 069 079 069 093 072 056 083 136 198 099 18917 099 098 099 099 069 059 069 069 090 069 085 082 189 189 102 09818 098 099 099 097 078 069 069 079 082 072 058 080 098 198 136 18919 099 099 097 099 069 069 069 069 088 070 058 086 099 136 136 09820 096 098 099 099 080 069 079 059 086 070 054 087 102 136 198 13621 099 098 099 099 069 069 069 069 093 068 057 082 136 098 198 09822 098 099 099 099 069 069 059 069 093 072 058 072 136 198 189 19823 099 099 099 098 069 069 069 079 091 073 056 087 098 098 098 19824 098 099 099 099 078 059 069 069 090 079 057 082 189 143 136 18925 097 098 099 099 079 069 069 079 091 072 058 080 098 198 098 18926 090 098 099 098 088 069 069 079 093 070 058 087 136 098 135 09827 099 099 097 099 069 079 079 069 092 070 054 088 098 136 098 13628 098 099 099 099 069 069 069 059 097 070 047 088 135 098 099 09829 099 099 099 098 069 069 069 079 095 072 058 082 101 198 102 13530 098 098 099 099 069 069 069 069 091 072 056 089 098 198 136 189










1 1199042 119904

119873 and a set of


119872 where 119873 is



119862119873


sum

119895=1

119902119895times 119882119895 (8)










MOPSO Parameters




w=05 Inertia Weight








mu=01 Mutation Rate

Initialization


for i=1nPop







end





for i=1numel(rep)


end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop







Apply Mutation










end



for e=1Extra


end

end



Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra


























1198772 RMSE 119877


1 099 098 099 099 069 078 069 063 093 072 078 072 136 198 198 1982 098 098 098 099 078 078 078 069 091 072 066 082 098 198 156 1903 097 098 099 099 079 059 069 069 080 069 067 072 090 102 098 1024 099 099 099 099 069 069 069 069 092 072 068 080 098 198 143 1025 099 098 098 099 069 079 059 059 088 070 078 086 099 136 198 1986 097 098 099 099 079 069 069 069 086 070 064 082 102 098 199 1987 099 098 099 099 069 069 069 069 093 068 077 082 136 136 136 0988 099 098 098 099 069 059 059 069 093 072 068 074 136 198 098 0999 097 099 099 098 079 059 069 089 091 072 076 074 098 198 143 10210 099 098 099 099 069 069 069 069 090 069 067 072 143 133 136 13611 099 098 099 099 069 069 069 069 072 072 068 070 135 198 136 19812 099 098 099 099 069 069 069 069 098 070 068 076 099 133 189 19813 099 099 099 098 059 069 059 080 096 070 064 072 133 136 098 18914 099 099 099 099 069 069 069 059 093 068 067 072 136 098 136 13615 099 098 099 098 069 069 069 079 093 072 058 072 136 198 098 13616 099 098 099 099 058 069 079 069 093 072 056 083 136 198 099 18917 099 098 099 099 069 059 069 069 090 069 085 082 189 189 102 09818 098 099 099 097 078 069 069 079 082 072 058 080 098 198 136 18919 099 099 097 099 069 069 069 069 088 070 058 086 099 136 136 09820 096 098 099 099 080 069 079 059 086 070 054 087 102 136 198 13621 099 098 099 099 069 069 069 069 093 068 057 082 136 098 198 09822 098 099 099 099 069 069 059 069 093 072 058 072 136 198 189 19823 099 099 099 098 069 069 069 079 091 073 056 087 098 098 098 19824 098 099 099 099 078 059 069 069 090 079 057 082 189 143 136 18925 097 098 099 099 079 069 069 079 091 072 058 080 098 198 098 18926 090 098 099 098 088 069 069 079 093 070 058 087 136 098 135 09827 099 099 097 099 069 079 079 069 092 070 054 088 098 136 098 13628 098 099 099 099 069 069 069 059 097 070 047 088 135 098 099 09829 099 099 099 098 069 069 069 079 095 072 058 082 101 198 102 13530 098 098 099 099 069 069 069 069 091 072 056 089 098 198 136 189










1 1199042 119904

119873 and a set of


119872 where 119873 is



119862119873


sum

119895=1

119902119895times 119882119895 (8)










MOPSO Parameters




w=05 Inertia Weight








mu=01 Mutation Rate

Initialization


for i=1nPop







end





for i=1numel(rep)


end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop







Apply Mutation










end



for e=1Extra


end

end



Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra













1 1199042 119904

119873 and a set of


119872 where 119873 is



119862119873


sum

119895=1

119902119895times 119882119895 (8)










MOPSO Parameters




w=05 Inertia Weight








mu=01 Mutation Rate

Initialization


for i=1nPop







end





for i=1numel(rep)


end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop







Apply Mutation










end



for e=1Extra


end

end



Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra










MOPSO Parameters




w=05 Inertia Weight








mu=01 Mutation Rate

Initialization


for i=1nPop







end





for i=1numel(rep)


end

MOPSO Main Loop

for it=1MaxIt

for i=1nPop







Apply Mutation










end



for e=1Extra


end

end



Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra










Core network

UMTS

WLAN1 WLAN2

AP1AP2

Server

Node B









6 Conclusion



minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra










minus20

minus60

minus40

minus80

minus100

minus120

RSS

(dBm

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

(a)

ABR

(bps

)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

1000

1500

2000

500

0

(b)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

100

80

60

40

20

SNR

(dB)

(c)

Distance (m)0 200 400 600 800 1000 1200 1400

UMTSWLAN1

WLAN2

Thro

ughp

ut (b

ps)

500

400

300

200

100

0

(d)


2(1 + SNR) 119894 = 1 2 3 [22] and SNR SNR

119894= 119864119887119873119900








01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra










01020304050607080

6 8 10 12 14 16

Num

ber o

f ver

tical

han

dove

rs

Arival rate



0

1

2

3

0 200 400 600 800 1000 1200 1400

Net

wor

k

Distance


Acknowledgments


References





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra



























Volume 2014




Journal of











Journal of


Function Spaces






Algebra









Research Article A Novel Model on Curve Fitting and ...

Documents

Transcript of Research Article A Novel Model on Curve Fitting and ...