Research Article Compensatory Analysis and …downloads.hindawi.com/journals/jece/2016/7539454.pdfResearch Article Compensatory Analysis and Optimization for MADM for Heterogeneous

Research ArticleCompensatory Analysis and Optimization for MADM forHeterogeneous Wireless Network Selection

Jian Zhou12 and Can-yan Zhu1

1 Institute of Intelligent Structure and System Soochow University Soochow 215006 China2Department of Information Engineering Suzhou Global Institute of Software Technology Soochow 215163 China

Correspondence should be addressed to Can-yan Zhu qiwuzhusudaeducn

Received 3 February 2016 Accepted 5 May 2016

Academic Editor Rajesh Khanna

Copyright copy 2016 J Zhou and C-y Zhu 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

In the next-generation heterogeneous wireless networks a mobile terminal with a multi-interface may have network access fromdifferent service providers using various technologies In spite of this heterogeneity seamless intersystem mobility is a mandatoryrequirement One of the major challenges for seamless mobility is the creation of a network selection scheme which is for usersthat select an optimal network with best comprehensive performance between different types of networks However the optimalnetwork may be not the most reasonable one due to compensation of MADM (Multiple Attribute Decision Making) and thenetwork is called pseudo-optimal network This paper conducts a performance evaluation of a number of widely used MADM-based methods for network selection that aim to keep the mobile users always best connected anywhere and anytime wheresubjective weight and objective weight are all considered The performance analysis shows that the selection scheme based onMEW (weightedmultiplicative method) and combination weight can better avoid accessing pseudo-optimal network for balancingnetwork load and reducing ping-pong effect in comparison with three other MADM solutions

1 Introduction

With the emerging and development of all kinds of wirelessaccess technology including 2G 3G WLANWiMax (WorldInteroperability for Microwave Access) and MBWA (MobileBroadband Wireless Access) wireless networks overlap andcomplement each other forming a hybrid wireless net-work called heterogeneous wireless networks [1] To supportseamless mobility while a mobile station roams within aheterogeneous wireless network VHO (Vertical Handoff)necessity estimation and decision to select a best targetnetwork are two important aspects of the overall mobilityframework The handoff necessity estimation is importantin order to keep the unnecessary handoffs and their failuresat a low level On the other hand to maximize the end-usersrsquo satisfaction level the decision to select the best networkamong other available candidates plays an important role aswell The network selection process consists of three majorsubservices (1) network monitoring monitors the currentnetwork conditions (network availability signal strength

current call connection etc) and provides the data gatheredtogether with information related to the user preferencescurrent running applications on the userrsquos mobile deviceand their QoS requirements (2) handover decision handlesthe network selection process (which ranks the candidatenetworks and selects the best target) and is initiated eitherby an automatic trigger for handover for an existing callconnection or by a request for a new connection on themobile device and (3) handover execution once a new targetnetwork is selected the connection is set up on the targetcandidate network (and the old connection torn down)

Network selection algorithm has become amore complexproblem and combines multiple systemsrsquo attributes to choosethe target network that offers the highest overall performanceThis approach is considered optimal as compared to theother traditional approaches that rely on a single systemrsquosattributes like RSS (Received Signal Strength) or availablebandwidth tomake handoff decisions As all of these parame-ters present different ranges and units of measurements theyneed to be normalized in order to make them comparable

Hindawi Publishing CorporationJournal of Electrical and Computer EngineeringVolume 2016 Article ID 7539454 9 pageshttpdxdoiorg10115520167539454

2 Journal of Electrical and Computer Engineering

Utility functions are used for normalization to map all theparameters into dimensionless units within the range [0 1][2ndash7] This normalized information is then used in thedecision making process in order to compute a ranked list ofthe best available network choices MADM including SAWMEW GRA and TOPSIS is widely used as score functionmethods for network selection [8ndash15] User or networkoperator preferences for the main trade-off criteria can berepresented by the use of different weights in weighted scorefunctions The candidate network with the highest score isselected as the target network if that differs from the currentnetwork connection (or it is for a new connection) it promptshandover execution (or new network connection setup)

However the optimal network with best comprehensiveperformancemay not be themost reasonable one due to com-pensation of MADM we call the network pseudo-optimalnetwork For example if network with best comprehensiveperformance and heavy load is selected which may furtheraggravate network congestion end-user cannot enjoy goodnetwork quality Moreover it is argued that an appropriateMADM should not make end-user access pseudo-optimalnetwork in [10] Hence performance of accessing pseudo-optimal network is firstly analyzed for SAW (Simple AdditiveWeighting Method) TOPSIS (Technique for Order Prefer-ence by Similarity to Ideal Solution) GRA (Grey RelationalAnalysis) and MEW then network selection based on MEWand combinational weight is proposed It can be seen fromsimulation that the proposed algorithm can make end-userbetter avoid assessing pseudo-optimal network and has betterperformance in network load balance and reducing ping-pingeffect

2 Compensatory Analysis for MADM

21 Common MADM MADM algorithms can be dividedinto compensatory and noncompensatory ones Noncom-pensatory algorithms are used to find acceptable alterna-tives which satisfy the minimum cutoff On the contrarycompensatory algorithms combinemultiple attributes to findthe best alternative Most MADM algorithms that have beenstudied for the network selection problem are compensatoryalgorithms

SAW is widely used by most studies of the networkselection problem using cost or utility functions generallygiven by

119880SAW =

119899

sum119895=1

119908119895119906119894119895 (1)

where 119908119895represents the weight of attribute 119888

119895and 119906

119894119895repre-

sents the adjusted value of attribute 119888119895of the network 119903

119894

MEW is to calculate the coefficient by multiplicativeoperation given by

119880MEW =

119899

prod119895=1

119906119894119895

119908119895 (2)

Other twoMADM algorithms used for network selection areTOPSIS and GRA which both consider the distance from

Table 1 Network selection based on SAW

Utility 119908119895

Network A Network B119880(119862) 13 05 08119880(119864) 13 05 08119880(119871) 13 05 005Aggregate utility 05 055

the evaluated network to one or multiple reference networksCoefficient of TOPSIS can be calculated as

119880TOPSIS =119863120572

119863120572 + 119863120573 (3)

where119863120572 and119863120573 represent the Euclidean distances from thecurrent network to the worst and best reference networksrespectively given by

119863120572

= radic

119899

sum119895=1

1199082119895(119906119894119895minus V120572119895)2

(4)

119863120573

= radic

119899

sum119895=1

1199082119895(119906119894119895minus V120573119895)2

(5)

where V120572119895and V120573

119895represent the values of attribute 119888

119895of the

worst and best reference networks respectivelyDifferent from TOPSIS GRA uses only the best reference

network to calculate the coefficient given by

119880GRA =1

sum119899

119895=1119908119895

100381610038161003816100381610038161003816119906119894119895minus V120573119895

100381610038161003816100381610038161003816+ 1

(6)

22 Compensatory Analysis According to the principle ofnetwork selection the optimal network is the network withbest comprehensive performance but it may be not the mostreasonable one due to compensation ofMADM For examplethe result is the fact that end-user chooses best network fromnetworksA andB by SAWas shown inTable 1 where decisionattributes are cost (119862) power consumption (119864) and load (119871)Assume that only subjective weight is considered

Table 1 shows that the comprehensive performance ofnetwork B is superior to network A and end-user will choosenetwork B as access networkHowever load utility of networkB in Table 1 is 005 which is close to 0 namely networkB is not suitable for new access due to its heavy load andotherwise it may lead to congestion for network B and notbalance load between networksA andB So networkA shouldbe the reasonable choice for the above situation But end-user chooses network B due to performance compensationbetween attributes which means the excellent performanceof cost and power consumption compensates the bad per-formance of load in network B and we call the network likenetwork B pseudo-optimal network

The results that end-user chooses best network fromnetworks A and B by TOPSIS and GRA are shown in Tables2 and 3 respectively It can be seen from Tables 2 and 3 that

Journal of Electrical and Computer Engineering 3

Table 2 Network selection based on TOPSIS

Utility 119908119895


Table 3 Network selection based on GRA

Utility 119908119895


Table 4 Network selection based on MEW (119908119897= 13)

Utility 119908119895



Utility 119908119895


TOPSIS andGRA also choose network Bwith heavy load likeSAW and they have the same limitation

The result that end-user chooses best network fromnetworks A and B by MEW is shown in Table 4 It can beseen from Table 4 unlike SAW TOPSIS and GRA that end-user chooses network A with light load as access networkHowever if we readjust attribute weight in Table 4 andreestimate the comprehensive performance of networks Aand B by MEW as shown in Table 5 it can be seen fromTable 5 that the comprehensive performance of network B issuperior to network A again end-user will choose network Bas access network Obviously MEW cannot ensure that end-user can always avoid accessing the pseudo-optimal network

The reason that end-user chooses optimal network fromnetwork A to network B when attribute weight of load isfrom one-third to 005 is shown in Figure 1 In Figure 1the expression of two curves is 05119908 and 005119908 respectivelydifference of utility value between two curves is largerwhen119908is larger and difference is close to 0 when119908 is smaller hencethe comprehensive performance of network Bwill be reducedwell due to multiplicative features of MEM and network A isselected as optimal network when difference of load utilitybetween networks A and B is larger because weight of load isone-third However network B is selected as optimal networkagain when difference of load utility between networks A and

U(L) = 05U(L) = 005

0

01

02

03

04

05

06

07

08

09

1

U(L)w

09 08 07 06 05 04 03 01 02 01w

Figure 1 Performance comparison of load with different weight

B is smaller because weight of load is 005 and network Awith light load is not selected Through the above analysisto make end-user avoid accessing the network with attributewith poor performance weight of the attribute should beadjusted in real-time to prevent its value frombeing too smallandMEWshould be used as decisionmakingmethod to rankalternative network

3 Network Selection Based on MEW andCombination Weight

As described in Section 2 attribute weight should be adjustedin real-time to make end-user avoid accessing the pseudo-optimal network while objective weight method can calcu-late attribute weight according to attribute value of alternativenetwork and recalculate attribute weight when attribute datachanges However subjective weight must also be consideredto reflect experience and subjective importance for attributeof decision makers Hence combination weight which inte-grates the subjective weight and objective weight will beconsidered The steps for calculating combination weight areas follows

Step 1 Objective weight is determined by entropy weightingmethod calculated as follows

(1) Construct normalized matrix R given by

R =

[[[[[[[[[[[

[

11990611

sdot sdot sdot 1199061119895


sdot sdot sdot

1199061198941

sdot sdot sdot 119906119894119895

sdot sdot sdot 119906119894119899

sdot sdot sdot

1199061198981

sdot sdot sdot 119906119898119895

sdot sdot sdot 119906119898119899

]]]]]]]]]]]

]

(7)


Weight 1

WLANUMTSWiMax

05

06

07

08

09

1N

etw

ork

load

40 60 8020Number of new end-users

Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

Figure 2 Change of network load by SAW

(2) Calculate information entropy of attribute 119888119895 given by

119864119895=

minus1

ln119898

119898

sum119894=1

119906119894119895ln 119906119894119895 (8)

(3) Calculate objective weight of attribute 119888119895 given by

119908119900119895=

1 minus 119864119895

sum119899

119896=1(1 minus 119864

119896) (9)

Step 2 Subjective weight is determined by experience andassigned directly in this paper denoted by 119908

119904119895

Step 3 For considering subjective weight and objectiveweight combination weight can be expressed as

119908119888119895= 120572119908119900119895+ 120573119908119904119895 (10)

where 120572 and 120573meet 120572 + 120573 = 1 and 120572 120573 ge 0

Considering that weighted attribute values determined bysubjective weight and objective weight should be consistentoptimal mathematical model can be constructed to solve 120572and120573 Deviation degree of subjective andobjective evaluationvalue of alternative networks is given by

119889119894=

119899

sum119895=1

(120572119906119894119895119908119900119895minus 120573119906119894119895119908119904119895)2

(11)

The smaller the value of 119889119894is the more consistent the sub-

jective and objective evaluations tend to be Hence optimalmathematical model can be constructed as follows

min119898

sum119894=1

119889119894=

119898

sum119894=1

119899

sum119895=1

(120572119906119894119895119908119904119895minus 120573119906119894119895119908119900119895)2

st 120572 + 120573 = 1

120572 120573 ge 0

(12)


Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax

Figure 3 Change of network load by TOPSIS

and its solution can be obtained as

120572 =sum119898

119894=1sum119899

119895=11199062

119894119895119908119900119895(119908119904119895+ 119908119900119895)

sum119898

119894=1sum119899

119895=11199062119894119895(119908119904119895+ 119908119900119895)2

120573 =sum119898

119894=1sum119899

119895=11199062

119894119895119908119904119895(119908119904119895+ 119908119900119895)

sum119898

119894=1sum119899

119895=11199062119894119895(119908119904119895+ 119908119900119895)2

(13)

Substitute (13) into (10) combination weight can be deter-mined

Step 4 Substitute 119908119888119895into (2) rank for alternative networks

can be obtained by MEW

4 Simulation and Analysis

In this section two groups of simulations are used to validateperformance of the proposed algorithm one simulation isfor performance evaluation of network load balance and the

Table 6 Measurement value of attribute and parameter setting forutility function

Attribute WLAN UMTS WiMax 119909120572

119909120573

119862 (centMb) 10 50 30 0 100119864 (w) 2 6 3 0 10119871 () 80 60 70 0 100119863 (kbps) 220 400 350 200 800Notes 119909

120572and 119909120573are lower limit and upper limit of linear utility function

other is for performance evaluation of reducing ping-pongeffect

In simulation environment three networks WLANUMTS and WiMax are selected as alternative networksand cost power consumption load and data rate (B) aredecision attribute used for wireless network selection Linearutility function is adopted as utility function for all attributesMeasurement value of attribute and parameter setting forutility function is shown in Table 6 and measurement


Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax

Figure 4 Change of network load by GRA

Table 7 Utility used for load balance simulation

Utility WALN UMTS WiMax119880(119862) 09 05 07119880(119864) 08 04 07119880(119871) 02 04 03

values of load and data rate will change after beginning ofsimulation

41 Simulation for Performance of Network Load BalanceAccording to Table 6 price power consumption and load areselected as decision attributes utility and weight of decisionattributes are shown in Tables 7 and 8 respectively

Assume that there are new requests coming constantly inoverlapping area of WLAN UMTS and WiMax and end-user accesses the optimal network based on network selectionalgorithm Then Figures 2ndash5 show change of network load

Table 8 Weight for network attributes

Weight (119862 119864 119871)Weight 1 Only subjective weight (13 13 13)Weight 2 Only subjective weight (05 045 005)

Weight 3Combination of subjective weight (1313 13) and objective weightcomputed by (9)


of alternative network when end-user selects optimal net-work by SAW TOPSIS GRA and MEW respectively whereattribute weight is in case of weight 1 weight 2 weight 3 andweight 4 respectively In Figures 2ndash5 there are a number ofnew end-users on horizontal coordinates and network loadon vertical coordinates


Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax

Figure 5 Change of network load by MEW

Figure 2 shows change of network load when optimalnetwork is selected by SAW The order of optimal networkselected by new end-users is much the same for four weightcases It is failed to balance network load well and it is easyto cause congestion for network with heavy load so it cannotprovide good quality for end-user

Changes of network load when optimal network isselected by TOPSIS and GRA respectively are shown inFigures 3 and 4 where the performance is like SAW

Figure 5 shows change of network load when optimalnetwork is selected by MEW It can be seen from Figure 5that network load can be balanced better by MEW whenattribute weight is in case of weight 1 weight 3 and weight4 Hence no matter what value of subject weight is end-usercan avoid better accessing of the pseudo-optimal network andbalance network load between three alternative networks bythe scheme based on MEW and combination weight

42 Simulation for Performance of Ping-Pong Effect Accord-ing to Table 6 price power consumption and data rate are

Table 9 Utility used for ping-pong effect simulation


selected as decision attributes utility of decision attributes isshown in Table 9 and weight of decision attributes is as inTable 8

Assume that data rate will vary by around 50 kbps afterstart of simulation Figures 6ndash9 show performance of ping-pong effect of alternative networkwhen end-user selects opti-mal network by SAW TOPSIS GRA andMEW respectivelywhere attribute weight is in case of weight 1 weight 2 weight3 and weight 4 respectively In Figures 6ndash9 there are anumber of data rate fluctuations on horizontal coordinatesand a number of network handoffs on vertical coordinates


SAWGRA

TOPSISMEW

10 20 30 40 50 60 70 80 90 1000Number of data rate fluctuations

0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s

Figure 6 Network handoffs in case of weight 1

SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s


It can be seen from Figures 6ndash9 that network handoffis more frequent when optimal network is selected by SAWTOPSIS and GRA because data rate ofWLANmay be lowerthan the minimum requirements of end-user for data ratefluctuation and end-user must select other eligible networksHowever there are fewer network handoffs when optimalnetwork selected by MEW is in case of weight 1 weight 3and weight 4 Hence no matter what value of subject weightis end-user can avoid better accessing of the pseudo-optimalnetwork and reduce network handoffs between alternativenetworks by the scheme based on MEW and combinationweight which is a better solution for ping-pong effect causedby data rate fluctuations

43 Analysis and Comparison of Algorithm Complexity Inthis section network selection algorithm based on SAW

SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50N

umbe

r of n

etw

ork

hand

offs


Table 10 Comparison of computational overhead

SAW TOPSIS GRA MEWWeight 3 1471 us 1562 us 1521 us 1575 usWeight 4 1470 us 1560 us 1522 us 1578 us

TOPSIS GRA and MEW is implemented on DSP devicerespectively and computational overhead of network selec-tion process is computed In Table 10 mean computationaloverhead of one network selection process is showed whenthree networks are alternative and three decision attributesare chosen

It can be seen from Table 10 that computational overheadof MEW is about 7 08 and 3more than SAW TOPSISand GRA respectively but the performance of balancingnetwork load and reducing ping-pong effect ofMEW is vastlybetter than the others


5 Conclusions

In this paper network selection algorithm based on MEWand combination weight is proposed for the problem thatpseudo-optimal network may be chosen because of com-pensation of attribute performance when optimal networkis selected by SAW TOPSIS and GRA Simulation showsthat the proposed algorithm not only makes end-user avoidaccessing pseudo-optimal network but also balances net-work load to prevent network congestion and reduce ping-pong effect which improves system performance

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work is financially supported by NSFC no 61071214

References

[1] M Corici J Fiedler T Magedanz and D Vingarzan ldquoAccessnetwork discovery and selection in the future wireless commu-nicationrdquo Mobile Networks and Applications vol 16 no 3 pp337ndash349 2011

[2] Q-T Nguyen-Vuong G D Yacine and N Agoulmine ldquoOnutility models for access network selection in wireless heteroge-neous networksrdquo in Proceedings of the IEEE Network OperationsandManagement Symposium (NOMS rsquo08) pp 44ndash151 SalvadorBrazil April 2008

[3] I Chamodrakas and D Martakos ldquoA utility-based fuzzy TOP-SIS method for energy efficient network selection in heteroge-neous wireless networksrdquo Applied Soft Computing Journal vol12 no 7 pp 1929ndash1938 2012

[4] L Wang and G-S G S Kuo ldquoMathematical modeling for net-work selection in heterogeneous wireless networksmdasha tutorialrdquoIEEE Communications Surveys and Tutorials vol 15 no 1 pp271ndash292 2013

[5] R Trestian O Ormond and G-M Muntean ldquoEnhancedpower-friendly access network selection strategy for multi-media delivery over heterogeneous wireless networksrdquo IEEETransactions on Broadcasting vol 60 no 1 pp 85ndash101 2014

[6] P Kosmides A Rouskas and M Anagnostou ldquoUtility-basedRAT selection optimization in heterogeneous wireless net-worksrdquo Pervasive and Mobile Computing vol 12 no 6 pp 92ndash111 2014

[7] R Chai H Zhang X Dong Q Chen and T SvenssonldquoOptimal joint utility based load balancing algorithm forheterogeneous wireless networksrdquo Wireless Networks vol 20no 6 pp 1557ndash1571 2014

[8] J Liu and X-N Li ldquoHandover algorithm for WLANcellularnetworks with analytic hierarchy processrdquo Journal on Commu-nications vol 34 no 2 pp 65ndash72 2013

[9] S Zhang and Q Zhu ldquoHeterogeneous wireless network selec-tion algorithm based on group decisionrdquo Journal of ChinaUniversities of Posts and Telecommunications vol 21 no 3 pp1ndash9 2014

[10] Q-T Nguyen-Vuong N Agoulmine E H Cherkaoui and LToni ldquoMulticriteria optimization of access selection to improve

the quality of experience in heterogeneous wireless accessnetworksrdquo IEEE Transactions on Vehicular Technology vol 62no 4 pp 1785ndash1800 2013

[11] R Trestian O Ormond and G-M Muntean ldquoPerformanceevaluation of MADM-based methods for network selection ina multimedia wireless environmentrdquoWireless Networks vol 21no 5 pp 1745ndash1763 2014

[12] L Xu and Y Li ldquoA network selection scheme based on topsisin heterogeneous network environmentrdquo Journal of HarbinInstitute of Technology vol 21 no 1 pp 43ndash48 2014

[13] M Drissi and M Oumsis ldquoPerformance evaluation of multi-criteria vertical handover for heterogeneous wireless networksrdquoin Proceedings of the 1st International Conference on IntelligentSystems and Computer Vision (ISCV rsquo15) pp 1ndash5 IEEE FezMorocco March 2015

[14] R Verma and N P Singh ldquoGRA based network selection inheterogeneous wireless networksrdquoWireless Personal Communi-cations vol 72 no 2 pp 1437ndash1452 2013

[15] N P Singh and B Singh ldquoVertical handoff decision in 4G wire-less networks using multi attribute decision making approachrdquoWireless Networks vol 20 no 5 pp 1203ndash1211 2014

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



Utility functions are used for normalization to map all theparameters into dimensionless units within the range [0 1][2ndash7] This normalized information is then used in thedecision making process in order to compute a ranked list ofthe best available network choices MADM including SAWMEW GRA and TOPSIS is widely used as score functionmethods for network selection [8ndash15] User or networkoperator preferences for the main trade-off criteria can berepresented by the use of different weights in weighted scorefunctions The candidate network with the highest score isselected as the target network if that differs from the currentnetwork connection (or it is for a new connection) it promptshandover execution (or new network connection setup)

However the optimal network with best comprehensiveperformancemay not be themost reasonable one due to com-pensation of MADM we call the network pseudo-optimalnetwork For example if network with best comprehensiveperformance and heavy load is selected which may furtheraggravate network congestion end-user cannot enjoy goodnetwork quality Moreover it is argued that an appropriateMADM should not make end-user access pseudo-optimalnetwork in [10] Hence performance of accessing pseudo-optimal network is firstly analyzed for SAW (Simple AdditiveWeighting Method) TOPSIS (Technique for Order Prefer-ence by Similarity to Ideal Solution) GRA (Grey RelationalAnalysis) and MEW then network selection based on MEWand combinational weight is proposed It can be seen fromsimulation that the proposed algorithm can make end-userbetter avoid assessing pseudo-optimal network and has betterperformance in network load balance and reducing ping-pingeffect

2 Compensatory Analysis for MADM

21 Common MADM MADM algorithms can be dividedinto compensatory and noncompensatory ones Noncom-pensatory algorithms are used to find acceptable alterna-tives which satisfy the minimum cutoff On the contrarycompensatory algorithms combinemultiple attributes to findthe best alternative Most MADM algorithms that have beenstudied for the network selection problem are compensatoryalgorithms

SAW is widely used by most studies of the networkselection problem using cost or utility functions generallygiven by

119880SAW =

119899

sum119895=1

119908119895119906119894119895 (1)

where 119908119895represents the weight of attribute 119888

119895and 119906

119894119895repre-

sents the adjusted value of attribute 119888119895of the network 119903

119894

MEW is to calculate the coefficient by multiplicativeoperation given by

119880MEW =

119899

prod119895=1

119906119894119895

119908119895 (2)

Other twoMADM algorithms used for network selection areTOPSIS and GRA which both consider the distance from

Table 1 Network selection based on SAW

Utility 119908119895


the evaluated network to one or multiple reference networksCoefficient of TOPSIS can be calculated as

119880TOPSIS =119863120572

119863120572 + 119863120573 (3)

where119863120572 and119863120573 represent the Euclidean distances from thecurrent network to the worst and best reference networksrespectively given by

119863120572

= radic

119899

sum119895=1

1199082119895(119906119894119895minus V120572119895)2

(4)

119863120573

= radic

119899

sum119895=1

1199082119895(119906119894119895minus V120573119895)2

(5)

where V120572119895and V120573

119895represent the values of attribute 119888

119895of the

worst and best reference networks respectivelyDifferent from TOPSIS GRA uses only the best reference

network to calculate the coefficient given by

119880GRA =1

sum119899

119895=1119908119895

100381610038161003816100381610038161003816119906119894119895minus V120573119895

100381610038161003816100381610038161003816+ 1

(6)

22 Compensatory Analysis According to the principle ofnetwork selection the optimal network is the network withbest comprehensive performance but it may be not the mostreasonable one due to compensation ofMADM For examplethe result is the fact that end-user chooses best network fromnetworksA andB by SAWas shown inTable 1 where decisionattributes are cost (119862) power consumption (119864) and load (119871)Assume that only subjective weight is considered

Table 1 shows that the comprehensive performance ofnetwork B is superior to network A and end-user will choosenetwork B as access networkHowever load utility of networkB in Table 1 is 005 which is close to 0 namely networkB is not suitable for new access due to its heavy load andotherwise it may lead to congestion for network B and notbalance load between networksA andB So networkA shouldbe the reasonable choice for the above situation But end-user chooses network B due to performance compensationbetween attributes which means the excellent performanceof cost and power consumption compensates the bad per-formance of load in network B and we call the network likenetwork B pseudo-optimal network

The results that end-user chooses best network fromnetworks A and B by TOPSIS and GRA are shown in Tables2 and 3 respectively It can be seen from Tables 2 and 3 that



Utility 119908119895



Utility 119908119895



Utility 119908119895



Utility 119908119895





U(L) = 05U(L) = 005

0

01

02

03

04

05

06

07

08

09

1

U(L)w

09 08 07 06 05 04 03 01 02 01w







R =

[[[[[[[[[[[

[

11990611



sdot sdot sdot

1199061198941

sdot sdot sdot 119906119894119895

sdot sdot sdot 119906119894119899

sdot sdot sdot

1199061198981

sdot sdot sdot 119906119898119895

sdot sdot sdot 119906119898119899

]]]]]]]]]]]

]

(7)


Weight 1

WLANUMTSWiMax

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax



119864119895=

minus1

ln119898

119898

sum119894=1

119906119894119895ln 119906119894119895 (8)


119908119900119895=

1 minus 119864119895

sum119899

119896=1(1 minus 119864

119896) (9)


119904119895


119908119888119895= 120572119908119900119895+ 120573119908119904119895 (10)



119889119894=

119899

sum119895=1

(120572119906119894119895119908119900119895minus 120573119906119894119895119908119904119895)2

(11)



min119898

sum119894=1

119889119894=

119898

sum119894=1

119899

sum119895=1

(120572119906119894119895119908119904119895minus 120573119906119894119895119908119900119895)2

st 120572 + 120573 = 1

120572 120573 ge 0

(12)


Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax



120572 =sum119898

119894=1sum119899

119895=11199062

119894119895119908119900119895(119908119904119895+ 119908119900119895)

sum119898

119894=1sum119899

119895=11199062119894119895(119908119904119895+ 119908119900119895)2

120573 =sum119898

119894=1sum119899

119895=11199062

119894119895119908119904119895(119908119904119895+ 119908119900119895)

sum119898

119894=1sum119899

119895=11199062119894119895(119908119904119895+ 119908119900119895)2

(13)








119909120573






Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax













Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax











SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s




SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50N

umbe

r of n

etw

ork

hand

offs







5 Conclusions


Competing Interests


Acknowledgments


References



















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











Utility 119908119895



Utility 119908119895



Utility 119908119895



Utility 119908119895





U(L) = 05U(L) = 005

0

01

02

03

04

05

06

07

08

09

1

U(L)w

09 08 07 06 05 04 03 01 02 01w







R =

[[[[[[[[[[[

[

11990611



sdot sdot sdot

1199061198941

sdot sdot sdot 119906119894119895

sdot sdot sdot 119906119894119899

sdot sdot sdot

1199061198981

sdot sdot sdot 119906119898119895

sdot sdot sdot 119906119898119899

]]]]]]]]]]]

]

(7)


Weight 1

WLANUMTSWiMax

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax



119864119895=

minus1

ln119898

119898

sum119894=1

119906119894119895ln 119906119894119895 (8)


119908119900119895=

1 minus 119864119895

sum119899

119896=1(1 minus 119864

119896) (9)


119904119895


119908119888119895= 120572119908119900119895+ 120573119908119904119895 (10)



119889119894=

119899

sum119895=1

(120572119906119894119895119908119900119895minus 120573119906119894119895119908119904119895)2

(11)



min119898

sum119894=1

119889119894=

119898

sum119894=1

119899

sum119895=1

(120572119906119894119895119908119904119895minus 120573119906119894119895119908119900119895)2

st 120572 + 120573 = 1

120572 120573 ge 0

(12)


Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax



120572 =sum119898

119894=1sum119899

119895=11199062

119894119895119908119900119895(119908119904119895+ 119908119900119895)

sum119898

119894=1sum119899

119895=11199062119894119895(119908119904119895+ 119908119900119895)2

120573 =sum119898

119894=1sum119899

119895=11199062

119894119895119908119904119895(119908119904119895+ 119908119900119895)

sum119898

119894=1sum119899

119895=11199062119894119895(119908119904119895+ 119908119900119895)2

(13)








119909120573






Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax













Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax











SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s




SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50N

umbe

r of n

etw

ork

hand

offs







5 Conclusions


Competing Interests


Acknowledgments


References



















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Weight 1

WLANUMTSWiMax

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax



119864119895=

minus1

ln119898

119898

sum119894=1

119906119894119895ln 119906119894119895 (8)


119908119900119895=

1 minus 119864119895

sum119899

119896=1(1 minus 119864

119896) (9)


119904119895


119908119888119895= 120572119908119900119895+ 120573119908119904119895 (10)



119889119894=

119899

sum119895=1

(120572119906119894119895119908119900119895minus 120573119906119894119895119908119904119895)2

(11)



min119898

sum119894=1

119889119894=

119898

sum119894=1

119899

sum119895=1

(120572119906119894119895119908119904119895minus 120573119906119894119895119908119900119895)2

st 120572 + 120573 = 1

120572 120573 ge 0

(12)


Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax



120572 =sum119898

119894=1sum119899

119895=11199062

119894119895119908119900119895(119908119904119895+ 119908119900119895)

sum119898

119894=1sum119899

119895=11199062119894119895(119908119904119895+ 119908119900119895)2

120573 =sum119898

119894=1sum119899

119895=11199062

119894119895119908119904119895(119908119904119895+ 119908119900119895)

sum119898

119894=1sum119899

119895=11199062119894119895(119908119904119895+ 119908119900119895)2

(13)








119909120573






Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax













Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax











SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s




SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50N

umbe

r of n

etw

ork

hand

offs







5 Conclusions


Competing Interests


Acknowledgments


References



















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax



120572 =sum119898

119894=1sum119899

119895=11199062

119894119895119908119900119895(119908119904119895+ 119908119900119895)

sum119898

119894=1sum119899

119895=11199062119894119895(119908119904119895+ 119908119900119895)2

120573 =sum119898

119894=1sum119899

119895=11199062

119894119895119908119904119895(119908119904119895+ 119908119900119895)

sum119898

119894=1sum119899

119895=11199062119894119895(119908119904119895+ 119908119900119895)2

(13)








119909120573






Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax













Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax











SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s




SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50N

umbe

r of n

etw

ork

hand

offs







5 Conclusions


Competing Interests


Acknowledgments


References



















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax













Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax











SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s




SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50N

umbe

r of n

etw

ork

hand

offs







5 Conclusions


Competing Interests


Acknowledgments


References



















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Weight 1

05

06

07

08

09

1N

etw

ork

load


Weight 2

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 4

05

06

07

08

09

1

Net

wor

k lo

ad


Weight 3

05

06

07

08

09

1

Net

wor

k lo

ad


WLANUMTSWiMax

WLANUMTSWiMax











SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s




SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50N

umbe

r of n

etw

ork

hand

offs







5 Conclusions


Competing Interests


Acknowledgments


References



















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f net

wor

k ha

ndoff

s




SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

Num

ber o

f net

wor

k ha

ndoff

s


SAWGRA

TOPSISMEW


0

5

10

15

20

25

30

35

40

45

50N

umbe

r of n

etw

ork

hand

offs







5 Conclusions


Competing Interests


Acknowledgments


References



















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










5 Conclusions


Competing Interests


Acknowledgments


References



















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Documents

Research Article Compensatory Analysis and …downloads.hindawi.com/journals/jece/2016/7539454.pdfResearch Article Compensatory Analysis and Optimization for MADM for Heterogeneous