Research Article An Adaptive Handover Prediction Scheme...

Research ArticleAn Adaptive Handover Prediction Scheme forSeamless Mobility Based Wireless Networks

Ali Safa Sadiq1 Norsheila Binti Fisal2 Kayhan Zrar Ghafoor3 and Jaime Lloret4

1Faculty of Computer Systems and Software Engineering Universiti Malaysia Pahang Lebuhraya Tun Razak Gambang26300 Kuantan Pahang Malaysia2UTMMIMOS CoE in Telecommunication Technology Faculty of Electrical Engineering Universiti Teknologi Malaysia (UTM)81310 Johor Bahru Johor Darul Takzim Malaysia3Faculty of Engineering Koya University Danielle Mitterrand Boulevard Koya Kurdistan Region Iraq4Instituto de Investigacion para la Gestion Integrada de Zonas Costeras Universidad Politecnica de Valencia 46022 Valencia Spain

Correspondence should be addressed to Norsheila Binti Fisal sheilafkeutmmy

Received 15 June 2014 Accepted 14 September 2014 Published 4 December 2014

Academic Editor Weifeng Sun

Copyright copy 2014 Ali Safa Sadiq et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

We propose an adaptive handover prediction (AHP) scheme for seamless mobility based wireless networks That is the AHPscheme incorporates fuzzy logic with AP prediction process in order to lend cognitive capability to handover decision makingSelection metrics including received signal strength mobile node relative direction towards the access points in the vicinity andaccess point load are collected and considered inputs of the fuzzy decision making system in order to select the best preferableAP around WLANs The obtained handover decision which is based on the calculated quality cost using fuzzy inference systemis also based on adaptable coefficients instead of fixed coefficients In other words the mean and the standard deviation ofthe normalized network prediction metrics of fuzzy inference system which are collected from available WLANs are obtainedadaptively Accordingly they are applied as statistical information to adjust or adapt the coefficients of membership functions Inaddition we propose an adjustable weight vector concept for input metrics in order to cope with the continuous unpredictablevariation in their membership degrees Furthermore handover decisions are performed in each MN independently after knowingRSS direction toward APs and AP load Finally performance evaluation of the proposed scheme shows its superiority comparedwith representatives of the prediction approaches

1 Introduction

Promising applications provided by emerging wireless net-works are preferable as long as they can offer uninterruptedservice during mobile nodersquos (MN) roaming between accessnetworks Besides the support of an efficient mobility man-agement is considered one of very important issues forthe future generation of wireless and mobile networks andservices [1] As expected the production of wireless networksIEEE 80211 which is known as wireless fidelity (WiFi) devicesreached nearly 11 billion in 2011 which is also predictedto be doubled by 2015 [2] Therefore it is quite challeng-ing to provide Internet connection with high quality-of-service (QoS) as a way to supply the running applicationssuch as video voice-over-IP (VoIP) navigation and traffic

monitoring Normally whenMN roaming among access linkcandidates which offer different QoS level MN must be ableto chose the most appropriate network candidate to campon This can be achieved by obtaining the good networkprediction technique with low handover delay which cansupport the desired QoS of ongoing applications [3ndash5]

Basically network access link prediction is performingits processes during the time of handover decision making(wireless channel scanning period) This process started byperforming the passive and active scanning and then select-ing one network candidate as a way to perform the han-dover In handover decision making system the link layerhandover process is required to be completed by the timeknown as a link layer delay Thus the handover signalingprocesses that are used for obtaining the new IPv6 address

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 610652 17 pageshttpdxdoiorg1011552014610652

2 The Scientific World Journal

from visited network (network layer handover procedure)must wait until the link layer handover has performed itsown process Accordingly the initiation time is increasedafterwards the overall latency will be increased as well Thiscan be considered as a main reason of degrading the QoS inWLANs [6 7] Therefore the accurate handover predictionbased on link quality and mobility aspects is challenging toachieve a seamless mobility with low handover delay [4 8ndash12]

For this reason there is a pressing need to developan adaptive prediction technique in order to predict themost qualified AP An adaptive fuzzy logic system has beenproposed in order to address the issues of handover processeswithin wireless networks Thus the handover in link layercan be performed in predictive mode with low delay Theelaboratedmetrics including received signal strengthmobilenode relative direction towards access points in the vicinityand access point load are considered inputs of the fuzzydecision making system in order to select the best prefer-able AP around wireless local area network (WLAN) Theobtained handover decision which is based on the calculatedquality cost using fuzzy inference system is based on adaptiveinstead of fixed coefficients In other words the mean andstandard deviation of the normalized fuzzy inference systemrsquosinput metrics are applied as statistical information to adjustor adapt to the coefficients In addition this paper proposesan adjustable weight vector concept for inputmetrics in orderto cope with the continuous unpredictable variation in theirmembership degrees Furthermore handover decisions areperformed in each MN independently after RSS directiontoward APs and AP load are determined

2 Related Work

One of the essential issues in wireless communications isthe handover delay Hence many studies have tried to comeout with appropriate solution to decrease the associated timedelay during handover processes For this reason the pre-diction of next wireless network link with best QoS is highlyrequired which can be achieved by decreasing the handoverdelay issue A number of studies that have been publishedearlier proposed several handover prediction techniques forWLANs Some of these techniques are based on link qualityaspects whereas the mobility aspects are considered in theother techniques in order to obtain the handover decisionTherefore this section tries to present the related works thatwere elaborated to improve the handover prediction for APselection in WLANs

In order to achieve the desirable QoS of ongoing appli-cations a handover decision making process based on pre-dictive decision concept is proposed by [13] A fuzzy logicbased handoff decision algorithm was proposed in thisstudy for maintaining the handover decision within wirelessnetworks The decision making parameters were data rateRSS and mobile speed which have been selected as inputsfor the proposed fuzzy-based system as a way to select thebest candidate AP A handover scenario was introduced tobe performed between WiFi and global service for mobile(GSM) The output of the proposed fuzzy algorithm in this

method was setted to be a parameter called AP candidacyvalue (APCV) Afterwards APCV was defined as a realnumber in order to rank the value of the candidacy level ofthe APs in scanning range

However the proposed fuzzy logic algorithm in [13] didnot cover some other aspects that can improve the handoverdecision accuracy For instance when an AP is overloadedwith many associated MNs this can lead to a handoverfailureMoreover the authors did not consider the adaptationconcept of membership functions of the input parametersin their method which plays effective roles to maintain thehandover decision under different circumstances of wirelessnetwork quality changing In other words the adaptablemembership functions in the proposed fuzzy logic algorithmwere not considered to maintain the adaptability in fuzzyhandover system Furthermore in the proposed fuzzy logicinference system the authors did not maintain a weightvector technique giving more impact to the parameterthat has more variance behaviour compared to the otherinputs

The authors in [14] introduced dual-mode handsetsand multimode terminals that are generating demand forsolutions that enable convergence and seamless handoveracross heterogeneous access networks Besides the fuzzylogic approach has been proposed by [15] in order tohandle the handovers betweenWLANs and universal mobiletelecommunication systems (UMTS) The current RSS Pre-dicted RSS and the bandwidth have been fuzzyificated andnormalized to be used in handover decision making Thisproposed fuzzy logic system reduced the number of han-dovers between WLANs and UTMS during MN roamingYet in [15] the performance evaluation criteria such as han-dover delay AP load MN related direction towards each APand MN velocity are not addressed as key parameters in theproposed fuzzy logic system In other words by consideringsuch key parameters the probability of handover success willincrease

In contrast the authors in [2] proposed a predictivefuzzy logic controller to reduce the channel scanning pro-cess The proposed fuzzy system was designed based onMamdani-type as a way to predict the next AP from a groupof available APs obtained from scanning processes Twoinput parameters utilized by the proposed fuzzy system arethe average signal intensity (ASI) and the signal intensityvariation (SIV) The ASI input is calculated at two-secondintervals from the time beacon signal received by the MNwhich is normally broadcasted by APs with an interval of 100milliseconds

However the proposed predictive scheme basically relieson ASI metric that is normally used by MNs to estimatethe need of performing the handover with available APsMoreover the SIV is elaborated to demonstrate the behaviourof the direction of MN towards available APs Thereforeit can be observed that the obtained handover predictionusing proposed scheme by [2] totally relies on the intensityof received signal from available APsThus the probability ofexperiencing unnecessary and wrong handover prediction ishigh due to unpredictable movements and different channelpropagation aspects For instance by the time the ASI and

The Scientific World Journal 3

SIV are calculated and the handover decision has been trig-gered with an AP having the maximum values the directionofMN is changedThis can yield connection breakdownwiththe new obtained AP due to the new movement directionthat is unrelated to this AP then ASI and SIV values starteddecreasing To this end the author in [2] could not efficientlyaddress the issue of handover prediction within IEEE 80211WLANs

Whereas in [16] the authors proposed the Dopplerfrequency and a fuzzy logic system in the handover decisionalgorithm called the Adaptive Fuzzy Logic Based HandoverAlgorithm for Hybrid Networks their approach supposesthat if the MN speed is high then triggering handover timewill be decreased Thus avoid handover latency which isbelonging to handover procedure On the other hand whenMN speed is low the trigger handover time will be increasedto get more suitable networks On the contrary the proposedalgorithm does not consider MN speed suitability to the nextAPs with different wireless technologiesTherefore when thespeed is high the handover will fail Moreover the algorithmdoes not consider the load of each AP which leads to ahandover failure as well

In [17] the authors focused on the handover in APdense 80211 networks Through this study the AP scanningprocess has been highlighted in order to achieve an improvedscan technique for 80211 networks Two key features proberesponse arrival time and AP signal quality were discoveredin this study in a way to reduce the active probing timeMoreover an improved version of D-Scan has been pro-posed by the aforementioned authors The authors focusedin the first stage on the probing wait time which is theMaxChannelTime The authors decreased the active probingtime by identifying the correlation between probe responsearrival time and RSS quality In this proposed approach thehandover is triggered based on the link quality of the currentassociatedAP It also performed a regular detection of the linkquality of current AP Whenever the current link quality ofassociated AP is poor enoughwhichmeans that the handoveris needed that is RSSI lt HANDOFF-THRESHOLD anactual handover process is enforced Otherwise if it is lowerthan a certain threshold (SCAN-THRESHOLD) the networkinterface card (NIC) is started to perform the backgroundprescan Eventually all obtained APs are stored in a local APdatabase Thus the scan process was trying to find a certainnumber of APs with acceptable RSSI (gtminus75 dBm) WhenD-Scan process cannot find good Apsrsquo RSSI quality on thecurrent channel it is switched to the next channel to scanuntil the whole frequency has been searched Afterwards thehandover will be initiated with the AP that has more signalquality compare with others

However the authors in this study did not consider asmart prediction technique in the proposedD-Scan approachwhich can give high impact in handover process Moreoverthe D-Scan approach relies on the link quality of associatedAP by monitoring the RSSI that cannot be reliable inAPs-dense wireless networks due to fluctuations normallyoccurring during MN movement Thus the RSS value ofcollected APs independently changes then the handover

decision obtained based on one metric (RSS quality) will beinaccurate

3 Proposed AHP Scheme Overview

Fuzzy logic basedmechanisms performwell in decisionmak-ing systems control estimation and prediction processesFor instance in [18] Shih et al proposed a productioninventory model to precisely estimate seasonal demand andtotal demand Other researchers in [19] utilized fuzzy logicin parallel interference cancellation (FLPIC) for frequency-selective fading channels in wireless CDMA communicationsystems In addition in [20] the authors used fuzzy logicin geographical routing when making packet forwardingdecisions In light of these applications fuzzy logic has beenapplied in this study to select the most qualified AP in termsof RSS MN direction and AP load based on WLAN

Figure 1 illustrates the systematic architecture of AHPdesign implementation and evaluation phases of the AHPscheme It can be seen that the AHPrsquos process begins withthe collection of fuzzy inferencersquos input parameters In otherwords in the first turn the GPS set-up process will beperformed in each MN in order to obtain the 119909-axis and119910-axis of each AP in the simulated scenario along withupdatedMNrsquosmovement vectors From the obtained data thedirection angle can be calculated in such a way as to observethe current MNrsquos direction in relation to each AP in theroaming area It should be noted that RSS monitoring is onwhenever theWLANrsquos interface is ldquoonrdquo in order to ensure thehighest quality of each available APThis process is performedvia wireless channelrsquos passive and active scanning for thethree nonoverlapped IEEE 80211b channels (Channels 1 6and 11)

Furthermore the current load of each AP is calculatedand broadcasted via beacon frame The RSS and AP loadvalues will be extracted from the beacon frame of each APwithin scanning range After MN measures the received RSSof the current associated AP the value RSS119862 is comparedwith threshold value 119879 When RSS

119862is less than 119879 the

AHP algorithm begins the process of obtaining the handoverdecision for the next predicted AP candidate

Therefore the proposed AHPrsquos fuzzy inference enginewas utilized to obtain the quality cost of each collected AP119865119906119911119911119910-119876-119862119900119904119905

119894 After the defuzzification process the AHP

checked whether 119860119875-119876-119862119900119904119905 119900119891 119888119906119903119903119890119899119905 119860119875-119865119906119911119911119910-119876-119862119900119904119905119894gt ℎ if 119910119890119904 the handover was initiated with selected

AP119894

and the 119862119906119903119903119890119899119905-119860119875-119876-119862119900119904119905 was replaced with119865119906119911119911119910-119876-119862119900119904119905

119894 ℎ is the identified unnecessary handover

restriction threshold which represents a quality cost thresh-old value Should the obtained 119865119906119911119911119910-119876-119862119900119904119905

119894exceed

this value the handover is not needed and afterwardsrestricted Hence based on AHP processes the handoverdecision is taken with the AP with highest QoS withconsideration given at the same time to restrict unnecessaryhandovers

31 Fuzzification of AP Selection Input Metrics and OutputAs discussed previously the selection criterion is one of the


Set-up GPS service in current MN and collect

Calculate current MNrsquos direction

Get current MNrsquos updated x-axis and y-axis

AP Q Cost-Fuzzy Q Costi gt h

Initiate the handover with predicted APi with Fuzzy Q Costi

AP Q Cost = current AP Q Cost

Get Fuzzy Q Costi (RSS direction and AP-load)

Send association req frame to predicted APi

Current AP Q Cost = Fuzzy Q Costi

Scan available WLANsrsquo channels (RSS monitoring)

Extract RSS and AP-load from beacon frame

the positions of all available APs

Run AHPrsquos fuzzy inference system and calculate the weight vector for each input metric

No

Yes

and movement vector

Start

broadcast via beacon frame

towards all available APs

Yes

No

No

Read adaptation variables of membership functions for each

Yes

Yes

Yes

No

No

input metric (RSS D and L)

Calculate the AP-load in each APi and

i = 1

i++

Perform authentication phase with predicted APi

Predicted APi authenticated

Beacon frame is received and RSSC lt T

Association response received

Figure 1 Flowchart steps of the proposed AHP scheme

most challenging areas in the handover process essentiallyaffecting the handover delay inWLANsHandover procedureshould support always-best-connected (ABC) and always-best-satisfying (ABS) when selecting the target access stationTherefore the lack of precise evaluation of the QoS metrics

of the available WLAN candidates could be responsible formany of the drawbacks of ABC and ABS For examplehandover decisions which rely only on the quality of RSSfor each AP selection process regardless of MNrsquos directionin relation to each AP can actually increase the number


05

1

Weak Average Strong

0

RSS value

RSSmin Ath Wmax Sth Amax RSSmax

Adap

tive m

embe

rshi

p fu

nctio

ns o

f (120583

RSS)

Figure 2 The adaptive membership functions of normalized RSSinput metric

of unnecessary or incorrect handovers Similarly when theselectedAP currently servingmany of theMNs is overloadedhandover failure phenomena may result In other wordswhen the association request from the newly reached MNarrives at the overloaded AP the probability that the associ-ation request will be discarded is quite high Hence in thefollowing subsection a fuzzy logic input metrics based AHPalgorithm is discussed

311 Received Signal Strength RSS Received signal strengthRSS is one of the most common metrics used in handoverdecision making [10 17 21] By monitoring RSS the qualityand distance of each AP in the range can be analyzedWhen MN moves away from or towards an AP (ping-pong movement) the RSS for the AP will either increase ordecrease Therefore during the passive scanning phase theRSS in AHP scheme for each available AP will be capturedand entered into the fuzzy inference system to be fuzzifiedThe range of RSS membership functions is considered tobe in adaptive form In other words in order to achieveadaptivemembership functions for RSS inputmetric the RSSvalue to be distributed is to be between identified RSSMinand RSSMax normalized values When the RSS for eachAP in MNrsquos scanning range has been collected during thepassive scanning process the RSS values are normalized andcategorized using (1) (2) and (3)

Figure 2 shows the membership functions for RSS inputmetric There are three RSS levels identified as Weak Aver-age and Strong and the range of them was identified usingthe aforementioned assumed variables By using variables119860 th threshold value of Average membership function 119882Maxmaximumvalue ofWeakmembership function 119878th thresholdvalue of Strong membership function and 119860Max maximumvalue of Average membership function in addition to mini-mum and maximum values of RSS RSSMin and RSSMax thepiecewise linear membership functions could be obtainedas shown in (1) (2) and (3) By expanding these equationsthe degrees between (0 to 1) of RSSrsquos membership values

are calculated respectively as shown in the vertical axis inFigure 2 Consider

120592WeakRSS =

1 (RSSMin le RSS le 119860 th) 10038161003816100381610038161003816100381610038161003816

RSS minus 119860 th119882Max minus 119860 th

10038161003816100381610038161003816100381610038161003816

(119860 th le RSS le 119882Max)

0 (RSS gt 119882Max)

(1)

120592AverageRSS =

1 (119882Max le RSS le 119878th) 10038161003816100381610038161003816100381610038161003816

119860 th minus RSS119860Max minus 119860 th

10038161003816100381610038161003816100381610038161003816

(119860 th le RSS le 119882Max)

or (119878th le RSS le 119860Max)


or (RSS lt 119860 th)

(2)

120592StrongRSS

0 (RSSMin le RSS le 119878th) 10038161003816100381610038161003816100381610038161003816

119878th minus RSSRSSMax minus 119878th

10038161003816100381610038161003816100381610038161003816

(119878th le RSS le 119860Max)

1 (119860Max le RSS le RSSMax)

(3)

Therefore the proposed adaptive fuzzy system is usedin the AHP scheme whereby the membership functions areidentified adaptivelyThe RSSMin RSSMax119860 th119882Max 119878th and119860Max coefficients are designed in a way which can be setup by the user thus the membershiprsquos coefficients can beobtained adaptively For instance after identifying the valuefor each of the aforementioned coefficients (supposing minus130minus10 minus85 minus70 minus50 and minus40 dBm resp) when the collectedRSS value of an AP is minus75 dBm this value will be checkedto identify which membership function it belongs to Basedon Figure 2 and after substituting the supposed values ofeach particular coefficient the value minus75 dBm is allocated inthe triangle with rib of (119860 th 119882Max) In other words when119860 th = minus85 and 119882Max = minus70 (minus85 le minus75 le minus70) hence(1) is applied to range between 0 and 1 Thus by substitutingnumerical values in |(RSS minus 119860 th)(119882Max minus 119860 th)| it will be|((minus75) minus (minus85))((minus70) minus (minus85))| = 066 This obtained valueutilizing (1) implies that the collected RSS value of minus75 dBm isallocated inside the weak membership function of RSS inputmetric and conflicts with medium membership function aswell so the degree of weak value is 066 Accordingly theconsidered membership value is adaptively calculated for theinput values neither totally inside nor outside any particularmembership function

312 Relative Direction between MN and AP The secondfuzzy input metric is the related MN direction towards eachAP Basically when an MN starts roaming across differentAps it could determinemore than one APwith a high qualityRSSOn the other hand this identifiedAPmaynot be situatedin the same direction as the MN In other words the MNmovement direction is not towards this particular AP Thisscenario shows that the MN can obtain the wrong handoverdecision if the AP does not share the related direction withthe MN Therefore the related MN direction towards each


05

1

Related MN direction angle towards each AP

0MDth LDmax HDth MDmax DmaxDmin

High directedMedium directedLess directed

Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583D

)

Figure 3The adaptivemembership functions of normalized relatedMN direction towards each AP input metric

AP has been assigned as an inputmetric in the proposedAHPadaptive fuzzy inference system

In simulation experiment scenario the MN is equippedwith a GPS system in order to obtain the updated 119909-axisand 119910-axis with every movement in addition to the currentMNrsquos movement vector The location of all APs in simulationscenario is defined as fixed positions in advance When MNis in the first round all AP positions will be collected throughGPS navigator map and the coordinates of these APs willbe saved inside its own basic service set ID (BSSID) Thuseach time an MN roams through the network the updatedGPS map will be downloaded automatically from the serverwith 119909-axis and 119910-axis for all APs in the range and its owncurrent position as well With this process the MN is awareof its own position in relation to themovement vector and thepositions ofAPs in the rangewhich enables us to calculate thedirection angle between the current MN position and eachAP

Formula (4) is used to calculate the direction anglersquosdegree for each AP from the current MN position duringits movement Suppose that the current MN position is(MN1198831 MN1198841) the position of AP is (AP1198832 AP1198842) MN119909and MN119910 represent the current position of the MN Formula(4) describes how the direction angle has been calculatedThe obtained crisp angle degree value will be entered to thefuzzy inference system as a second input parameter to befuzzified with the other two inputs Figure 3 demonstratesthe membership functions of related MN direction towardseach AP input metric distributed as Less-Directed Medium-Directed and High-Directed

The bearing angle (120579) between a MN and AP can becalculated as follows

cos 120579 =

MN1199091 sdot AP1199092 +MN1199101 sdot AP1199102

radicMN21199091

+MN21199101

sdot radicAP21199092

+ AP21199102

(4)

The range of membership function for directions selectedto be between 119863Min which equals minus1 reflects an MN directedless to one particular AP up to 119863Max which equals 1 andis highly directed towards AP On the other hand thevariables LDMax Low-Directed membership functionrsquos max-imum value MDth Medium-Directedmembership functionrsquosthreshold MDMax Medium-Directed membership functionrsquos

maximum value and HDth High-Directed membershipfunctionrsquos threshold are identified in order to achieve adap-tive direction membership functions Using these identi-fied variables the coefficients of membership functions fordirection input metric are obtained in an adaptive wayUtilizing (5) (6) and (7) the degrees of membershiprsquosvalues of direction metric are calculated based on identifiedcoefficients

For instance when the identified coefficients in Figure 3are set as 119863Min = minus1 119863Max = 1 MDth = minus04 LDMax =minus02 HDth = 06 and MDMax = 07 the obtained 119863 valueusing (4) is 069 degree of 120579 It is obvious that the 119863 valueof 069 is allocated in the highlighted triangle with therib of (HDth MDMax) in horizontal axis in Figure 3 whichis considered to be a conflict area between medium andhigh directionmembership functionsTherefore by applying(7) (HDth le 069 le MDMax) the degree of high directionmembership function is thus calculated to be in the range of0 lt MembershipDegree lt 1 Accordingly by substitutingthe given coefficients in |(HDth minus 119863)(119863Max minus HDth)| theobtained degree is |(06 minus 069)(1 minus 06)| = 0225 High-Directed membership degree

313 AP Load In order to achieve an accurate handoverdecision a third input metric the load in each AP hasbeen considered in the proposed adaptive fuzzy inferencesystem of AHP In some cases the handover decision makingmechanism could assign high quality cost to one AP whichhas a good RSS and is with a high direction anglersquos degree120579 towards this AP On the other hand the selected APmight be overloaded In other words based only on thetwo aforementioned metrics with this particular AP andregardless of the number ofMNs that are currently associatedwith it (by sending and receiving the traffic) the obtainedhandover decision is considered inaccurate The implicationis that when a new MN intends to establish a new handoverprocess with an AP the handover might fail due to the highload currently borne by that AP For this reason the APload has been assigned as an additional input metric in theproposed adaptive fuzzy inference system of AHP in order tosupport an accurate handover decision

To identify the membership functionrsquos range of AP loadmetric a simulation experiment has been conducted Theoutdoor campus of 500 lowast 500m is utilized to simulate theconducted wireless network scenario In addition the APrsquostransmit power and data rate are set at 60 milliwatt and11Mbps respectively Voice-over-IP traffic has been gener-ated between MNs in the simulation area in order to testthe load in the AP with real time applications In order toprecisely identify the maximum number of MNs that an APcan serve with reasonable throughput the number of MNsis increased gradually in the simulated network area and thethroughput has been collected in the AP side

The simulator experiment has been conducted four timesand the APrsquos throughput is collected as shown in Figures 4(a)4(b) 4(c) and 4(d) The AP throughput has been capturedeach time that MNrsquos number increased Figure 4(a) showsthat when the number of MNs was 12 the AP throughputafter 105 seconds reached 96000 bitssec with constant value


0

20000

40000

60000

80000

100000

120000

0 100 200 300 400 500Time (s)

Throughput (bitss) in AP with 12 MN

(a) APrsquos throughput with 12 MNs

0

50000

100000

150000

200000

250000

0 100 200 300 400 500Time (s)


(b) APrsquos throughput with 27 MNs

0

100000

200000

300000

400000

500000

600000

0 100 200 300 400 500Time (s)


(c) APrsquos throughput with 40 MNs

0

500

1000

1500

2000

2500

0 100 200 300 400 500Time (s)


(d) APrsquos throughput with 43 MNs

Figure 4 Analyses of the impact of increasing MNs number on APrsquos throughput

until end of simulation On the other hand in Figure 4(b)the AP throughput was 192000 bitssec when the number ofMNs increased to 27 When the number of MNs reached40 in Figure 4(c) the AP throughput after 110 secondsincreased to 527424 bitssec but rapidly decreased after 5seconds to between 192768 and 145536 bitssec However inFigure 4(d) when the number of MNs increased to 43 theAP throughput continued to decrease to the range of 192 to1920 bitssec From this experiment it can be concluded thatthe AP throughput with real-time traffic begins to decreasesharply when the number of associated MNs reaches 40Consider

120592Less119863

1 (minus1 le 119863 le MDth) 10038161003816100381610038161003816100381610038161003816

119863 minusMDthLDMax minusMDth

10038161003816100381610038161003816100381610038161003816

(MDth le 119863 le LDMax)

0 (119863 gt LDMax)

(5)

120592Medium119863

=

1 (LDMax le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

MDth minus 119863

MDMax minusMDth

10038161003816100381610038161003816100381610038161003816


or (HDth le 119863 le MDMax)

0 (119863 gt MDMax)

or (119863 lt MDth)

(6)

120592High119863

=

0 (minus1 le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

HDth minus 119863

119863Max minusHDth

10038161003816100381610038161003816100381610038161003816

(HDth le 119863 le MDMax)

1 (MDMax le 119863 le 1)

(7)

Therefore the AP load input metric range has beenassigned to between 119871Min = 0 and 119871Max = 40 which represents


Table 1 The Modified AP load element in beacon frame

Octets1 1 2 1 2

AP load Length 7octets

Stationcount

Channelutilization

Availableadmissioncapacity

05

1

Low Medium High

0Loadmin MLth HLthLLmax MLmax Loadmax

AP load value

Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583

load

)

Figure 5 The adaptive membership functions of normalized APload input metric

the number of associated MNs In other words no MNcurrently associated with the AP indicates that the AP iscurrently with 0 load and is now ldquopreferredrdquo On the otherhand when the number of MNs reaches 40 the AP currentlywith maximum load is ldquonot-preferredrdquo The load range isnormalized to be between 0 and 1 using the adaptationprocess for membership functions Elaborating the received119878Count value from each AP which is the Station Countcollected from AP load elements in the beacon frame thenormalized AP load is obtained via adapted membershipdegree

Basically the APload value which consists the 119878Count(number of associatedMNs) is periodically broadcast viaAPrsquosbeacons in the wireless network area The AP load elementin the AP beacon frame has been modified by adding apredetermined number ranging between 0 and 40 Table 1shows the modified AP load element in the beacon frame ineach particular AP Thus when MN receives this amount ofAP load the adaptation process is applied in order to obtainthe membership functionsrsquo degree of load input metric

Figure 5 shows the adaptive membership functions ofnormalized AP load input metric categorized as LowMedium and High It can be seen that the variables MLthmedium load membership function threshold LLMax lowload membership function maximum value HLth high loadmembership function threshold and MLmax medium loadmembership function maximum value are identified in away that allows for the calculation of the degree of eachmembership function Equations (8) (9) and (10) are appliedas a piecewise linear function to calculate the degree for eachLow Medium and High membership function

Similarly a numerical example is illustrated in thisparagraph in order to present the process of obtaining themembership functionsrsquo degree for AP load input metricSuppose that MLth = 035 LLMax = 04 HLth = 073 MLmax =073 and the collected 119871 value from an APrsquos beacon frameis 05 When the given value 05 is compared among the

identified coefficients as presented in (8) (9) and (10) theappropriate membership function is subsequently selectedThus using (9) it can be observed that (LLMax lt 05 lt HLth)which implies that the given 119871 value is completely underthe Medium membership function Therefore based on thepreceding equation the value 1 is given as the input value ofthe medium membership function degree of 05119871

32 Design of Adaptive Fuzzy Logic for Handover PredictionSystem The first step in designing a fuzzy inference systemis to determine input and output variables and their fuzzyset of membership functions An adaptive process is appliedin order to obtain the degree of membership functions foreach input metric In addition the adaptive weight vectoris obtained by calculating the weight impact caused by thevariance of each input metric and then determining thevector which helps to obtain the final fuzzy quality cost foreach AP This is followed by designing fuzzy rules for thesystem Furthermore a group of rules are used to representthe inference engine (knowledge base) to express the controlaction in linguistic form The adaptive input metrics of thefuzzy inference system which are elaborated in AP selectionand prediction process are presented in Section 31 Consider

120592LowLoad =

1 (0 le 119863 le MLth) 119871 minusMLth

LLMax minusMLth (MLth le 119871 le LLMax)

0 (119871 gt LLMax)

(8)

120592MediumLoad

=

1 (LLMax le 119871 le HLth) MLth minus 119871

MLMax minusMLth (MLth le 119871 le LLMax)

or (HLth le 119871 le MLMax)

0 (119871 gt MLMax) or (119871 lt MLth) (9)

120592High119871

=

0 (0 le 119871 le HLth) HLth minus 119871

119871Max minusHLth (HLth le 119871 le MLMax)

1 (MLMax le 119871 le 40)

(10)

321 Adjustable Weight Vector for Input Metrics of FuzzyInference Engine in AHP In this section the process ofobtaining the weight vector 119882 which was calculated via (13)is presented Taking into account various conditions thevalues of each RSS119863 and 119871 in addition to their membershipdegrees continuously vary in an unpredictable manner Inorder to achieve the best handover decision under differentconditions the following features have been considered toobtain adaptable weight vector

(1) The weight vector for each input metric should not befixed meaning that it must be adjustable according tovarying conditions


(2) The input metric whose value varies to a greaterdegree compared with other metrics this metric isconsidered to be more important and it must have ahigher weight

For instance assume that the RSS candidate value (MN1

MN2 MN

119899) has the maximum variance compared to the

variance in the value of other metrics 119863 and 119871 Thereforeit will be adjusted to the largest value in terms of weightvector Equation (11) shows the calculation of the weightvector adjustment process for fuzzy input metrics In thiscase 119860RSS 119860119863 and 119860

119871are the adjusted values of each input

metric (RSS119863 and 119871) and 120590RSS 120590119863 and 120590119871 are the standarddeviations of each input metric respectively Consider

119860 = (119860RSS 119860119863 119860119871) = (

120590RSS119894120590119894

120590119863

119894120590119894

120590119871

119894120590119894)

119894 isin RSS 119863 119871

(11)

The standard deviation of the membership values havebeen normalized in (11) where the120590

119894is the standard deviation

of 1205921198941 1205921198942 120592

119894119899and 119872

119894is their mean calculated utilizing

the following equations (12) Consider

119872119894=

1

119899

119899119895=1

120592119894119895 119894 isin RSS 119863 119871

120590119894=

1

119899

119899119895=1

(120592119894119895minus119872119894)

2

119894 isin RSS 119863 119871

(12)

In addition the RSS is considered a basic input metricfor decision making in the handover process thus it shouldbe highlighted that low variance of RSS metric should not bereflected in a decrease in its own weight vector For instancewhen the overall average of 120592RSS119895 (119895 = 1 2 119899) is low theadaptation of membership degree of input metric RSS mustbe tackled more seriously For this reason the weight vectorof RSS input metric 119882RSS should be given a higher weightvalue among the other two input metrics (119863 and 119871)Thus thelink-breakdown probability is reduced by ensuring that thehandover decision based on RSS parameters during the timeof link quality is weak ranking value is in overall averageOn the other hand when the mean value of RSS 120592RSS119895 ishigh in overall average the effects of its weight vector will bemoderated among the other two input metrics (119863 and 119871)

Moreover a weight vector119882 is identifiedwhich presentsthe weight of input metrics RSS119863 and 119871 as follows

119882 = [119882RSS119882119863119882119871] (13)

The detailed weight vector calculation is presented usingthe following equation (14) whereas the average variance istackled seriously with RSS input metric rather than othertwo input metrics (119863 and 119871) as presented in the examplein the previous paragraph The important point to note hereis that (14) is adjustable based on the input metric that ismore variable during theMNrsquos roaming process For instancewhen theMN ismovingwithmany changes in direction angle120579 the mean of direction input metric119872

119863will be considered

03 04 05 06 07 08 09 1

05

1Hcost VHcost

0

Mem

bers

hip

func

tion

020

LcostVLcost Mcost

Output variable ldquoAP Q Costrdquo

Figure 6 The membership function for AP quality cost outputmetric

in (14) Hence the accuracy throughout this feature improvesthe handover decision making process Consider

119882 = (119908RSS 119908119863 119908119871)

= (

119860RSS119872RSS

119872RSS times 119860119863119872RSS times 119860

119871)

(14)

Suppose that for the 119895th (119895 = 1 2 119899) mobile stationnamely MN

119895 the membership degree vector 119880

119895has been

defined from combination of three inputmetrics (RSS119863 and119871) Consider

119880119895 =

120592RSS119895120592119863119895

120592119871119895

(15)

Based on (15) and (13) the final fuzzy cost FuzzyCost119894 ofmobile station 119895th can be obtained utilizing (16) Consider

Fuzzycost = 119880119895 sdot 119882 (16)

The output AP quality cost from fuzzy inference systemFuzzyCost119894 is configured to range between (0 to 1 Rank) fromlower value to the higher value of quality cost for each APFor instance Figure 6 shows that the division of quality costoutput of AP has five levels of rank VLcost Lcost McostHcost andVHcostThefinal fuzzy inference decision is basedon the adaptive membership degree vector of each inputmetric and the weight vector as presented in (16) Moreovertriangular functions are used as membership functions asthey have been widely used in real-time applications dueto their simple formulas and computational efficiency It isimportant to highlight that a good membership functiondesign has a significant impact on the performance of thefuzzy decision making process

322 Adaptive Fuzzy Inference Engine In the proposed AHPscheme the adaptive membership function is proposed andutilized in the design of a fuzzy inference system Moreoverit is important to mention that the precise design of member-ship function has a major impact on the overall performanceof the fuzzy prediction process Furthermore the proposed


Table 2 Knowledge structure based on fuzzy rules

Rule IF THENRSS Direction AP load AP-Q-Cost

1 Weak Less-Directed High VLcost2 Weak Less-Directed Medium Lcost3 Weak Less-Directed Low Lcost

27 Strong Medium-Directed Low VHcost

weight vector concept and the best AP selection processcontribute positively to increase the quality of the obtainedfinal handover decision Table 2 demonstrates the utilizedfuzzy rules in the proposed fuzzy inference system

323 Defuzzifcation Defuzzification refers to the way thata crisp value is extracted from a fuzzy set value In theproposed fuzzy decision making system in AHP the centroidof area strategy for defuzzification has been considered Thisdefuzzifier method is based on Formula (17) as followsConsider

Fuzzycost =sumAll Rules 119880119895 times119882

sumAll Rules 119880119895 (17)

where Fuzzycost is used to specify the degree of decisionmaking119882 is theweight vector variable of inputmetrics (RSS119863 and 119871) and 119880119895 is their adaptive degree of membershipfunctions Based on this defuzzification method the outputof the AP-Q-Cost is changed to a crisp value

33 The Best AP Selection Process in AHP The handoverdecision is performed in local host mode as each MN mea-sures the received RSS and its direction degree towards eachavailable AP In addition to the received AP load value whichis broadcast via each AP the handover decision utilizing theimplemented AHP (whenever RSS from current serving APRSS119888 degrades below a threshold 119879) is then carried outConsider

RSS119888lt 119879 (18)

Afterwards the decision factor AHP119865 based on the

calculated FuzzyQCost119894for all the candidates is obtained and

the AP119894candidate is chosen for handover initiation if the

following condition is satisfied

AHP119865 = APCost minus FuzzyCost119894 gt ℎ (19)

where ℎ is the threshold value which helps to avoid unnec-essary handovers FuzzyCost119894 is the final decision metric ofmaximum quality cost of AP

119894 candidate APCost is the qualitycost of the currently serving AP and AHP

119865is the difference

between decision factor of the serving AP and the AP119894target

Table 3 Simulation parameters

Parameters ValueSimulation time 700 sSimulation area 2500 times 1500mMobility model Rectangle and mass modelsNumber of MN 50MN Speed Maximum 60 kmhTransmitted power WLAN 17 dbmTransmission range of eachAP 400 meter

Maximum packetgeneration rate 1350 packetsecond

Maximum packet size 1000 byteChannel bandwidth WLAN 11MbpsMAC protocol of WLAN IEEE 80211b PCF

4 Performance Evaluation

To evaluate the performance of the proposed AHP scheme asimulation scenario is created employing OMNET++ simu-lator and the AHP scheme was implemented along with thestate of the art which are the existing AP predictionmethodsin wireless networks The evaluation is conducted based onseveral metrics which are the impact of MNrsquos number onaverage handover delay impact of MNrsquos number on averagehandover delay AP load total number of handovers numberof failed handovers handover failure probability averageMAC-layer delay the impact of MNrsquos number on packetloss ratio and adaptive Fuzzy-Quality-Cost of available APsin simulation scenario Table 3 demonstrates the simulationparameters that were utilized in simulation AHP scheme byOMNeT++

In order to achieve simplicity in presenting the simulationresults the two compared methods are represented by short-form style The method proposed in [13] is denoted asaccess point candidacy value (APCV) whereas the othermethod in [22] Scan in AP-dense 80211 networks is calledD-Scan On the other hand adaptive handover predictionhas been previously identified as an AHP scheme A detaileddiscussion of all the aforementioned evaluation metrics ispresented in the subsections below

41 Impact of MNrsquos Number on Average Handover DelayFigure 10(a) illustrates the impact of MNrsquos number onobtained average handover delay based on 5 simulationruns As a function of MNrsquos number increasing up to amaximum of 50 MNs graphs of average handover delay inseconds are collected and presented for each AHP schemeand APCV and D-Scan methods in Figure 10(a) As can beseen as the number of MNs is increased to 50 the AHPscheme performed the best in decreasing the overall averagehandover delayMore precisely the handover delay withAHPscheme maintained an average of 006 to 009 sec when thenumber of MNs increased from 1 to 18 In contrast whenthe number of MNs increased to 19 the average handover


delay increased to 01 secThe handover delay kept increasingslightly on average as the number of MNs reached 22 theaverage delay was fixed to 023 sec up to 50 MNs

On the other hand the achieved average handover delayutilizing APCVmethod was very similar to the one obtainedbyAHP schemewith 10MNs running in a simulated scenarioThis delay started to increase sharply after 18 MNs It can beobserved from the resulting graph of APCV method that theaverage handover delay reached 1098 sec when the numberof MNrsquos reached 43 However the delay keep increasingsimilar to the increase which occurred in MNrsquos number untilreaching 284 sec with 50 MNs In contrast although D-Scan method is designed to decrease the handover delay byincorporating smart scanning processing in the link layera worse performance is observed with respect to both theAHP scheme and APCV method when the number of MNsis increasing As observed from the results presented inFigure 10(a) note that the average handover delay beganto increase sharply after 29 MNs (more than 1 sec delay)compared to both AHP and APCV results The averageobtained handover delay by D-Scan method continued toincrease as the number of MNs increased until it is reached299 sec after 43 MNs

In fact the serious improvement in decreasing averagehandover delay which was achieved using the proposed AHPscheme is due to the fact that the handover decision inthe AHP scheme is obtained in cooperation with adaptiveAP load input metric Therefore the handover process didnot encounter any overloaded APs keeping the averagehandover delay low regardless of the increase in the numberof MNs However this feature was not considered in eitherthe APCV or D-Scan method which resulted in the failureto reduce the handover delay in the low average range inresponse to increases in the number of MNs Finally it isworth mentioning that the AHP scheme could efficientlydecrease the average handover delay as the number of MNscontinued to increase This was achieved by developingadaptive coefficients of the mean and standard deviation ofthe normalized fuzzy inference systemrsquos input metrics

42 AP Load Asmentioned earlier AP load is an importantmetric that must be considered during the handover decisionmaking process Handover decisions obtained with oneparticular AP with high load cause high handover delay andmight cause handover failure Hence the AP load consideredin the proposed AHP is an important metric that contributespositively to the AP rankings This leads to making handoverdecisions with the most qualified AP candidate by takinginto account its current load Moreover the AHP schemeusing AP load metric made an essential contribution insupport of wireless networks by creating load balancingamong APs ensuring or improving accuracy in handoverdecisionmakingTherefore as can be seen from Figure 10(b)the AHP scheme reduced the load balance that was tackledby each AP and distributed it fairly among all 9 APs in thesimulation scenario

In order to provide a perspective example of calculatedAP load the average load obtained from AP1 is highlightedin this paragraph Alternatively Figure 10(c) presents the

0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV

Figure 7 Number of total handovers

average of obtained AP load of AP1 utilizing AHP schemeand APCV and D-Scan methods measured in bitssec duringsimulation time As a function of load (bitssec) tackledby AP1 during simulation time the output graphs of AHPscheme and APCV and D-Scan methods over the first 100seconds all acting on the same load are shown The reasonis that during this period of time AP1 is still servingonly the first round of MNs When the new MNs begin toassociate with AP1 as a result of handovers beginning after100 seconds the obtained load by both APCV and D-Scanis increased sharply At the same time the load obtained byemploying AHP scheme continued to decrease throughoutthe simulation time comparedwith APCV andD-Scan whichobtained higher loads respectively

In percentage form the AP1 load as presented inFigure 10(b) indicates that the achieved load is the lowestusing the AHP scheme followed by D-Scan and APCVmethods respectively Similarly in Figure 10(c) the AHPscheme is superior in terms of decreasing the load (bitssec)performed by AP1 during simulation time compared with thestate of the art or the existing APrsquos prediction methods Thishas a tremendous effect on the load balancing among theavailable APs in the simulated area Thereby the handoverprocess avoids overloadingAPs as long as there are alternativeAPs with better quality cost obtained using the proposedadaptive fuzzy inference system

43 Total Number of Handovers In order to evaluate theproposed AHP scheme in terms of the ability to maintainthe total number of handovers at an acceptable level fiveMNs have been selected to observe the average number ofhandovers that are performed with each simulation runThroughout this evaluationmetric the level of improvementsin prediction accuracy can be studied and analyzed as a wayto validate the proposed AHP schemersquos performance Thetotal number of handovers processed during the simulationtime by the five selected MNs has been captured and thencalculated Figure 7 illustrates the total number of handoverdecisions triggered by each of the five MNs (successful andfailure handovers) employing AHP APCV and D-Scan It


0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan

Figure 8 Number of failed handovers

was observed that by using the proposed APH scheme theaverage number of handovers that are obtained by MN1 is 5whereas MN2 and MN4 are achieved 4 On the other handthe obtained average handovers within MN3 and MN5 was 3handovers Utilizing APCV and D-Scan the average numberof total handovers were 9 6 5 6 and 7 and 7 7 6 5 and 5 asa sequence of five selected MNs respectively

It is obvious that proposed AHP scheme performs betterthan both APCV and D-Scan methods in terms of reducingthe total number of handovers In other words by using theAHP scheme unnecessary and incorrect handover decisionshave been significantly reduced or avoided This is due to thefact that in proposed AHP scheme the MN calculates thequality cost of each neighbour AP using the proposed adap-tive fuzzy inference system before performing the handoverprocess Thus the obtained handover decision is based onthe correlation between three fuzzified input metrics (RSSrelated direction and AP load) and is more accurate

44 Number of Failed Handovers From another perspectiveto evaluate the proposed AHP scheme in terms of reducingthe number of unsuccessful handovers the average numberof failed handovers in each of 5 selected MNs has beencalculated Through conducting this performance test theability in obtaining correct handover predictions in WLANscan be examined which subsequently contributes in reducingthe handover delay By looking at Figure 8 it can be observedthat MN2 MN3 and MN5 using the proposed AHP schemedid not face any handover failure during simulation timeHowever MN1 and MN4 obtained one handover failureIn contrast the number of failed handovers in each of 5MNs using both APCV and D-Scan was 3 1 0 1 and 1and 3 2 1 2 and 3 respectively In different form whencounting the average number of failed handovers out ofthe five MNs as presented in Figure 8 for the three appliedschemes the obtained average number using each imple-mented scheme was 04 AHP 12 APCV and 22 utilizing D-ScanThis indicates that the proposed AHP scheme achievedthe lowest average of failed handovers while APCV and D-Scan methods followed in rank order This is not surprisingsince the proposed AHP scheme relies on a predictive fuzzyinference system based on three input metrics (RSS related

02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan

Figure 9 Handover failure probability

direction and AP load) Hence the handover process wasperformed each time with the most qualified AP candidatein the scanning area

On the other hand in APCV method the MN obtainedthe handover decisions with APs based on the candidacyvalue obtained via fuzzy logic regardless of APrsquos currentload factor and its related direction aspect In contrast D-Scan method relies only in a predictive way on performinga fast and active scan for existing APs which contributesto reducing the Maximum Channel scanning time Thesimulation experiment conducted in this regard shows thatthe D-Scan method achieves low total handover latency incomparison with APCV while at the same time the numberof failed handovers increased This is due to the fact thatthe D-Scan method focused on performing scanning processin less time than obtaining the handover decision with anAP collected from APs list by comparing their RSS withthe current AP An additional weakness is that this type ofdecision making system can fall into inaccurate handoverdecisions easily Alternatively it can be concluded fromFigure 8 that the proposed AHP scheme could achieve a lownumber of failed handovers in comparison with both APCVand D-Scan methods due to accurate handover decisionsbased on an adaptable fuzzy inference system

45 Handover Failure Probability The probability of han-dover failure in unit of zero (Low) to 1 (High) for the fiveselected MNs in the experiments is considered under thissection The simulation outcome of varying number of failedhandovers using proposed AHP scheme in comparison toD-Scan and APCV methods was calculated to demonstratethe probability of failure It is under such circumstancesthat the probability of handover failure can be calculatedbased on the mean of obtained failed handovers that werepreviously recordedHandover failure probabilities have beencalculated for each of the five selectedMNs and are illustratedin Figure 9 In Figure 9 the probability of handover failureis shown in comparison form and the probability of failure


0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV

(a) Average handover delay against number ofMNs

D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27

(b) The load for 9 APs in the simulated scenario

0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP

(c) AP load (bitssec) of AP1

0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3

(d) Average MAC-layer delay

0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan

(e) Normalized packet loss ratio against number ofMNs using each AHP APCV andD-Scanmethods

0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)

Called partyCalling party

times10minus3

(f) The packet delay variation between calling andcalled party during simulation time using AHPscheme

Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]

(g) Simulation scenario

0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3

(h) The obtained Fuzzy-Quality-Cost of AP1 AP2and AP3 in simulation scenario using proposedAHP scheme

020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6

(i) The obtained Fuzzy-Quality-Cost of AP4 AP5and AP6 in simulation scenario using proposedAHP scheme

020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9

(j) Th obtained Fuzzy-Quality-Cost of AP7 AP8and AP9 in simulation scenario using proposedAHP scheme

Figure 10 Performance evaluation

achieved by the AHP scheme APCV and D-scanMN1 is 02033 and 042 respectively This implies that MN1 explainedin the proposed AHP scheme could obtain the lowest failureprobability compared with the other two methods In MN2the probability of failure was 0 016 and 028 respectivelywhich shows that the AHP scheme achieved zero handoverfailure probability with MN2 The calculated failure proba-bility for MN3 was zero in both AHP scheme and APCVwhile it was 016 in D-Scan method On the other handAPCV method with MN4 obtained 016 as the lowest failureprobability in comparisonwith proposedAHP scheme at 024and the D-Scan method at 04

To sum up the calculated failure probability in MN5 waszero by usingAHP scheme while it were 014 with APCV and06 with Scan method It can be summarized from Figure 9

that the handover failure probability using the AHP schemein MN1 MN2 and MN5 was the lowest compared with theother two methods In contrast the probability in MN3 wasthe same as AHP and APCV which was zero probabilityLast but not least in MN4 the APCV achieved less failureprobability compared with AHP and D-Scan Therefore interms of the overall probability of handover failure the AHPscheme is superior in comparison to the state of the art inreducing the probability of failure due to the aforementionedreasons

46 AverageMAC-Layer Delay Figure 10(d) shows the aver-age MAC-layer delay (measured in seconds) in comparisonformbetween the proposedAHP schemeD-Scan andAPCVduring simulation time Generally it can be observed that


the proposed AHP scheme obtained the lowest averageMAC-layer delay out of 5 simulation runs compared withAPCV and D-Scan methods The graph presented inFigure 10(d) illustrates the varying rate in MAC-layer delaywith the impact of handovers which are performed duringsimulation time Therefore by looking at the fluctuations inthe depicted graphs the behaviour of the MAC-layer delayduring the handover time is split between MN and APs inthe simulated area On the other hand a straight line graphindicates that theMAC-layer is currently not in the handoverprocess (MN is currently active with one AP)

In fact the proposed AHP scheme performed handoverdecisions accurately avoiding unnecessary handovers Aspresented in Figure 7 the total number of handovers is lessthan that of the other twomethods and theMAC-layer delayis critically decreased Hence the MAC-layer delay which issusceptible to high delay due to many handover processeshas been saved from having to do so many fluctuations as theAHP scheme reduces the number of handovers In contrastAPCV method obtained higher delay compared with AHPscheme since APCV does not consider any adaptationprocess or weight vectors in order to improve handoverdecision making in fuzzy inference systems Subsequentlythe delay increased during simulation time compared withthe AHP scheme On the other hand the MAC-layer delayincreased sharply after 50 seconds using the D-Scan methodbeing in the average higher than both APCV and AHPscheme It should be noted that the D-Scan method focuseson smartness during the scanning process in MAC-layerregardless of mobility and APrsquos aspects such as MNrsquos relateddirection with the AP and current APrsquos load

47 Impact of MNrsquos Number on Packet Loss RatioFigure 10(e) shows the packet loss ratio of AHP schemeAPCV andD-ScanThepacket loss ratio has beennormalizedbetween 0 and 1 It can be noted that the AHP schemeobtained the lowest ratio followed by APCV and D-Scanmethods As mentioned in the previous subsections theAHP scheme achieved the lowest average handover delayin addition to low handover delay associated with real-timeapplications For this reason when MN performs thehandover process with low delay the connection is less likelyto be broken during the handover procedure Therefore theprobability of incurring a high packet loss ratio is low

Furthermore the impact of MNs increasing duringsimulation time on packet loss ratio is decreased by theproposed AHP scheme since load balancing is consideredin the proposed adaptive fuzzy inference system This is notsurprising since the packet loss ratio continued to decreaseas the number of MNs increased which implies that thereare no handovers being processed by overloaded APs whichdecreases packet loss ratio From a different perspective inorder to place greater emphasis on the reasons behind theimprovement in the ratio of packet loss utilizing proposedAHP scheme Figure 10(f) presents the packet delay variationbetween calling and called party obtained by the AHPscheme

Figure 10(f) illustrates the collected results of one exam-ple of two MNs communicating with each other by running

a VoIP application type pulse-code modulation (PCM) withbit generation rate at 64 kbps Throughout this example thevariance in time delay in delivering VoIP application packetsis very similar between both the calling and the called partywhere the calling party is the MN which initiates the calland the called party is the MN which answers the call Fromthe presented graphs which were collected during a handoverprocess during simulation time it can be observed that after100 seconds the delay associated with the called party startedto increase slightly as another handover process was startedat that moment

Afterwards the delay increased when the packet ratio ofVoIP increased during the calling session After 100 secondshowever the variation in the delay time between both thecalling and the called parties was insignificant Subsequentlyafter 490 seconds of simulation time the second handover isstarted and the variance in delay experienced between callingand called parties was in the range of 1 to 2 milliseconds

48 Adaptive Fuzzy-Quality-Cost of Available APs in Simu-lation Scenario In order to present the calculated Fuzzy-Q-Cost output of input metrics (RSS119863 and 119871) that is obtainedwith each run of adaptive fuzzy inference engine one MNwas selected in the simulation scenario The calculated costsby the selectedMNof all available APs usingAHP scheme arepresented in this subsection Figure 10(g) shows the scenarioof the selected MNrsquos movement trajectory (Host1) crossing 9specific APs The APs are sorted in the movement trajectoryas sequence AP1 AP2 AP3 AP4 AP5 AP6 AP7 AP8 andAP9 Throughout these 9 APs the obtained Fuzzy-Q-Costby Host1 illustrates that an average of 5 simulation runs aresufficient to serve as a numerical example of how to calculatecost in an AHP scheme

Figure 10(h) illustrates the outputs of the proposed adap-tive fuzzy inference engine which was employed in the AHPscheme to calculate the quality cost of AP1 AP2 and AP3 Itis worth mentioning that the Fuzzy-Q-Cost of the APs wascalculated every 10 seconds during the scanning process byHost1 during its movement As can be seen from Figure 10(h)the Fuzzy-Q-Cost ranges between 0 and 1 and is presented asa function of distance At the first point of Host1 movement(started its trajectory fromAP1 as shown in Figure 10(g)) theobtained Fuzzy-Q-Cost is 1 (best quality cost) during the first100 meters Afterwards the cost began to gradually decreaseas the time distance was increasing until sharply dropping to0 beyond 800 meters

The reason is that as the time distance increased betweenHost1 and AP1 many quality aspects may have varied Forinstance when the MN is moving out of an APrsquos coveragearea the RSS will experience quality attenuation due tolarge scale fading Moreover direction can be changed tobe more likely in low directed membership Therefore theobtained cost decreased as distance increased with AP1 Onthe other hand the obtained costs from both AP2 and AP3fluctuate by the increased distance during Host1 movementIt is under such circumstances that the calculated cost byemploying proposed adaptive fuzzy inference system can beeither increased or decreased For instance as the load factors


begin to vary between AP2 andAP3with respect to two otherinput metrics (RSS and119863) different costs result for AP2 andAP3 as plotted in Figure 10(h)

Figure 10(i) illustrates the obtained Fuzzy-Q-Cost forAP4 AP5 and AP6 during Host1rsquos movement trajectory asshown in Figure 10(g) As can be seen from the figure atthe first 100 meters of Host1 movement the Fuzzy-Q-Costfor AP4 AP5 and AP6 was low cost This is not surprisingsince the allocated positions of the aforementioned APs inthe movement trajectory are at a farther distance comparedto AP1 AP2 and AP3 Therefore the cost of AP4 and AP5started to increase evenly by the time of Host1 movementtowardsAP5 crossingAP4 Subsequently the obtained cost ofAP4 sharply decreased after 930 meters of Host1 movementIn fact two main AP4 ranking factors which are RSS qualitydue to the movement out of AP4rsquos coverage area and Host1moving in different direction with AP4 at this distance aredegraded Therefore the maximum achieved Fuzzy-Q-Costfor AP4 during simulation is a cost score of 0635

On the other hand Figure 10(j) plots the obtained Fuzzy-Q-Cost of the last APs in Host1 trajectory (AP7 AP8 andAP9) in the presented scenario in Figure 10(g) Based on thepresented Fuzzy-Q-Cost graphs the cost for these three APswas zero during the first 1200 meters of Host1 movementwhich increased the AP7 cost to 0208 after 1220 metersThis sort of increase in the quality cost of AP7 fluctuatedslightly between small cost values to a maximum value of002 However as the distance increased the cost is increasedas well and at 1400 meters AP8 started to obtain smallamounts of Fuzzy-Q-Cost as well

The maximum Fuzzy-Q-Cost is obtained from AP7which was 0416 at 1910 meters of Host1 movement Beyondthis distance the calculated quality cost of AP7 started togradually decrease In this regard it can be mentioned thatbased on definedHost1rsquos movement trajectory the movementis adjusted close to the borders of AP7rsquos coverage area Forthis reason as Host1 moves farther distance away from AP7after 1910 meters the quality of two input metrics RSS andDirection is deducted In the meantime the Fuzzy-Q-Costof AP8 and AP9 increased to a maximum achieved cost forAP8 of 098 at 1930 meters and the cost value of AP9 reachedits maximum value (1) at 1980 meters

Finally the important point to note here is that through-out the obtained Fuzzy-Q-Cost as presented in the exampleof Host1 all the available APs in MNrsquos coverage area areranked between 0 and 1 and are ready to choose the bestcandidate AP to camp on Utilizing AP selection process aspresented in Section 33 the best candidate AP can be chosenaccordingly Subsequently all the associated delays in thehandover process are decreased and the QoS of the ongoingapplications are insured

5 Conclusion

This paper proposed an adaptive handover prediction (AHP)scheme that predicts the best AP candidate considering theRSS of AP candidate mobile node relative direction towardsthe access points in the vicinity and access point load Asdiscussed in detail the proposed AHP scheme relies on

an adaptive fuzzy inference system to obtain predictionsin the handover decision making process This is achievedwhereby coefficients are designed in a form and can be set upby the user Afterwards the membership functions of eachparticular input metric are calculated adaptively employingthe developed piecewise linear equations In the meantimethe weight vector for each input metric is proposed inan adjustable way to insure the accuracy of the obtainedhandover decision Subsequently the AHP scheme selectsthe final handover decision where AP obtained the highestquality cost (Fuzzy-Q-Cost) with respect to ℎ which is thethreshold value which helps to reduce unnecessary han-dovers Simulation results show that proposed AHP schemeperforms the best in contrast to the state of the art

Conflict of Interests

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

References

[1] I You Y H Han Y S Chen and H C Chao ldquoNext generationmobility managementrdquo Wireless Communications and MobileComputing vol 11 no 4 pp 443ndash445 2011

[2] R Sepulveda O Montiel-Ross J Quinones-Rivera and EE Quiroz ldquoWLAN cell handoff latency abatement using anFPGA fuzzy logic algorithm implementationrdquo Advances inFuzzy Systems vol 2012 Article ID 219602 10 pages 2012

[3] W Song ldquoResource reservation for mobile hotspots in vehic-ular environments with cellularWLAN interworkingrdquo EurasipJournal on Wireless Communications and Networking vol 2012no 1 article 18 2012

[4] A S Sadiq K Abu Bakar K Z Ghafoor J Lloret and RKhokhar ldquoAn intelligent vertical handover scheme for audioand video streaming in heterogeneous vehicular networksrdquoMobile Networks and Applications vol 18 no 6 pp 879ndash8952013

[5] A S Sadiq K Abu Bakar K Z Ghafoor and J Lloret ldquoIntelli-gent vertical handover for heterogeneous wireless networkrdquo inProceedings of theWorld Congress on Engineering and ComputerScience vol 2 pp 774ndash779 San Francisco Calif USA October2013

[6] K Nahrstedt Quality of Service in Wireless Networks over Unli-censed Spectrum Synthesis Lectures on Mobile an d PervasiveComputing 2011

[7] V Visoottiviseth and S Siwamogsatham ldquoEnd-to-end QoS-aware handover in fast handovers formobile IPv6 with DiffServusing IEEE80211eIEEE80211krdquo in Proceedings of the 10th Inter-national Conference on Advanced Communication Technology(ICACT rsquo08) pp 1548ndash1553 Mahidol University BangkokThailand February 2008

[8] L A Magagula H A Chan and O E Falowo ldquoHandoverapproaches for seamless mobility management in next gener-ation wireless networksrdquo Wireless Communications and MobileComputing vol 12 no 16 pp 1414ndash1428 2012

[9] M A Amin K Abu Bakar A H Abdullah and R H Kho-khar ldquoHandover latency measurement using variant of capwapprotocolrdquo Network Protocols and Algorithms vol 3 no 2 pp87ndash101 2011


[10] A S Sadiq K Abu Bakar and K Z Ghafoor ldquoA fuzzy logicapproach for reducing handover latency in wireless networksrdquoNetwork Protocols and Algorithms vol 2 no 4 pp 61ndash87 2011

[11] A Canovas D Bri S Sendra and J Lloret ldquoVertical WLANhandover algorithm and protocol to improve the IPTV QoSof the end userrdquo in Proceedings of the IEEE InternationalConference on Communications (ICC rsquo12) pp 1901ndash1905 June2012

[12] A S Sadiq K A Bakar K Z Ghafoor J Lloret and S MirjalilildquoA smart handover prediction system based on curve fittingmodel for Fast Mobile IPv6 in wireless networksrdquo InternationalJournal of Communication Systems vol 27 no 7 pp 969ndash9902014

[13] C Ceken S Yarkan and H Arslan ldquoInterference aware verticalhandoff decision algorithm for quality of service support inwireless heterogeneous networksrdquo Computer Networks vol 54no 5 pp 726ndash740 2010

[14] A Dutta S Das D Famolari et al ldquoSeamless proactive han-dover across heterogeneous access networksrdquoWireless PersonalCommunications vol 43 no 3 pp 837ndash855 2007

[15] L Xia L Jiang and C He ldquoA novel fuzzy logic vertical handoffalgorithm with aid of differential prediction and pre-decisionmethodrdquo in Proceedings of the IEEE International Conference onCommunications (ICC rsquo07) pp 5665ndash5670 IEEE June 2007

[16] A Majlesi and B H Khalaj ldquoAn adaptive fuzzy logic basedhandoff algorithm for hybrid networksrdquo inProceedings of the 6thInternational Conference on Signal Processing vol 2 pp 1223ndash1228 IEEE Beijing China August 2002

[17] C Xu J Teng and W Jia ldquoEnabling faster and smootherhandoffs in AP-dense 80211 wireless networksrdquo ComputerCommunications vol 33 no 15 pp 1795ndash1803 2010

[18] T-S Shih J-S Su and H-M Lee ldquoFuzzy seasonal demand andfuzzy total demand production quantities based on interval-valued fuzzy setsrdquo International Journal of Innovative Comput-ing Information and Control vol 7 no 5B pp 2637ndash2650 2011

[19] Y-F Huang Y-F Chen T-H Tan and H-L Hung ldquoAppli-cations of fuzzy logic for adaptive interference canceller inCDMAwireless communication systemsrdquo International Journalof Innovative Computing Information and Control vol 6 no 4pp 1749ndash1761 2010

[20] K Z Ghafoor K A Bakar S Salleh et al ldquoFuzzy logic-assistedgeographical routing over Vehicular AD HOC NetworksrdquoInternational Journal of Innovative Computing Information andControl vol 8 no 7 pp 5095ndash5120 2012

[21] J Holis and P Pechac ldquoElevation dependent shadowing modelfor mobile communications via high altitude platforms in built-up areasrdquo IEEE Transactions on Antennas and Propagation vol56 no 4 pp 1078ndash1084 2008


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



from visited network (network layer handover procedure)must wait until the link layer handover has performed itsown process Accordingly the initiation time is increasedafterwards the overall latency will be increased as well Thiscan be considered as a main reason of degrading the QoS inWLANs [6 7] Therefore the accurate handover predictionbased on link quality and mobility aspects is challenging toachieve a seamless mobility with low handover delay [4 8ndash12]

For this reason there is a pressing need to developan adaptive prediction technique in order to predict themost qualified AP An adaptive fuzzy logic system has beenproposed in order to address the issues of handover processeswithin wireless networks Thus the handover in link layercan be performed in predictive mode with low delay Theelaboratedmetrics including received signal strengthmobilenode relative direction towards access points in the vicinityand access point load are considered inputs of the fuzzydecision making system in order to select the best prefer-able AP around wireless local area network (WLAN) Theobtained handover decision which is based on the calculatedquality cost using fuzzy inference system is based on adaptiveinstead of fixed coefficients In other words the mean andstandard deviation of the normalized fuzzy inference systemrsquosinput metrics are applied as statistical information to adjustor adapt to the coefficients In addition this paper proposesan adjustable weight vector concept for inputmetrics in orderto cope with the continuous unpredictable variation in theirmembership degrees Furthermore handover decisions areperformed in each MN independently after RSS directiontoward APs and AP load are determined

2 Related Work

One of the essential issues in wireless communications isthe handover delay Hence many studies have tried to comeout with appropriate solution to decrease the associated timedelay during handover processes For this reason the pre-diction of next wireless network link with best QoS is highlyrequired which can be achieved by decreasing the handoverdelay issue A number of studies that have been publishedearlier proposed several handover prediction techniques forWLANs Some of these techniques are based on link qualityaspects whereas the mobility aspects are considered in theother techniques in order to obtain the handover decisionTherefore this section tries to present the related works thatwere elaborated to improve the handover prediction for APselection in WLANs

In order to achieve the desirable QoS of ongoing appli-cations a handover decision making process based on pre-dictive decision concept is proposed by [13] A fuzzy logicbased handoff decision algorithm was proposed in thisstudy for maintaining the handover decision within wirelessnetworks The decision making parameters were data rateRSS and mobile speed which have been selected as inputsfor the proposed fuzzy-based system as a way to select thebest candidate AP A handover scenario was introduced tobe performed between WiFi and global service for mobile(GSM) The output of the proposed fuzzy algorithm in this

method was setted to be a parameter called AP candidacyvalue (APCV) Afterwards APCV was defined as a realnumber in order to rank the value of the candidacy level ofthe APs in scanning range

However the proposed fuzzy logic algorithm in [13] didnot cover some other aspects that can improve the handoverdecision accuracy For instance when an AP is overloadedwith many associated MNs this can lead to a handoverfailureMoreover the authors did not consider the adaptationconcept of membership functions of the input parametersin their method which plays effective roles to maintain thehandover decision under different circumstances of wirelessnetwork quality changing In other words the adaptablemembership functions in the proposed fuzzy logic algorithmwere not considered to maintain the adaptability in fuzzyhandover system Furthermore in the proposed fuzzy logicinference system the authors did not maintain a weightvector technique giving more impact to the parameterthat has more variance behaviour compared to the otherinputs

The authors in [14] introduced dual-mode handsetsand multimode terminals that are generating demand forsolutions that enable convergence and seamless handoveracross heterogeneous access networks Besides the fuzzylogic approach has been proposed by [15] in order tohandle the handovers betweenWLANs and universal mobiletelecommunication systems (UMTS) The current RSS Pre-dicted RSS and the bandwidth have been fuzzyificated andnormalized to be used in handover decision making Thisproposed fuzzy logic system reduced the number of han-dovers between WLANs and UTMS during MN roamingYet in [15] the performance evaluation criteria such as han-dover delay AP load MN related direction towards each APand MN velocity are not addressed as key parameters in theproposed fuzzy logic system In other words by consideringsuch key parameters the probability of handover success willincrease

In contrast the authors in [2] proposed a predictivefuzzy logic controller to reduce the channel scanning pro-cess The proposed fuzzy system was designed based onMamdani-type as a way to predict the next AP from a groupof available APs obtained from scanning processes Twoinput parameters utilized by the proposed fuzzy system arethe average signal intensity (ASI) and the signal intensityvariation (SIV) The ASI input is calculated at two-secondintervals from the time beacon signal received by the MNwhich is normally broadcasted by APs with an interval of 100milliseconds

However the proposed predictive scheme basically relieson ASI metric that is normally used by MNs to estimatethe need of performing the handover with available APsMoreover the SIV is elaborated to demonstrate the behaviourof the direction of MN towards available APs Thereforeit can be observed that the obtained handover predictionusing proposed scheme by [2] totally relies on the intensityof received signal from available APsThus the probability ofexperiencing unnecessary and wrong handover prediction ishigh due to unpredictable movements and different channelpropagation aspects For instance by the time the ASI and
















AP119894




119894exceed

















No

Yes

and movement vector

Start



Yes

No

No


Yes

Yes

Yes

No

No



i = 1

i++









05

1

Weak Average Strong

0

RSS value


Adap

tive m

embe

rshi

p fu

nctio

ns o

f (120583

RSS)






120592WeakRSS =



10038161003816100381610038161003816100381610038161003816

(119860 th le RSS le 119882Max)


(1)

120592AverageRSS =

1 (119882Max le RSS le 119878th) 10038161003816100381610038161003816100381610038161003816


10038161003816100381610038161003816100381610038161003816

(119860 th le RSS le 119882Max)




(2)

120592StrongRSS



10038161003816100381610038161003816100381610038161003816



(3)




05

1




Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583D

)






cos 120579 =


radicMN21199091

+MN21199101


+ AP21199102

(4)








0

20000

40000

60000

80000

100000

120000

0 100 200 300 400 500Time (s)



0

50000

100000

150000

200000

250000

0 100 200 300 400 500Time (s)



0

100000

200000

300000

400000

500000

600000

0 100 200 300 400 500Time (s)



0

500

1000

1500

2000

2500

0 100 200 300 400 500Time (s)





120592Less119863

1 (minus1 le 119863 le MDth) 10038161003816100381610038161003816100381610038161003816


10038161003816100381610038161003816100381610038161003816


0 (119863 gt LDMax)

(5)

120592Medium119863

=

1 (LDMax le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

MDth minus 119863

MDMax minusMDth

10038161003816100381610038161003816100381610038161003816



0 (119863 gt MDMax)

or (119863 lt MDth)

(6)

120592High119863

=

0 (minus1 le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

HDth minus 119863

119863Max minusHDth

10038161003816100381610038161003816100381610038161003816


1 (MDMax le 119863 le 1)

(7)




Octets1 1 2 1 2


Stationcount

Channelutilization


05

1

Low Medium High


AP load value

Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583

load

)








120592LowLoad =



0 (119871 gt LLMax)

(8)

120592MediumLoad

=





120592High119871

=



1 (MLMax le 119871 le 40)

(10)






MN2 MN





119860 = (119860RSS 119860119863 119860119871) = (

120590RSS119894120590119894

120590119863

119894120590119894

120590119871

119894120590119894)

119894 isin RSS 119863 119871

(11)



of 1205921198941 1205921198942 120592

119894119899and 119872



119872119894=

1

119899

119899119895=1

120592119894119895 119894 isin RSS 119863 119871

120590119894=

1

119899

119899119895=1

(120592119894119895minus119872119894)

2

119894 isin RSS 119863 119871

(12)



119882 = [119882RSS119882119863119882119871] (13)



03 04 05 06 07 08 09 1

05

1Hcost VHcost

0

Mem

bers

hip

func

tion

020

LcostVLcost Mcost




119882 = (119908RSS 119908119863 119908119871)

= (

119860RSS119872RSS


119871)

(14)



119895has been


119880119895 =

120592RSS119895120592119863119895

120592119871119895

(15)


Fuzzycost = 119880119895 sdot 119882 (16)











sumAll Rules 119880119895 (17)



RSS119888lt 119879 (18)
























0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV






0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation
























AP119894




119894exceed

















No

Yes

and movement vector

Start



Yes

No

No


Yes

Yes

Yes

No

No



i = 1

i++









05

1

Weak Average Strong

0

RSS value


Adap

tive m

embe

rshi

p fu

nctio

ns o

f (120583

RSS)






120592WeakRSS =



10038161003816100381610038161003816100381610038161003816

(119860 th le RSS le 119882Max)


(1)

120592AverageRSS =

1 (119882Max le RSS le 119878th) 10038161003816100381610038161003816100381610038161003816


10038161003816100381610038161003816100381610038161003816

(119860 th le RSS le 119882Max)




(2)

120592StrongRSS



10038161003816100381610038161003816100381610038161003816



(3)




05

1




Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583D

)






cos 120579 =


radicMN21199091

+MN21199101


+ AP21199102

(4)








0

20000

40000

60000

80000

100000

120000

0 100 200 300 400 500Time (s)



0

50000

100000

150000

200000

250000

0 100 200 300 400 500Time (s)



0

100000

200000

300000

400000

500000

600000

0 100 200 300 400 500Time (s)



0

500

1000

1500

2000

2500

0 100 200 300 400 500Time (s)





120592Less119863

1 (minus1 le 119863 le MDth) 10038161003816100381610038161003816100381610038161003816


10038161003816100381610038161003816100381610038161003816


0 (119863 gt LDMax)

(5)

120592Medium119863

=

1 (LDMax le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

MDth minus 119863

MDMax minusMDth

10038161003816100381610038161003816100381610038161003816



0 (119863 gt MDMax)

or (119863 lt MDth)

(6)

120592High119863

=

0 (minus1 le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

HDth minus 119863

119863Max minusHDth

10038161003816100381610038161003816100381610038161003816


1 (MDMax le 119863 le 1)

(7)




Octets1 1 2 1 2


Stationcount

Channelutilization


05

1

Low Medium High


AP load value

Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583

load

)








120592LowLoad =



0 (119871 gt LLMax)

(8)

120592MediumLoad

=





120592High119871

=



1 (MLMax le 119871 le 40)

(10)






MN2 MN





119860 = (119860RSS 119860119863 119860119871) = (

120590RSS119894120590119894

120590119863

119894120590119894

120590119871

119894120590119894)

119894 isin RSS 119863 119871

(11)



of 1205921198941 1205921198942 120592

119894119899and 119872



119872119894=

1

119899

119899119895=1

120592119894119895 119894 isin RSS 119863 119871

120590119894=

1

119899

119899119895=1

(120592119894119895minus119872119894)

2

119894 isin RSS 119863 119871

(12)



119882 = [119882RSS119882119863119882119871] (13)



03 04 05 06 07 08 09 1

05

1Hcost VHcost

0

Mem

bers

hip

func

tion

020

LcostVLcost Mcost




119882 = (119908RSS 119908119863 119908119871)

= (

119860RSS119872RSS


119871)

(14)



119895has been


119880119895 =

120592RSS119895120592119863119895

120592119871119895

(15)


Fuzzycost = 119880119895 sdot 119882 (16)











sumAll Rules 119880119895 (17)



RSS119888lt 119879 (18)
























0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV






0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation























No

Yes

and movement vector

Start



Yes

No

No


Yes

Yes

Yes

No

No



i = 1

i++









05

1

Weak Average Strong

0

RSS value


Adap

tive m

embe

rshi

p fu

nctio

ns o

f (120583

RSS)






120592WeakRSS =



10038161003816100381610038161003816100381610038161003816

(119860 th le RSS le 119882Max)


(1)

120592AverageRSS =

1 (119882Max le RSS le 119878th) 10038161003816100381610038161003816100381610038161003816


10038161003816100381610038161003816100381610038161003816

(119860 th le RSS le 119882Max)




(2)

120592StrongRSS



10038161003816100381610038161003816100381610038161003816



(3)




05

1




Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583D

)






cos 120579 =


radicMN21199091

+MN21199101


+ AP21199102

(4)








0

20000

40000

60000

80000

100000

120000

0 100 200 300 400 500Time (s)



0

50000

100000

150000

200000

250000

0 100 200 300 400 500Time (s)



0

100000

200000

300000

400000

500000

600000

0 100 200 300 400 500Time (s)



0

500

1000

1500

2000

2500

0 100 200 300 400 500Time (s)





120592Less119863

1 (minus1 le 119863 le MDth) 10038161003816100381610038161003816100381610038161003816


10038161003816100381610038161003816100381610038161003816


0 (119863 gt LDMax)

(5)

120592Medium119863

=

1 (LDMax le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

MDth minus 119863

MDMax minusMDth

10038161003816100381610038161003816100381610038161003816



0 (119863 gt MDMax)

or (119863 lt MDth)

(6)

120592High119863

=

0 (minus1 le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

HDth minus 119863

119863Max minusHDth

10038161003816100381610038161003816100381610038161003816


1 (MDMax le 119863 le 1)

(7)




Octets1 1 2 1 2


Stationcount

Channelutilization


05

1

Low Medium High


AP load value

Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583

load

)








120592LowLoad =



0 (119871 gt LLMax)

(8)

120592MediumLoad

=





120592High119871

=



1 (MLMax le 119871 le 40)

(10)






MN2 MN





119860 = (119860RSS 119860119863 119860119871) = (

120590RSS119894120590119894

120590119863

119894120590119894

120590119871

119894120590119894)

119894 isin RSS 119863 119871

(11)



of 1205921198941 1205921198942 120592

119894119899and 119872



119872119894=

1

119899

119899119895=1

120592119894119895 119894 isin RSS 119863 119871

120590119894=

1

119899

119899119895=1

(120592119894119895minus119872119894)

2

119894 isin RSS 119863 119871

(12)



119882 = [119882RSS119882119863119882119871] (13)



03 04 05 06 07 08 09 1

05

1Hcost VHcost

0

Mem

bers

hip

func

tion

020

LcostVLcost Mcost




119882 = (119908RSS 119908119863 119908119871)

= (

119860RSS119872RSS


119871)

(14)



119895has been


119880119895 =

120592RSS119895120592119863119895

120592119871119895

(15)


Fuzzycost = 119880119895 sdot 119882 (16)











sumAll Rules 119880119895 (17)



RSS119888lt 119879 (18)
























0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV






0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










05

1

Weak Average Strong

0

RSS value


Adap

tive m

embe

rshi

p fu

nctio

ns o

f (120583

RSS)






120592WeakRSS =



10038161003816100381610038161003816100381610038161003816

(119860 th le RSS le 119882Max)


(1)

120592AverageRSS =

1 (119882Max le RSS le 119878th) 10038161003816100381610038161003816100381610038161003816


10038161003816100381610038161003816100381610038161003816

(119860 th le RSS le 119882Max)




(2)

120592StrongRSS



10038161003816100381610038161003816100381610038161003816



(3)




05

1




Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583D

)






cos 120579 =


radicMN21199091

+MN21199101


+ AP21199102

(4)








0

20000

40000

60000

80000

100000

120000

0 100 200 300 400 500Time (s)



0

50000

100000

150000

200000

250000

0 100 200 300 400 500Time (s)



0

100000

200000

300000

400000

500000

600000

0 100 200 300 400 500Time (s)



0

500

1000

1500

2000

2500

0 100 200 300 400 500Time (s)





120592Less119863

1 (minus1 le 119863 le MDth) 10038161003816100381610038161003816100381610038161003816


10038161003816100381610038161003816100381610038161003816


0 (119863 gt LDMax)

(5)

120592Medium119863

=

1 (LDMax le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

MDth minus 119863

MDMax minusMDth

10038161003816100381610038161003816100381610038161003816



0 (119863 gt MDMax)

or (119863 lt MDth)

(6)

120592High119863

=

0 (minus1 le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

HDth minus 119863

119863Max minusHDth

10038161003816100381610038161003816100381610038161003816


1 (MDMax le 119863 le 1)

(7)




Octets1 1 2 1 2


Stationcount

Channelutilization


05

1

Low Medium High


AP load value

Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583

load

)








120592LowLoad =



0 (119871 gt LLMax)

(8)

120592MediumLoad

=





120592High119871

=



1 (MLMax le 119871 le 40)

(10)






MN2 MN





119860 = (119860RSS 119860119863 119860119871) = (

120590RSS119894120590119894

120590119863

119894120590119894

120590119871

119894120590119894)

119894 isin RSS 119863 119871

(11)



of 1205921198941 1205921198942 120592

119894119899and 119872



119872119894=

1

119899

119899119895=1

120592119894119895 119894 isin RSS 119863 119871

120590119894=

1

119899

119899119895=1

(120592119894119895minus119872119894)

2

119894 isin RSS 119863 119871

(12)



119882 = [119882RSS119882119863119882119871] (13)



03 04 05 06 07 08 09 1

05

1Hcost VHcost

0

Mem

bers

hip

func

tion

020

LcostVLcost Mcost




119882 = (119908RSS 119908119863 119908119871)

= (

119860RSS119872RSS


119871)

(14)



119895has been


119880119895 =

120592RSS119895120592119863119895

120592119871119895

(15)


Fuzzycost = 119880119895 sdot 119882 (16)











sumAll Rules 119880119895 (17)



RSS119888lt 119879 (18)
























0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV






0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










05

1




Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583D

)






cos 120579 =


radicMN21199091

+MN21199101


+ AP21199102

(4)








0

20000

40000

60000

80000

100000

120000

0 100 200 300 400 500Time (s)



0

50000

100000

150000

200000

250000

0 100 200 300 400 500Time (s)



0

100000

200000

300000

400000

500000

600000

0 100 200 300 400 500Time (s)



0

500

1000

1500

2000

2500

0 100 200 300 400 500Time (s)





120592Less119863

1 (minus1 le 119863 le MDth) 10038161003816100381610038161003816100381610038161003816


10038161003816100381610038161003816100381610038161003816


0 (119863 gt LDMax)

(5)

120592Medium119863

=

1 (LDMax le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

MDth minus 119863

MDMax minusMDth

10038161003816100381610038161003816100381610038161003816



0 (119863 gt MDMax)

or (119863 lt MDth)

(6)

120592High119863

=

0 (minus1 le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

HDth minus 119863

119863Max minusHDth

10038161003816100381610038161003816100381610038161003816


1 (MDMax le 119863 le 1)

(7)




Octets1 1 2 1 2


Stationcount

Channelutilization


05

1

Low Medium High


AP load value

Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583

load

)








120592LowLoad =



0 (119871 gt LLMax)

(8)

120592MediumLoad

=





120592High119871

=



1 (MLMax le 119871 le 40)

(10)






MN2 MN





119860 = (119860RSS 119860119863 119860119871) = (

120590RSS119894120590119894

120590119863

119894120590119894

120590119871

119894120590119894)

119894 isin RSS 119863 119871

(11)



of 1205921198941 1205921198942 120592

119894119899and 119872



119872119894=

1

119899

119899119895=1

120592119894119895 119894 isin RSS 119863 119871

120590119894=

1

119899

119899119895=1

(120592119894119895minus119872119894)

2

119894 isin RSS 119863 119871

(12)



119882 = [119882RSS119882119863119882119871] (13)



03 04 05 06 07 08 09 1

05

1Hcost VHcost

0

Mem

bers

hip

func

tion

020

LcostVLcost Mcost




119882 = (119908RSS 119908119863 119908119871)

= (

119860RSS119872RSS


119871)

(14)



119895has been


119880119895 =

120592RSS119895120592119863119895

120592119871119895

(15)


Fuzzycost = 119880119895 sdot 119882 (16)











sumAll Rules 119880119895 (17)



RSS119888lt 119879 (18)
























0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV






0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










0

20000

40000

60000

80000

100000

120000

0 100 200 300 400 500Time (s)



0

50000

100000

150000

200000

250000

0 100 200 300 400 500Time (s)



0

100000

200000

300000

400000

500000

600000

0 100 200 300 400 500Time (s)



0

500

1000

1500

2000

2500

0 100 200 300 400 500Time (s)





120592Less119863

1 (minus1 le 119863 le MDth) 10038161003816100381610038161003816100381610038161003816


10038161003816100381610038161003816100381610038161003816


0 (119863 gt LDMax)

(5)

120592Medium119863

=

1 (LDMax le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

MDth minus 119863

MDMax minusMDth

10038161003816100381610038161003816100381610038161003816



0 (119863 gt MDMax)

or (119863 lt MDth)

(6)

120592High119863

=

0 (minus1 le 119863 le HDth) 10038161003816100381610038161003816100381610038161003816

HDth minus 119863

119863Max minusHDth

10038161003816100381610038161003816100381610038161003816


1 (MDMax le 119863 le 1)

(7)




Octets1 1 2 1 2


Stationcount

Channelutilization


05

1

Low Medium High


AP load value

Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583

load

)








120592LowLoad =



0 (119871 gt LLMax)

(8)

120592MediumLoad

=





120592High119871

=



1 (MLMax le 119871 le 40)

(10)






MN2 MN





119860 = (119860RSS 119860119863 119860119871) = (

120590RSS119894120590119894

120590119863

119894120590119894

120590119871

119894120590119894)

119894 isin RSS 119863 119871

(11)



of 1205921198941 1205921198942 120592

119894119899and 119872



119872119894=

1

119899

119899119895=1

120592119894119895 119894 isin RSS 119863 119871

120590119894=

1

119899

119899119895=1

(120592119894119895minus119872119894)

2

119894 isin RSS 119863 119871

(12)



119882 = [119882RSS119882119863119882119871] (13)



03 04 05 06 07 08 09 1

05

1Hcost VHcost

0

Mem

bers

hip

func

tion

020

LcostVLcost Mcost




119882 = (119908RSS 119908119863 119908119871)

= (

119860RSS119872RSS


119871)

(14)



119895has been


119880119895 =

120592RSS119895120592119863119895

120592119871119895

(15)


Fuzzycost = 119880119895 sdot 119882 (16)











sumAll Rules 119880119895 (17)



RSS119888lt 119879 (18)
























0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV






0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











Octets1 1 2 1 2


Stationcount

Channelutilization


05

1

Low Medium High


AP load value

Adap

tive m

embe

rshi

pfu

nctio

ns o

f (120583

load

)








120592LowLoad =



0 (119871 gt LLMax)

(8)

120592MediumLoad

=





120592High119871

=



1 (MLMax le 119871 le 40)

(10)






MN2 MN





119860 = (119860RSS 119860119863 119860119871) = (

120590RSS119894120590119894

120590119863

119894120590119894

120590119871

119894120590119894)

119894 isin RSS 119863 119871

(11)



of 1205921198941 1205921198942 120592

119894119899and 119872



119872119894=

1

119899

119899119895=1

120592119894119895 119894 isin RSS 119863 119871

120590119894=

1

119899

119899119895=1

(120592119894119895minus119872119894)

2

119894 isin RSS 119863 119871

(12)



119882 = [119882RSS119882119863119882119871] (13)



03 04 05 06 07 08 09 1

05

1Hcost VHcost

0

Mem

bers

hip

func

tion

020

LcostVLcost Mcost




119882 = (119908RSS 119908119863 119908119871)

= (

119860RSS119872RSS


119871)

(14)



119895has been


119880119895 =

120592RSS119895120592119863119895

120592119871119895

(15)


Fuzzycost = 119880119895 sdot 119882 (16)











sumAll Rules 119880119895 (17)



RSS119888lt 119879 (18)
























0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV






0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












MN2 MN





119860 = (119860RSS 119860119863 119860119871) = (

120590RSS119894120590119894

120590119863

119894120590119894

120590119871

119894120590119894)

119894 isin RSS 119863 119871

(11)



of 1205921198941 1205921198942 120592

119894119899and 119872



119872119894=

1

119899

119899119895=1

120592119894119895 119894 isin RSS 119863 119871

120590119894=

1

119899

119899119895=1

(120592119894119895minus119872119894)

2

119894 isin RSS 119863 119871

(12)



119882 = [119882RSS119882119863119882119871] (13)



03 04 05 06 07 08 09 1

05

1Hcost VHcost

0

Mem

bers

hip

func

tion

020

LcostVLcost Mcost




119882 = (119908RSS 119908119863 119908119871)

= (

119860RSS119872RSS


119871)

(14)



119895has been


119880119895 =

120592RSS119895120592119863119895

120592119871119895

(15)


Fuzzycost = 119880119895 sdot 119882 (16)











sumAll Rules 119880119895 (17)



RSS119888lt 119879 (18)
























0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV






0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation

















sumAll Rules 119880119895 (17)



RSS119888lt 119879 (18)
























0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV






0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation















0123456789

10

MN1 MN2 MN3 MN4 MN5

Num

ber o

f tot

al h

ando

vers

AHPD-Scan

APCV






0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










0

05

1

15

2

25

3

35

MN1 MN2 MN3 MN4 MN5

Num

ber o

f fai

led

hand

over

s

AHPAPCV

D-Scan





02

0 0

024

0

033

016

0

016

014

042

028

016

04

06

0

01

02

03

04

05

06

07

MN1 MN2 MN3 MN4 MN5

Han

dove

r fai

lure

pro

babi

lity

Mobile node

AHPAPCV

D-Scan






0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










0 5 10 15 20 25 30 35 40 45 500

05

1

15

2

25

3

Number of MN

Aver

age h

ando

ver d

elay

(s)

AHPD-Scan

APCV


D-ScanAPCV

AHP

0102030405060708090

100

1 2 3 4 5 6 7 8 9

AP

load

()

Access point

36

24

42

100

82

62 73

1

22

48

82

22

65 74

53

100

33

5

11 21

20

32

7

21

9

19 27


0 100 200 300 400 500 600 7000

500

1000

1500

2000

2500

3000

Time (s)

AP

load

(bits

s)

APCVD-Scan

AHP


0 100 200 300 400 500 600 7000

05

1

15

2

25

3

35

4

Time (s)

Aver

age M

AC-la

yer d

elay

(s)

AHPAPCV

D-Scan

times10minus3


0 10 20 30 40 50 600

010203040506070809

1

Number of MN

Nor

mal

ized

pac

ket l

oss

AHPAPCV

D-Scan


0 100 200 300 400 500 600 7000

05

1

15

2

25

Time (s)

Pack

et d

elay

var

iatio

n of

VoI

P (s

)


times10minus3


Figure 10 Continued


mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










mIPv6Network

Host43Host5

Host8

Host28

Host3

Host45

Host39

Host33

Host23

Host20

Host15

Host31Host12

Host25Host13 Host30

Host48Host49

Host50Host21Host16

Host27Host32

Host37

Host47Host4Host40

Host22

Host11

Host29

Host26

Host14

Host42Host41

Host9 Host6

Host24

Host2 Host45

Host17Host44

Host35

Host34

Host38

Host19

Host7

Host18Host1

Host10

Host36

AP6

AP9

AP8

AP7

AP4

AP5

AP2

AP3

AP1 Configurator

Channel control

Default getway

Hub

R 5

R 4

R 3

R 1

R 2

CN [total cn]


0

200

400

600

800

1000

1200

1400

1600

1800

20000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP1AP2

AP3


020

040

060

080

010

0012

0014

0016

0018

0020

00

0010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP4AP5

AP6


020

040

060

080

010

0012

0014

0016

0018

0020

000

010203040506070809

1

Distance (m)

AH

P Fu

zzy-

Qua

lity-

Cos

t

AP7AP8

AP9
























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation


























5 Conclusion





References


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation















5 Conclusion





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











RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Date post:	05-Jul-2020
Category:	Documents
Upload:	others
View:	2 times
Download:	0 times

Research Article An Adaptive Handover Prediction Scheme...

Documents